{"site": "浅译快报", "updated": "2026-09-29T11:49:08+00:00", "items": [{"id": "2026-09-21-comfyui-v0370-v0371-qwen-image21-moge3-hy-image35", "url": "https://www.transwonder.top/insights/#2026-09-21-comfyui-v0370-v0371-qwen-image21-moge3-hy-image35", "title": "ComfyUI v0.37.0/v0.37.1：Qwen-Image 2.1 原生三模板 + MoGe 3 几何估计 + HY Image 3.5 Preview（5 参考图·2K 输出）", "date": "2026-09-21", "category": "产品", "source": "ComfyUI 官方", "src": "https://github.com/Comfy-Org/ComfyUI/releases/tag/v0.37.0", "sources": null, "summary": "ComfyUI v0.37.0/v0.37.1 原生集成 Qwen-Image 2.1 三种生成模式节点，并接入 MoGe 3 单目几何估计与 HY Image 3.5 Preview 五参考图 2K 输出。", "reason": null, "tags": ["ComfyUI","v0.37","Qwen-Image","MoGe3","HY-Image","背景移除","几何估计","工作流"], "topics": ["阿里巴巴","ComfyUI","图像生成"], "firsthand": true, "editor_pick": false},{"id": "2026-09-20-alibaba-qwen-image-21-open-source", "url": "https://www.transwonder.top/insights/#2026-09-20-alibaba-qwen-image-21-open-source", "title": "阿里开源 Qwen-Image-2.1：7B 统一图像生成+编辑模型，原生 RGBA 透明通道，支持 10 图参考，首日同步 ComfyUI/Diffusers/vLLM", "date": "2026-09-20", "category": "模型", "source": "Qwen 官方博客 + GitHub + TechNode + Hugging Face（多源）", "src": "https://github.com/QwenLM/Qwen-Image-2.1", "sources": null, "summary": "阿里开源 Qwen-Image-2.1，7B 参数统一图像生成与编辑模型，首次原生支持 RGBA 透明通道，最多 10 张参考图输入，发布首日可通过 ComfyUI、Diffusers、vLLM 调用。", "reason": null, "tags": ["Qwen","阿里巴巴","图像生成","图像编辑","开源","RGBA透明","参考图像","ComfyUI","vLLM"], "topics": ["阿里巴巴","ComfyUI","图像生成"], "firsthand": true, "editor_pick": false},{"id": "2026-09-17-pika-creative-studio-model-router", "url": "https://www.transwonder.top/insights/#2026-09-17-pika-creative-studio-model-router", "title": "Pika 重构为全流程创意工作室：7 大 AI 模型自动路由，内置镜头清单生成、重打光、换装工具套件", "date": "2026-09-17", "category": "产品", "source": "Semafor（独家）/ AlphaSignal", "src": "https://www.semafor.com/article/09/17/2026/video-ai-company-pika-offers-tool-to-speed-up-ai-video-creation", "sources": null, "summary": "Pika 将产品从单一视频生成工具重构为多模型创意工作室，集成 7 个 AI 模型并新增逐镜头清单生成、重打光与换装工具套件。", "reason": null, "tags": ["Pika","视频生成","模型路由","分镜","重打光"], "topics": ["视频生成"], "firsthand": false, "editor_pick": false},{"id": "2026-09-16-midjourney-alpha-v8-2-korean-mobile-editor", "url": "https://www.transwonder.top/insights/#2026-09-16-midjourney-alpha-v8-2-korean-mobile-editor", "title": "Midjourney Alpha 9/16：韩文支持 + V8.2 编辑器优化 + 移动/平板布局", "date": "2026-09-16", "category": "产品", "source": "Midjourney 官方更新日志（官方单源）", "src": "https://updates.midjourney.com/alpha-changelog-9-16-26/", "sources": null, "summary": "Midjourney Alpha 9/16 更新：新增韩文语言界面支持", "reason": null, "tags": ["Midjourney","Alpha更新","V8.2","韩文支持"], "topics": ["Midjourney"], "firsthand": true, "editor_pick": false},{"id": "2026-09-16-figma-weave-node-shaders-beta-community-publishing", "url": "https://www.transwonder.top/insights/#2026-09-16-figma-weave-node-shaders-beta-community-publishing", "title": "Figma 九月新功能：Weave 节点（Design→AI 工作流）+ Shaders 开放测试版 + 社区工具发布", "date": "2026-09-16", "category": "产品", "source": "Figma 官方博客（release-notes）", "src": "https://www.figma.com/blog/how-we-built-generative-plugins-and-shaders/", "sources": null, "summary": "Figma 推出 Weave 节点支持将 Design 画框拖入 AI 工作流、Shaders 开放测试版通过提示词生成动效着色器，并上线 Weave 社区工具发布功能。", "reason": null, "tags": ["Figma","Weave","Shaders","AI视觉特效","工作流"], "topics": ["Figma","智能体与工作流"], "firsthand": true, "editor_pick": false},{"id": "2026-09-16-arxiv-design-space-visual-interfaces-genai", "url": "https://www.transwonder.top/insights/#2026-09-16-arxiv-design-space-visual-interfaces-genai", "title": "arXiv 2609.18065：为生成式图像模型构建视觉界面设计空间框架", "date": "2026-09-16", "category": "论文", "source": "arXiv.org", "src": "https://arxiv.org/abs/2609.18065", "sources": null, "summary": "arXiv 论文 2609.18065 提出系统性框架梳理生成式图像模型视觉界面的提示输入、控制粒度、迭代模式与输出展示四大核心维度。", "reason": null, "tags": ["arXiv","界面设计","设计空间","HCI","生成式图像模型"], "topics": ["研究论文"], "firsthand": true, "editor_pick": false},{"id": "2026-09-16-anthropic-claude-unified-chat-cowork", "url": "https://www.transwonder.top/insights/#2026-09-16-anthropic-claude-unified-chat-cowork", "title": "Anthropic Claude 统一界面：Chat + Cowork 合并单窗口，Claude Design 保持品牌一致性", "date": "2026-09-16", "category": "产品", "source": "TechCrunch + Anthropic 官方", "src": "https://techcrunch.com/2026/09/16/anthropic-merges-claude-chat-and-cowork-in-one-interface/", "sources": null, "summary": "Anthropic 将 Claude.ai 聊天与 Cowork 工作空间合并至单一窗口，并推出 Claude Design 支持跨项目品牌一致性管理。", "reason": null, "tags": ["Anthropic","Claude","Cowork","Claude Design","统一界面"], "topics": ["Anthropic"], "firsthand": true, "editor_pick": false},{"id": "2026-09-15-vidu-s2-realtime-avatar-streaming-edit", "url": "https://www.transwonder.top/insights/#2026-09-15-vidu-s2-realtime-avatar-streaming-edit", "title": "Vidu S2：实时 720p 数字人生成 + 流式视频风格编辑，支持 VR 空间", "date": "2026-09-15", "category": "模型", "source": "arXiv（arXiv:2609.11638）+ 生数科技 / PR Newswire", "src": "https://arxiv.org/abs/2609.11638", "sources": null, "summary": "生数科技发布 Vidu S2，包含实时 720p 数字人生成与无中断流式视频风格编辑两个子模型，在线演示同步开放。", "reason": null, "tags": ["Vidu S2","实时视频生成","数字人","流式编辑","换装换背景"], "topics": ["视频生成"], "firsthand": true, "editor_pick": false},{"id": "2026-09-15-elevenlabs-mcp-multimodal-creative-pipeline", "url": "https://www.transwonder.top/insights/#2026-09-15-elevenlabs-mcp-multimodal-creative-pipeline", "title": "ElevenLabs MCP 多模态扩展：语音+音乐+图像+视频 50+ 模型单次 OAuth 接入", "date": "2026-09-15", "category": "产品", "source": "ElevenLabs 官方博客", "src": "https://www.testingcatalog.com/elevenlabs-mcp-can-now-generate-voice-music-images-video/", "sources": null, "summary": "ElevenLabs 将 MCP 连接器从纯语音扩展至多模态全链路，单次 OAuth 安装接入 50 余个语音、音乐、图像与视频生成模型。", "reason": null, "tags": ["ElevenLabs","MCP","多模态","创意管线","音乐生成"], "topics": ["音频与音乐","智能体与工作流"], "firsthand": true, "editor_pick": false},{"id": "2026-09-15-coreldraw-text-to-vector-ai", "url": "https://www.transwonder.top/insights/#2026-09-15-coreldraw-text-to-vector-ai", "title": "CorelDRAW Graphics Suite 九月更新：Text-to-Vector AI 文字直生可编辑矢量图形 + PowerTRACE 提速", "date": "2026-09-15", "category": "产品", "source": "GlobeNewswire + PR Newswire", "src": "https://www.globenewswire.com/news-release/2026/09/15/3362194/0/en/ai-vector-graphics-tools-power-creative-workflows-in-coreldraw-graphics-suite-2026.html", "sources": null, "summary": "CorelDRAW Graphics Suite September 2026 Update 新增 Text-to-Vector AI，支持从文字描述直接生成可商用、可编辑矢量图形路径，并将 PowerTRACE 速度提升 4 倍。", "reason": null, "tags": ["CorelDRAW","Text-to-Vector","矢量生成","平面设计"], "topics": ["设计工具"], "firsthand": true, "editor_pick": false},{"id": "2026-09-15-comfyui-v0360-fasth3-bfl-flux-video-edit", "url": "https://www.transwonder.top/insights/#2026-09-15-comfyui-v0360-fasth3-bfl-flux-video-edit", "title": "ComfyUI v0.36.0：FastVideo FastH3 8步视频生成 + BFL FLUX 视频编辑节点 + YuE2 音乐 + Tripo P2 3D", "date": "2026-09-15", "category": "产品", "source": "GitHub (comfyanonymous/ComfyUI releases) + comfyui-wiki.com", "src": "https://github.com/comfyanonymous/ComfyUI/releases/tag/v0.36.0", "sources": null, "summary": "ComfyUI v0.36.0 新增 FastH3 8步蒸馏视频、BFL FLUX 视频编辑、YuE2 音乐生成与 Tripo P2 3D 等多项节点，单一工作流覆盖文字到视频配乐完整链路。", "reason": null, "tags": ["ComfyUI","FastH3","BFL FLUX Video Edit","YuE2","视频生成"], "topics": ["Black Forest Labs","ComfyUI","视频生成"], "firsthand": true, "editor_pick": false},{"id": "2026-09-15-bfl-flux-video-edit-api", "url": "https://www.transwonder.top/insights/#2026-09-15-bfl-flux-video-edit-api", "title": "BFL FLUX Video Edit API：$0.03/秒保留原片运动与音频的精准视频编辑端点", "date": "2026-09-15", "category": "模型", "source": "Black Forest Labs 官方文档 + ComfyUI v0.36.0 发布说明", "src": "https://bfl.ai/video-edit", "sources": null, "summary": "Black Forest Labs 推出 FLUX Video Edit API，$0.03/秒输出，可按文字描述精准修改视频中指定元素，同时完整保留原片运动轨迹与音频。", "reason": null, "tags": ["BFL","FLUX Video Edit","视频编辑","运动保留","ComfyUI"], "topics": ["Black Forest Labs","ComfyUI","音频与音乐"], "firsthand": true, "editor_pick": false},{"id": "2026-09-12-multimatte-feyn-text-guided-cutout", "url": "https://www.transwonder.top/insights/#2026-09-12-multimatte-feyn-text-guided-cutout", "title": "MultiMatte by Feyn：文字引导多层蒙版，一键输出精细分层", "date": "2026-09-12", "category": "产品", "source": "usefeyn.com", "src": "https://usefeyn.com/blog/multimatte/", "sources": null, "summary": "Feyn 发布 MultiMatte，通过文字描述同时生成多个精准蒙版层，输出与 Photoshop / After Effects 兼容的 alpha 通道分层文件，支持复杂场景下对多对象分层抠图。", "reason": null, "tags": ["Feyn","MultiMatte","蒙版","图像编辑"], "topics": ["图像生成"], "firsthand": true, "editor_pick": false},{"id": "2026-09-12-higgsfield-motion-designer-ae-premiere", "url": "https://www.transwonder.top/insights/#2026-09-12-higgsfield-motion-designer-ae-premiere", "title": "Higgsfield AI Motion Designer：原生内嵌 After Effects 与 Premiere Pro 的 AI 视频生成插件", "date": "2026-09-12", "category": "产品", "source": "X (@higgsfield) / AI media", "src": "https://x.com/higgsfield/status/2098409362041753708", "sources": null, "summary": "Higgsfield AI 推出 Motion Designer 插件，在 After Effects / Premiere Pro 时间线上选取片段、输入文字提示即可原地生成或替换运动画面，已开放候补名单申请。", "reason": null, "tags": ["Higgsfield","After Effects","Premiere Pro","AI视频生成"], "topics": ["Adobe","视频生成"], "firsthand": false, "editor_pick": false},{"id": "2026-09-10-relightformer-siggraph-asia-2026-multiview-relighting", "url": "https://www.transwonder.top/insights/#2026-09-10-relightformer-siggraph-asia-2026-multiview-relighting", "title": "RelightFormer：SIGGRAPH Asia 2026 多视角物体重打光生成式 Transformer，无需逆渲染", "date": "2026-09-10", "category": "论文", "source": "arXiv", "src": "https://arxiv.org/abs/2609.07414", "sources": null, "summary": "香港科技大学 vLAR 团队提出 RelightFormer，前馈生成式 Transformer 无需逆渲染，输入目标光照条件和多视角图像即可输出光照一致的多视角重打光结果，已被 SIGGRAPH Asia 2026 接收，权重开源。", "reason": null, "tags": ["RelightFormer","重打光","生成式AI","SIGGRAPH Asia 2026"], "topics": ["研究论文"], "firsthand": true, "editor_pick": false},{"id": "2026-09-10-prix-ars-electronica-2026-deer-hunter-golden-nica", "url": "https://www.transwonder.top/insights/#2026-09-10-prix-ars-electronica-2026-deer-hunter-golden-nica", "title": "Prix Ars Electronica 2026：AI 与档案影像融合作品「Deer Hunter」获新动画艺术金熊奖", "date": "2026-09-10", "category": "赛展", "source": "Ars Electronica 官方", "src": "https://ars.electronica.art/prix/en/winners/", "sources": null, "summary": "第 38 届 Prix Ars Electronica 2026 颁奖，Andrew Herzog 的「Deer Hunter」以档案素材与生成式 AI 融合叙事摘得新动画艺术金熊奖（Golden Nica）。", "reason": null, "tags": ["Ars Electronica","Prix","AI动画","媒体艺术"], "topics": ["视频生成","艺术与展览"], "firsthand": true, "editor_pick": false},{"id": "2026-09-sora-api-shutdown-disney-deal-collapse", "url": "https://www.transwonder.top/insights/#2026-09-sora-api-shutdown-disney-deal-collapse", "title": "OpenAI Sora API 9月24日关停：$10亿迪士尼合作流产，AI视频格局重塑", "date": "2026-09-09", "category": "行业", "source": "OpenAI官方/The Verge/TechCrunch", "src": "https://variety.com/2026/digital/news/openai-shutting-down-sora-video-disney-1236698277/", "sources": null, "summary": "OpenAI Sora API将于2026年9月24日终止服务，独立app已于4月下线，与迪士尼约$10亿规模的AI视频合作随之流产", "reason": null, "tags": ["OpenAI","Sora","API关停","迪士尼","AI视频","格局重塑"], "topics": ["OpenAI","视频生成","行业与融资"], "firsthand": true, "editor_pick": false},{"id": "2026-09-09-metadisruption-ai-art-design-festival", "url": "https://www.transwonder.top/insights/#2026-09-09-metadisruption-ai-art-design-festival", "title": "Metadisruption AI Art & Design Festival 2026：9月23–27日，英国威尔士斯旺西", "date": "2026-09-09", "category": "赛展", "source": "FilmFreeway/UWTSD（威尔士三一圣大卫大学）", "src": "https://filmfreeway.com/Metadisruption", "sources": null, "summary": "Metadisruption AI Art & Design Festival 2026定于9月23–27日在英国威尔士斯旺西（UWTSD）举办，聚焦AI艺术、算法设计与创意科技的跨学科展览与研讨", "reason": null, "tags": ["AI艺术","设计节","英国","UWTSD","威尔士","算法设计","国际展览"], "topics": ["艺术与展览"], "firsthand": true, "editor_pick": false},{"id": "2026-09-09-eccv2026-ai4va-workshop-visual-arts", "url": "https://www.transwonder.top/insights/#2026-09-09-eccv2026-ai4va-workshop-visual-arts", "title": "ECCV 2026第四届AI for Visual Arts工作坊今日举办：世界模型×视觉艺术、长视频生成前沿", "date": "2026-09-09", "category": "论文", "source": "ECCV 2026官方/AI4VA Workshop官方", "src": "https://ai4va-eccv.github.io/", "sources": null, "summary": "ECCV 2026第四届AI for Visual Arts专题工作坊于9月9日在瑞典马尔默举办，聚焦世界模型、长视频生成、神经三维等视觉艺术AI前沿方向", "reason": null, "tags": ["ECCV 2026","AI4VA","视觉艺术","世界模型","学术","长视频生成"], "topics": ["视频生成","3D 与世界模型","研究论文"], "firsthand": true, "editor_pick": false},{"id": "2026-09-09-design-ai-summit-creativepro", "url": "https://www.transwonder.top/insights/#2026-09-09-design-ai-summit-creativepro", "title": "Design + AI Summit 2026（CreativePro）：设计师AI工作流实战峰会，9月17日线上举办", "date": "2026-09-09", "category": "赛展", "source": "CreativePro", "src": "https://creativepro.com/event/design-ai-summit-2026/", "sources": null, "summary": "CreativePro主办的Design + AI Summit 2026定于2026年9月17日线上举办，聚焦平面、UI和出版设计师的AI工具实战工作流", "reason": null, "tags": ["Design+AI","Summit","CreativePro","设计师AI工作流","线上峰会","Firefly"], "topics": ["Adobe","智能体与工作流","艺术与展览"], "firsthand": true, "editor_pick": false},{"id": "2026-09-09-comfyui-v0350-comfy-compiler-seven-partner-nodes", "url": "https://www.transwonder.top/insights/#2026-09-09-comfyui-v0350-comfy-compiler-seven-partner-nodes", "title": "ComfyUI v0.35.0：Comfy Compiler 加速与七个前沿模型官方合作节点一键接入", "date": "2026-09-09", "category": "产品", "source": "ComfyUI 官方 / ai-tldr.dev", "src": "https://docs.comfy.org/changelog", "sources": null, "summary": "ComfyUI v0.35.0 引入 Comfy Compiler 内存加速，并新增 GPT-6、GPT Image 2.5、Claude Fable 5.1、Gemini Omni 1.1 等 7 个官方合作模型节点，同时扩展 3D 管线。", "reason": null, "tags": ["ComfyUI","工作流","多模型","3D生成"], "topics": ["ComfyUI","3D 与世界模型","智能体与工作流"], "firsthand": true, "editor_pick": false},{"id": "2026-09-08-runway-adobe-premiere-after-effects-plugin", "url": "https://www.transwonder.top/insights/#2026-09-08-runway-adobe-premiere-after-effects-plugin", "title": "Runway 发布 Adobe Premiere Pro 与 After Effects 原生插件", "date": "2026-09-08", "category": "产品", "source": "Runway 官方 / Adobe 官方", "src": "https://runway.com/news/company-news/runway-for-adobe", "sources": null, "summary": "Runway 推出 Adobe Premiere Pro 和 After Effects 官方原生插件，专业剪辑者可在时间轴内直接调用 Runway Aleph 2 完成 AI 视频生成、风格化与超分辨率，无需离开剪辑软件。", "reason": null, "tags": ["Runway","Adobe","Premiere Pro","AI视频生成"], "topics": ["Adobe","Runway","视频生成"], "firsthand": true, "editor_pick": false},{"id": "2026-09-08-runway-200m-arr-kinetix", "url": "https://www.transwonder.top/insights/#2026-09-08-runway-200m-arr-kinetix", "title": "Runway AI 年收入达 2 亿美元并购巴黎实验室 Kinetix", "date": "2026-09-08", "category": "行业", "source": "Bloomberg / TechTimes", "src": "https://www.bloomberg.com/news/articles/2026-09-08/ai-startup-runway-hits-200-million-in-annual-recurring-revenue", "sources": null, "summary": "Runway AI 年收入从 4 月 1 亿美元跃升至 9 月 2 亿美元，同时宣布收购专研三维人体动作的巴黎实验室 Kinetix。", "reason": null, "tags": ["Runway","AI视频","商业","收购"], "topics": ["Runway","视频生成","行业与融资"], "firsthand": false, "editor_pick": false},{"id": "2026-09-08-chatgpt-images-2-5-sketch-api", "url": "https://www.transwonder.top/insights/#2026-09-08-chatgpt-images-2-5-sketch-api", "title": "OpenAI 发布 ChatGPT Images 2.5：新增手绘 Sketch 参考与双轨 API 模型", "date": "2026-09-08", "category": "模型", "source": "OpenAI 官方博客", "src": "https://openai.com/index/introducing-chatgpt-images-2-5/", "sources": null, "summary": "ChatGPT Images 2.5 新增手绘草图 Sketch 参考功能，生成速度提升 50%，并发布 Flare（速度优先）和 Sunburst（精度优先）两个 API 模型，全面取代 GPT-Image-2。", "reason": null, "tags": ["OpenAI","图像生成","Sketch","API"], "topics": ["OpenAI","图像生成"], "firsthand": true, "editor_pick": false},{"id": "2026-09-08-adobe-premiere-generative-media-tool-ibc2026", "url": "https://www.transwonder.top/insights/#2026-09-08-adobe-premiere-generative-media-tool-ibc2026", "title": "Adobe Premiere Pro 内嵌五大 AI 视频模型与 AI 音效生成——IBC 2026 Generative Media Tool", "date": "2026-09-08", "category": "产品", "source": "Adobe 官方博客 / No Film School / TechTimes", "src": "https://blog.adobe.com/en/publish/2026/09/08/generate-create-directly-in-your-timeline-with-new-ai-powered-innovations-in-premiere-after-effects", "sources": null, "summary": "Adobe 于 IBC 2026 推出 Premiere Pro Generative Media Tool，时间轴内框选即可调用 Firefly、Runway、Kling、Veo、Luma 五大 AI 视频模型，并同步新增 AI 音效与声景生成。", "reason": null, "tags": ["Adobe","Premiere Pro","AI视频生成","IBC 2026"], "topics": ["Adobe","视频生成"], "firsthand": true, "editor_pick": false},{"id": "2026-09-05-comfyui-v034-meta-muse-comfy-cloud", "url": "https://www.transwonder.top/insights/#2026-09-05-comfyui-v034-meta-muse-comfy-cloud", "title": "ComfyUI v0.34.4+v0.34.5：Meta Muse首发、子工作流嵌套、Comfy Cloud视频直出", "date": "2026-09-05", "category": "产品", "source": "ComfyUI官方GitHub/ComfyUI Wiki", "src": "https://github.com/Comfy-Org/ComfyUI/releases", "sources": null, "summary": "ComfyUI v0.34.4首发Meta Muse视频生成节点；v0.34.5新增子工作流嵌套调用及Comfy Cloud原生视频流式预览", "reason": null, "tags": ["ComfyUI","v0.34.4","v0.34.5","Meta Muse","Comfy Cloud","子工作流","视频生成"], "topics": ["Meta","ComfyUI","视频生成"], "firsthand": true, "editor_pick": false},{"id": "2026-09-04-llada-image", "url": "https://www.transwonder.top/insights/#2026-09-04-llada-image", "title": "LLaDA-Image：开源6B统一文生图+参考编辑+中英文排版，Base+Turbo双档", "date": "2026-09-04", "category": "模型", "source": "HuggingFace", "src": "https://huggingface.co/inclusionAI/LLaDA-Image", "sources": null, "summary": "inclusionAI 发布 LLaDA-Image：6B 参数开源统一图像生成与编辑模型，分 Base（50步高质量）和 Turbo（4步快速，Twin-DMD 蒸馏）双档，支持文生图、参考图风格迁移、精准编辑以及中英文文字排版渲染。", "reason": null, "tags": ["图像生成","开源","中文","文字渲染","inclusionAI"], "topics": ["图像生成","设计工具"], "firsthand": true, "editor_pick": false},{"id": "2026-09-03-suno-download-limits-effective-ai-music-policy", "url": "https://www.transwonder.top/insights/#2026-09-03-suno-download-limits-effective-ai-music-policy", "title": "Suno正式实施下载限制：免费账户终身仅可下载7次，引发AI音乐版权争议", "date": "2026-09-03", "category": "行业", "source": "Suno官方/多家音乐媒体", "src": "https://suno.com/blog/suno-updates-tos", "sources": null, "summary": "Suno正式对免费账户实施终身7次下载上限，被普遍解读为回应版权诉讼压力的商业政策调整", "reason": null, "tags": ["Suno","AI音乐","下载限制","版权政策","商业模式"], "topics": ["音频与音乐","版权与政策"], "firsthand": true, "editor_pick": false},{"id": "2026-09-03-runway-gwm-worlds-2", "url": "https://www.transwonder.top/insights/#2026-09-03-runway-gwm-worlds-2", "title": "Runway GWM Worlds 2：全球首个实时可操控AI世界模型，720p/24fps持续生成并响应文字指令", "date": "2026-09-03", "category": "模型", "source": "Runway官方研究博客", "src": "https://runway.com/research/introducing-gwm-worlds-2", "sources": null, "summary": "Runway发布通用世界模型第二代GWM Worlds 2，持续生成720p/24fps视频并响应用户实时文字操控指令，世界从当前状态延续而非重置", "reason": null, "tags": ["Runway","GWM Worlds 2","世界模型","实时交互","AI视频","互动叙事"], "topics": ["Runway","视频生成","3D 与世界模型"], "firsthand": true, "editor_pick": false},{"id": "2026-09-03-krea-agents-mcp-creative-memory", "url": "https://www.transwonder.top/insights/#2026-09-03-krea-agents-mcp-creative-memory", "title": "Krea Agents：单一创意提示替代复杂节点图，支持 MCP 与长期记忆", "date": "2026-09-03", "category": "产品", "source": "alphasignal.ai", "src": "https://alphasignal.ai/news/krea-agents-replaces-complex-node-graphs-with-a-single-creative-prompt", "sources": null, "summary": "Krea 发布 Agents，单条自然语言指令驱动整套图像 / 视频生成工作流，内置 MCP 工具调用与跨会话长期记忆，面向 Krea Pro 订阅用户开放。", "reason": null, "tags": ["Krea","Agents","MCP","工作流自动化"], "topics": ["智能体与工作流"], "firsthand": true, "editor_pick": false},{"id": "2026-09-03-gpt6-astra-computer-use", "url": "https://www.transwonder.top/insights/#2026-09-03-gpt6-astra-computer-use", "title": "OpenAI发布GPT-6（Project Astra）：Computer Use能力正式上线，可操控3D/CAD软件", "date": "2026-09-03", "category": "模型", "source": "OpenAI官方/Fortune/CNBC", "src": "https://openai.com/index/gpt-6-astra/", "sources": null, "summary": "OpenAI正式发布GPT-6（Project Astra），新增Computer Use能力，官方演示包括操控3D建模软件和CAD程序执行多步骤设计任务", "reason": null, "tags": ["OpenAI","GPT-6","Project Astra","Computer Use","3D建模","CAD","多模态"], "topics": ["OpenAI","3D 与世界模型"], "firsthand": true, "editor_pick": false},{"id": "2026-09-03-adobe-for-slack", "url": "https://www.transwonder.top/insights/#2026-09-03-adobe-for-slack", "title": "Adobe for Slack：Firefly生图能力嵌入Slack工作流，商业授权直接可用", "date": "2026-09-03", "category": "产品", "source": "Adobe官方", "src": "https://9to5mac.com/2026/09/02/adobe-brings-photoshop-firefly-and-70-creative-tools-to-slack-with-mcp-app/", "sources": null, "summary": "Adobe推出Firefly for Slack集成，用户在Slack对话中可直接调用文生图、图生图和重新着色功能生成商业可用视觉资产，无需切换应用", "reason": null, "tags": ["Adobe","Firefly","Slack","工作流集成","文生图","商业授权","协作设计"], "topics": ["Adobe","图像生成","智能体与工作流"], "firsthand": true, "editor_pick": false},{"id": "2026-09-02-serpentine-art-tech-fellowship", "url": "https://www.transwonder.top/insights/#2026-09-02-serpentine-art-tech-fellowship", "title": "Serpentine首届Art+Technology Fellowship开放申请：Refik Anadol与曹斐担任评委", "date": "2026-09-02", "category": "赛展", "source": "The Art Newspaper/Serpentine官方", "src": "https://www.serpentinegalleries.org/whats-on/future-art-ecosystems-rd-fellowship-art-x-convergence/", "sources": null, "summary": "伦敦蛇形画廊首届Art+Technology Fellowship正式开放申请，Refik Anadol和曹斐担任评审，面向AI艺术与科技交叉领域创作者", "reason": null, "tags": ["Serpentine","Art+Technology Fellowship","Refik Anadol","曹斐","AI艺术","驻留","申请"], "topics": ["艺术与展览"], "firsthand": true, "editor_pick": false},{"id": "2026-09-02-arxiv-ai-support-novice-visual-artists-ideation-pedagogy", "url": "https://www.transwonder.top/insights/#2026-09-02-arxiv-ai-support-novice-visual-artists-ideation-pedagogy", "title": "arXiv研究：AI如何支持视觉艺术初学者的创意构思与教学介入", "date": "2026-09-02", "category": "论文", "source": "arXiv（论文编号 2509.24167）", "src": "https://arxiv.org/abs/2509.24167", "sources": null, "summary": "arXiv论文2509.24167考察AI工具如何支持视觉艺术初学者的创意构思过程，并探讨教学介入的适当边界", "reason": null, "tags": ["arXiv","AI艺术教育","视觉艺术","初学者","创意构思","教学法"], "topics": ["研究论文","设计教育"], "firsthand": true, "editor_pick": false},{"id": "2026-09-01-google-pics-workspace-ai-image-editor", "url": "https://www.transwonder.top/insights/#2026-09-01-google-pics-workspace-ai-image-editor", "title": "Google Pics 登陆 Workspace：Gemini 驱动的专业级 AI 图像创建与编辑", "date": "2026-09-01", "category": "产品", "source": "Google Workspace 官方博客", "src": "https://workspaceupdates.googleblog.com/2026/09/google-pics-brings-pro-level-ai-image-creation-and-editing-to-Google-Workspace.html", "sources": null, "summary": "Google 将 Pics 正式集成进 Workspace，以 Gemini 为底层提供文生图、背景替换、对象移除等专业级图像编辑，可在 Google Docs / Slides / Drive 内直接操作。", "reason": null, "tags": ["Google","Workspace","Gemini","图像编辑"], "topics": ["Google","图像生成"], "firsthand": true, "editor_pick": false},{"id": "2026-09-01-claude-fable-51-reasoning-cache-cut", "url": "https://www.transwonder.top/insights/#2026-09-01-claude-fable-51-reasoning-cache-cut", "title": "Anthropic发布Fable 5.1+Mythos 5.1：缓存读取成本降75%，维持输出价格", "date": "2026-09-01", "category": "模型", "source": "Anthropic官方", "src": "https://www.anthropic.com/claude-fable-and-mythos-5-1", "sources": null, "summary": "Anthropic更新Fable 5.1和Mythos 5.1，缓存读取（cache read）成本降低75%，输出token价格维持不变", "reason": null, "tags": ["Anthropic","Fable 5.1","Mythos 5.1","API定价","缓存成本","Claude"], "topics": ["Anthropic"], "firsthand": true, "editor_pick": false},{"id": "2026-08-31-runway-solaris-interface-world-model", "url": "https://www.transwonder.top/insights/#2026-08-31-runway-solaris-interface-world-model", "title": "Runway Solaris：首个界面世界模型，将实时网站与应用UI转化为视频体验", "date": "2026-08-31", "category": "模型", "source": "Runway官方研究博客", "src": "https://runwayml.com/research/solaris", "sources": null, "summary": "Runway发布研究级模型Solaris，定义为界面世界模型（Interface World Model），可将网页和应用UI实时渲染为连续视频体验，目前为研究Demo阶段", "reason": null, "tags": ["Runway","Solaris","界面世界模型","UI视频化","交互视频","研究发布"], "topics": ["Runway","3D 与世界模型"], "firsthand": true, "editor_pick": false},{"id": "2026-08-31-runway-ruby-video-color-grading", "url": "https://www.transwonder.top/insights/#2026-08-31-runway-ruby-video-color-grading", "title": "Runway Ruby：自然语言描述驱动的专业级视频色彩分级模型", "date": "2026-08-31", "category": "产品", "source": "Runway官方changelog", "src": "https://help.runwayml.com/hc/en-us/articles/54749891346579-Creating-HDR-videos-with-Runway-Ruby", "sources": null, "summary": "Runway发布Ruby专用视频调色模型，通过自然语言描述（如「70s胶片感」「清晨蓝调」）执行专业级色彩分级，理解光线情绪与时代氛围", "reason": null, "tags": ["Runway","Ruby","视频调色","色彩分级","视频后期","AI美学"], "topics": ["Runway"], "firsthand": true, "editor_pick": false},{"id": "2026-08-31-midjourney-v82-edit-mode-four-refs-inpaint-outpaint", "url": "https://www.transwonder.top/insights/#2026-08-31-midjourney-v82-edit-mode-four-refs-inpaint-outpaint", "title": "Midjourney V8.2 Edit Mode全量开放：4图参考+Inpaint+Outpaint统一界面", "date": "2026-08-31", "category": "产品", "source": "Midjourney官方", "src": "https://updates.midjourney.com/edit-model-for-v8/", "sources": null, "summary": "Midjourney V8.2 Edit Mode正式全量开放，在单一界面整合4张参考图输入、画布内绘（inpaint）与外延（outpaint）功能", "reason": null, "tags": ["Midjourney","V8.2","Edit Mode","inpaint","outpaint","多图参考"], "topics": ["Midjourney"], "firsthand": true, "editor_pick": false},{"id": "2026-08-pika-video-to-sound-scene-aware", "url": "https://www.transwonder.top/insights/#2026-08-pika-video-to-sound-scene-aware", "title": "Pika推出场景感知视频配音模型：画面驱动全场景音效与音乐同步生成", "date": "2026-08-27", "category": "模型", "source": "Pika官方", "src": "https://experiment.pika.art/blog/pika-audio-models", "sources": null, "summary": "Pika推出专属视频配音模型，分析视频画面语义和节奏后同步生成音效、环境声和音乐底层，无需手动对口型或检索素材库", "reason": null, "tags": ["Pika","视频配音","AI音效","音视频同步","视频后期","场景感知"], "topics": ["音频与音乐"], "firsthand": true, "editor_pick": false},{"id": "2026-08-27-wan30-prime-fast-text-video-image-video-audio", "url": "https://www.transwonder.top/insights/#2026-08-27-wan30-prime-fast-text-video-image-video-audio", "title": "阿里Wan3.0 Prime发布：$0.06–$0.30/秒的快速视频变体，支持文转视频、图转视频、音频输入", "date": "2026-08-27", "category": "模型", "source": "阿里巴巴官方", "src": "https://www.alibabacloud.com/help/en/model-studio/wan3-0-video-prime", "sources": null, "summary": "阿里Wan3.0 Prime快速变体上线，定价$0.06–$0.30/秒，支持文字转视频、图像转视频及音频驱动视频生成三类任务", "reason": null, "tags": ["Wan3.0","Prime","阿里巴巴","视频生成","快速模式","定价"], "topics": ["阿里巴巴","视频生成","音频与音乐"], "firsthand": true, "editor_pick": false},{"id": "2026-08-27-minimax-h3-max", "url": "https://www.transwonder.top/insights/#2026-08-27-minimax-h3-max", "title": "MiniMax H3 Max：fal.ai联合发布，35倍加速，Design Arena全球第一", "date": "2026-08-27", "category": "模型", "source": "MiniMax官方/fal.ai", "src": "https://blog.fal.ai/introducing-h3-max-by-fal/", "sources": null, "summary": "MiniMax与fal.ai联合发布H3 Max，相比H3标准版实现35倍推理加速，在Design Arena视频质量评测中位列全球第一", "reason": null, "tags": ["MiniMax","H3 Max","fal.ai","视频生成","加速","Design Arena"], "topics": ["视频生成","大模型"], "firsthand": true, "editor_pick": false},{"id": "2026-08-27-gemini-omni-11-flash", "url": "https://www.transwonder.top/insights/#2026-08-27-gemini-omni-11-flash", "title": "Gemini Omni 1.1 Flash 发布：40秒视频、首尾帧双锁定、4K分辨率、Draft Mode $0.03/秒", "date": "2026-08-27", "category": "模型", "source": "Google DeepMind官方", "src": "https://blog.google/innovation-and-ai/technology/developers-tools/build-with-gemini-omni-1-1-flash/", "sources": null, "summary": "Google发布Gemini Omni 1.1 Flash，支持40秒视频生成、首尾帧双重锁定、4K输出及$0.03/秒的Draft快速模式", "reason": null, "tags": ["Gemini","视频生成","4K","首尾帧锁定","Draft Mode"], "topics": ["Google","视频生成"], "firsthand": true, "editor_pick": false},{"id": "2026-08-26-recraft-v4-styles-reference-consistent-generation", "url": "https://www.transwonder.top/insights/#2026-08-26-recraft-v4-styles-reference-consistent-generation", "title": "Recraft V4 Styles：1–10张参考图驱动风格一致性图像生成", "date": "2026-08-26", "category": "模型", "source": "Recraft官方", "src": "https://www.recraft.ai/blog/meet-recraft-v4-styles", "sources": null, "summary": "Recraft发布V4 Styles功能，支持输入1至10张参考图提取视觉风格，生成与参考图高度一致的新图像", "reason": null, "tags": ["Recraft","V4","风格一致性","参考图","图像生成","品牌设计"], "topics": ["图像生成"], "firsthand": true, "editor_pick": false},{"id": "2026-08-26-figma-detachable-agent-chat-vector-eraser-recolor", "url": "https://www.transwonder.top/insights/#2026-08-26-figma-detachable-agent-chat-vector-eraser-recolor", "title": "Figma新增可拆卸AI对话面板、矢量路径橡皮擦、区域填色三大功能", "date": "2026-08-26", "category": "产品", "source": "Figma官方", "src": "https://www.figma.com/release-notes/", "sources": null, "summary": "Figma推出可拆卸AI对话面板、矢量路径橡皮擦及区域智能填色三项新功能，进一步整合AI辅助能力至核心设计工作流", "reason": null, "tags": ["Figma","AI对话","矢量编辑","区域填色","设计工具"], "topics": ["Figma","设计工具"], "firsthand": true, "editor_pick": false},{"id": "2026-08-26-comfyui-v034-trellis2-wan3prime-recraft-v4", "url": "https://www.transwonder.top/insights/#2026-08-26-comfyui-v034-trellis2-wan3prime-recraft-v4", "title": "ComfyUI v0.34系列：Pixal3D+TRELLIS2原生3D、Recraft V4、Gemini Omni 1.1 Flash、Wan3.0 Prime", "date": "2026-08-26", "category": "产品", "source": "ComfyUI官方GitHub/ComfyUI Wiki", "src": "https://github.com/Comfy-Org/ComfyUI/releases", "sources": null, "summary": "ComfyUI v0.34系列版本集成Pixal3D+TRELLIS2原生3D管线、Recraft V4风格生成、Gemini Omni 1.1 Flash视频及Wan3.0 Prime快速变体", "reason": null, "tags": ["ComfyUI","v0.34","TRELLIS2","Recraft V4","Gemini Omni","Wan3.0 Prime","3D生成"], "topics": ["Google","阿里巴巴","ComfyUI"], "firsthand": true, "editor_pick": false},{"id": "2026-08-25-stability-ai-76m-series-b-music-labels", "url": "https://www.transwonder.top/insights/#2026-08-25-stability-ai-76m-series-b-music-labels", "title": "Stability AI完成$76M B轮：环球音乐、索尼、华纳、EA联合参投", "date": "2026-08-25", "category": "行业", "source": "TechCrunch/多家财经媒体", "src": "https://stability.ai/news-updates/stability-ai-latest-funding-backed-by-entertainment-industry-biggest-names", "sources": null, "summary": "Stability AI完成$76M B轮融资，环球音乐集团、索尼音乐、华纳音乐及游戏巨头EA联合参投，此轮融资被解读为内容产业对AI公司授权路径的战略背书", "reason": null, "tags": ["Stability AI","融资","B轮","音乐公司","EA","版权合规"], "topics": ["Stability AI","音频与音乐","版权与政策"], "firsthand": true, "editor_pick": false},{"id": "2026-08-25-chinese-ink-ai-belgrade-exhibition", "url": "https://www.transwonder.top/insights/#2026-08-25-chinese-ink-ai-belgrade-exhibition", "title": "水墨×AI国际展贝尔格莱德站开幕：中国当代水墨与AI生成的跨文化对话", "date": "2026-08-25", "category": "赛展", "source": "展览主办方官方", "src": "https://english.news.cn/20260825/edaf524bbc5f4ebfb944c04acf1fa51b/c.html", "sources": null, "summary": "水墨×AI国际巡展贝尔格莱德站开幕，聚焦中国当代水墨艺术与AI生成技术的融合创作实践", "reason": null, "tags": ["水墨","AI艺术","国际展览","贝尔格莱德","中国当代艺术","跨文化"], "topics": ["艺术与展览"], "firsthand": true, "editor_pick": false},{"id": "2026-08-24-comfyui-v0334-wan3-meshy7-vcube", "url": "https://www.transwonder.top/insights/#2026-08-24-comfyui-v0334-wan3-meshy7-vcube", "title": "ComfyUI v0.33.4：Wan 3.0 30秒原生节点+Meshy-7 Ultra 3D生成+ByteDance vCube 8K视频增强", "date": "2026-08-24", "category": "产品", "source": "ComfyUI官方GitHub", "src": "https://github.com/comfy-org/ComfyUI/releases/tag/v0.33.4", "sources": null, "summary": "ComfyUI v0.33.4 发布，新增阿里 Wan 3.0 原生工作流节点（30秒/1080p视频+多模态输入）、Meshy-7 Ultra 3D生成节点、字节跳动 vCube 视频增强节点（至8K超分+帧率增强），同时修复若干UI稳定性问题。", "reason": null, "tags": ["ComfyUI","v0.33.4","Wan 3.0","Meshy-7","vCube"], "topics": ["阿里巴巴","ComfyUI","3D 与世界模型"], "firsthand": true, "editor_pick": false},{"id": "2026-08-24-alibaba-wan3-30s-multimodal-video", "url": "https://www.transwonder.top/insights/#2026-08-24-alibaba-wan3-30s-multimodal-video", "title": "阿里Wan3.0全球正式发布：30秒视频、文档转视频、$0.05–$0.20/秒", "date": "2026-08-24", "category": "模型", "source": "阿里巴巴官方/TechCrunch", "src": "https://www.alibabacloud.com/blog/wan3-0-30-second-ai-video-generation-from-any-input_603452", "sources": null, "summary": "阿里巴巴Wan3.0全球发布，支持最长30秒视频生成、PDF/PPT/Word文档直接转视频，定价$0.05–$0.20/秒", "reason": null, "tags": ["Wan3.0","阿里巴巴","视频生成","文档转视频","多模态"], "topics": ["阿里巴巴","视频生成"], "firsthand": true, "editor_pick": false},{"id": "2026-09-09-ars-electronica-2026-future-begins-negotiating-humanity", "url": "https://www.transwonder.top/insights/#2026-09-09-ars-electronica-2026-future-begins-negotiating-humanity", "title": "Ars Electronica 2026「Future Begins」9月9–13日林茨：AI不再是未来假设，「协商人类性」成47届核心议题", "date": "2026-08-23", "category": "赛展", "source": "Ars Electronica官方", "src": "https://ars.electronica.art/futurebegins/en/", "sources": null, "summary": "Ars Electronica 2026（第47届）将于9月9–13日在奥地利林茨举行，主题「Future Begins」聚焦「协商人类性」（Negotiating Humanity）：AI已从「奇异未来场景」渗透至社会各层面，媒体艺术的任务是持续协商技术与人性的边界。", "reason": null, "tags": ["Ars Electronica",2026,"Future Begins","媒体艺术"], "topics": ["艺术与展览"], "firsthand": true, "editor_pick": false},{"id": "2026-08-20-adobe-firefly-audio-ga", "url": "https://www.transwonder.top/insights/#2026-08-20-adobe-firefly-audio-ga", "title": "Adobe Firefly 音频全面开放：商业授权音乐/配音/音效三合一，创意全链路扩展至有声", "date": "2026-08-20", "category": "产品", "source": "Adobe官方博客 / 9to5Mac", "src": "https://blog.adobe.com/en/publish/2026/08/20/adobe-firefly-expands-its-creative-ai-studio-generate-music-speech-and-sound-effects-in-one-place", "sources": null, "summary": "Adobe Firefly 音频工具（Generate Music / Generate Speech / Generate Sound Effects）2026年8月20日正式对所有用户全量开放；三工具均为商业免许可证授权，Generate Music由Firefly Music Model驱动可生成与视频时长/情绪匹配的原创音乐，Generate Speech支持ElevenLabs可选模型+情绪/语速调节，Generate Sound Effects根据视频节奏自动同步音效；同日新增Gemini Omni Flash为合作伙伴模型。", "reason": null, "tags": ["Adobe Firefly","音频生成","音乐生成","配音"], "topics": ["Adobe","音频与音乐"], "firsthand": true, "editor_pick": false},{"id": "2026-08-19-walker-art-center-jeyifous-ai-misread", "url": "https://www.transwonder.top/insights/#2026-08-19-walker-art-center-jeyifous-ai-misread", "title": "Walker Art Center澄清：Jeyifous展览作品并非AI生成，认知鸿沟引发行业讨论", "date": "2026-08-19", "category": "赛展", "source": "Walker Art Center官方/多家艺术媒体", "src": "https://www.mprnews.org/story/2026/08/19/walker-art-center-clarifies-ai-use-in-olalekan-jeyifous-exhibition", "sources": null, "summary": "Walker Art Center正式澄清艺术家Olalekan Jeyifous展览作品并非AI生成，此事件暴露了观众对AI艺术与数字艺术的认知边界模糊", "reason": null, "tags": ["Walker Art Center","Jeyifous","AI误读","艺术认知","数字艺术"], "topics": ["艺术与展览"], "firsthand": true, "editor_pick": false},{"id": "2026-08-19-flux-video-upscale-4k-super-resolution-bfl", "url": "https://www.transwonder.top/insights/#2026-08-19-flux-video-upscale-4k-super-resolution-bfl", "title": "FLUX Video Upscale工具发布：视频超分1.5x–3x，原生输出至4K，双模式选择", "date": "2026-08-19", "category": "产品", "source": "Black Forest Labs官方", "src": "https://bfl.ai/blog/flux-video-upscale", "sources": null, "summary": "Black Forest Labs发布FLUX Video Upscale，支持对已生成视频进行1.5x至3x超分辨率处理，最高输出4K，提供质量优先和速度优先双模式", "reason": null, "tags": ["FLUX","视频超分","4K","Black Forest Labs","视频后期"], "topics": ["Black Forest Labs"], "firsthand": true, "editor_pick": false},{"id": "2026-08-18-mit-csail-attribution-decay-ai-art", "url": "https://www.transwonder.top/insights/#2026-08-18-mit-csail-attribution-decay-ai-art", "title": "MIT CSAIL《Nature Communications》：AI生成图像的归因衰减使版权追责在技术层面趋于不可能", "date": "2026-08-18", "category": "论文", "source": "MIT CSAIL/Nature Communications", "src": "https://news.mit.edu/2026/when-ai-art-has-no-author-generated-images-often-cant-be-traced-to-training-data-0818", "sources": null, "summary": "MIT CSAIL发表于《Nature Communications》的研究显示，AI图像生成的多步骤变换机制导致训练数据归因信号快速衰减，使版权追责在技术层面趋于不可能", "reason": null, "tags": ["MIT CSAIL","版权","归因衰减","AI生成图像","学术研究"], "topics": ["版权与政策","研究论文"], "firsthand": true, "editor_pick": false},{"id": "2026-08-18-coinve-edit-tencent-22b-video-editing", "url": "https://www.transwonder.top/insights/#2026-08-18-coinve-edit-tencent-22b-video-editing", "title": "腾讯CoinVE-Edit：22B参数区域感知视频编辑，单次Pass完成多处局部修改", "date": "2026-08-18", "category": "论文", "source": "腾讯研究院", "src": "https://arxiv.org/abs/2608.17566", "sources": null, "summary": "腾讯发布CoinVE-Edit，22B参数视频编辑模型，区域感知架构支持单次推理同时完成2-5处独立局部修改", "reason": null, "tags": ["腾讯","视频编辑","CoinVE-Edit","22B","区域感知","局部编辑"], "topics": ["腾讯","研究论文"], "firsthand": true, "editor_pick": false},{"id": "2026-08-17-google-imagen-retired-nano-banana-migration", "url": "https://www.transwonder.top/insights/#2026-08-17-google-imagen-retired-nano-banana-migration", "title": "Google于8月17日关闭全部Imagen 4 API，Figma插件/Make/Zapier工作流静默失效，强制迁移至Nano Banana", "date": "2026-08-17", "category": "产品", "source": "Google Firebase官方文档 + ud.hk + clauding.de", "src": "https://firebase.google.com/docs/ai-logic/imagen-models-migration", "sources": null, "summary": "Google于2026-08-17关闭全部Imagen 4系列API（imagen-4.0-generate-001 / ultra / fast），全线切换至Gemini 3.x图像族（Nano Banana）；API调用方式从专用接口改为通用Gemini content-generation接口（modality=image）；调用Imagen端点的Figma插件/Zapier/Make/n8n及大量小SaaS工具无声停摆；受影响用户须主动迁移端点+重写SDK调用逻辑。", "reason": null, "tags": ["Google","Imagen","Nano Banana","API退役"], "topics": ["Google","Figma","图像生成"], "firsthand": true, "editor_pick": false},{"id": "2026-08-17-anthropic-claude-code-design-command", "url": "https://www.transwonder.top/insights/#2026-08-17-anthropic-claude-code-design-command", "title": "Anthropic Claude Code新增/design命令研究预览版：编程会话中直接生成UI画板", "date": "2026-08-17", "category": "产品", "source": "Anthropic官方", "src": "https://the-decoder.com/claude-code-gets-a-design-command-that-lets-developers-create-ui-mockups-right-in-the-terminal/", "sources": null, "summary": "Anthropic在Claude Code中推出/design命令研究预览版，开发者可在编码会话内通过自然语言描述生成UI画板并直接导入代码流", "reason": null, "tags": ["Anthropic","Claude Code","UI设计","AI原生设计","研究预览","Figma"], "topics": ["Anthropic","Figma","设计工具"], "firsthand": true, "editor_pick": false},{"id": "2026-08-13-minimax-music3-open-weight-5min-song", "url": "https://www.transwonder.top/insights/#2026-08-13-minimax-music3-open-weight-5min-song", "title": "MiniMax Music 3开源：首个产品级AI音乐模型，凭歌词+描述独立生成5分钟完整歌曲", "date": "2026-08-13", "category": "模型", "source": "MiniMax官方Blog / MarkTechPost", "src": "https://www.minimax.io/blog/minimax-music-3-0-next-generation-open-weights-production-ready-versatile-music-model", "sources": null, "summary": "MiniMax于8月13-14日发布MiniMax Music 3开源权重——首个「产品级」AI音乐生成模型，给定歌词+风格描述即可输出最长5分钟结构完整的歌曲（人声+乐器+编曲）；8B Global LLM（基于Qwen3-8B）把控全曲语义进程，0.6B Local LLM精修声学细节；32kHz 16bit立体声输出，最低8GB显存可本地运行，权重已上HuggingFace与GitHub；ComfyUI v0.33.1（08-13）同步集成。", "reason": null, "tags": ["MiniMax","Music3","AI音乐","开源"], "topics": ["大模型","音频与音乐"], "firsthand": true, "editor_pick": false},{"id": "2026-08-13-figma-agent-skills-50plus-community", "url": "https://www.transwonder.top/insights/#2026-08-13-figma-agent-skills-50plus-community", "title": "Figma开放Agent Skills创作权限：设计师可编写自定义工作流技能，社区已发布50+技能", "date": "2026-08-13", "category": "产品", "source": "Figma官方博客", "src": "https://www.figma.com/blog/got-skills-make-the-figma-agent-a-better-collaborator/", "sources": null, "summary": "Figma向所有用户开放Agent Skills创作权限，设计师可编写自定义工作流技能并发布至社区，截至9月社区已积累50+可用技能", "reason": null, "tags": ["Figma","Agent Skills","工作流自动化","设计工具","设计平台"], "topics": ["Figma","设计工具","智能体与工作流"], "firsthand": true, "editor_pick": false},{"id": "2026-08-13-comfyui-v0-33-1-minimax-music3-cuda-graphs", "url": "https://www.transwonder.top/insights/#2026-08-13-comfyui-v0-33-1-minimax-music3-cuda-graphs", "title": "ComfyUI v0.31→v0.33.1：原生MiniMax Music 3文生音乐+CUDA图加速+Bria生图编辑节点", "date": "2026-08-13", "category": "产品", "source": "ComfyUI官方GitHub", "src": "https://github.com/Comfy-Org/ComfyUI/releases/tag/v0.33.1", "sources": null, "summary": "ComfyUI v0.33.1（8月13日）引入MiniMax Music 3原生支持（文生音乐+歌词驱动，最长5分钟），Anima Tunes视频音效增强，CUDA图原生加速（与动态VRAM协同），以及Bria GenFill/Eraser/Expand/Upscale四套合作伙伴节点；MiniMax H3 Context IR和Regenerate节点同步更新。", "reason": null, "tags": ["ComfyUI","v0.33.1","MiniMax Music3","CUDA"], "topics": ["ComfyUI","图像生成","大模型"], "firsthand": true, "editor_pick": false},{"id": "2026-08-07-comfyui-v031-v032-v033-update-wave", "url": "https://www.transwonder.top/insights/#2026-08-07-comfyui-v031-v032-v033-update-wave", "title": "ComfyUI 8月三连更（v0.31–v0.33.1）：Wan-Animate2 / LTX-2.5 / MiniMax Music 3原生支持，PSD分层合成上线", "date": "2026-08-13", "category": "产品", "source": "ComfyUI官方GitHub / ComfyUI Wiki", "src": "https://comfyui-wiki.com/en/news/2026-08-13-comfyui-v0-33-1", "sources": null, "summary": "ComfyUI在8月7至13日连续发布三个版本：v0.31.0（8/7）原生支持Wan-Animate2（姿态+参考控制）、SeeDance 2.5/Seedream 5.0 Layer Separation伙伴节点、PSD风格分层合成（混合模式+图层状态）；v0.32.0（8/11）原生LTX 2.5 Partner Nodes + PyTorch 2.7最低要求；v0.33.1（8/13）原生支持MiniMax Music 3开源权重（歌词文本编码+音频生成）、Bria伙伴节点、修复MiniMax H3/LTX/KSamplerAdvanced已知问题。", "reason": null, "tags": ["ComfyUI","v0.31","v0.32","v0.33.1"], "topics": ["阿里巴巴","ComfyUI","大模型"], "firsthand": true, "editor_pick": false},{"id": "2026-08-12-runway-gemini-omni-flash-api-router", "url": "https://www.transwonder.top/insights/#2026-08-12-runway-gemini-omni-flash-api-router", "title": "Runway接入Gemini Omni Flash并推出API Model Router，实现跨模型自动路由", "date": "2026-08-12", "category": "产品", "source": "Runway官方", "src": "https://runway.com/product/models/gemini-omni", "sources": null, "summary": "Runway新增Gemini Omni Flash接入并推出API Model Router，根据任务类型自动选择最优底层视频模型", "reason": null, "tags": ["Runway","Gemini","API Router","模型路由","视频生成"], "topics": ["Google","Runway","视频生成"], "firsthand": true, "editor_pick": false},{"id": "2026-08-11-ltx25-open-weights-68sec-4k-video", "url": "https://www.transwonder.top/insights/#2026-08-11-ltx25-open-weights-68sec-4k-video", "title": "LTX-2.5 开源：10 秒 720p 仅需 6.8 秒，支持 4K 与同步音频，ComfyUI 当日 Day-0 支持", "date": "2026-08-11", "category": "模型", "source": "Lightricks 官方 / VentureBeat", "src": "https://venturebeat.com/technology/ltx-2-5-open-weights-video-model-from-lightricks-68-seconds-720p/", "sources": null, "summary": "Lightricks 于 2026-08-11 开源 LTX-2.5（HuggingFace `Lightricks/LTX-Video-2.5`）：10 秒 720p 视频生成仅需 6.8 秒（单 RTX 4090），支持 4K 输出、同步音频、原生多镜头（multi-shot）叙事；ComfyUI 发布当日 day-0 官方支持（含节点与工作流模板）；商业营收 $10M ARR 以下完全免费。", "reason": null, "tags": ["LTX-2.5","Lightricks","开源视频","6.8秒"], "topics": ["ComfyUI","音频与音乐"], "firsthand": true, "editor_pick": false},{"id": "2026-08-11-comfyui-v0320-v0331-ltx25-music3", "url": "https://www.transwonder.top/insights/#2026-08-11-comfyui-v0320-v0331-ltx25-music3", "title": "ComfyUI v0.32.0 + v0.33.1 双版本更新：LTX-2.5 原生支持 + MiniMax Music 3 T2M 最长 5 分钟", "date": "2026-08-11", "category": "产品", "source": "ComfyUI 官方 GitHub / ComfyUI Wiki", "src": "https://github.com/Comfy-Org/ComfyUI/releases/tag/v0.32.0", "sources": null, "summary": "ComfyUI v0.32.0（2026-08-11）：SageAttention 级别显存优化（速度提升约 20-30%）+ LTX 2.5 原生支持（day-0）+ Qwen Image 3.0 + Grok Imagine Image 2.0 节点；v0.33.1（2026-08-13）：MiniMax Music 3 T2M 节点（文本→音乐，最长 5 分钟）+ CUDA Graph 计算图缓存进一步提速；两个版本合计大幅提升高端视频+音频生成的本地工作流能力。", "reason": null, "tags": ["ComfyUI","v0.32.0","v0.33.1","LTX-2.5"], "topics": ["ComfyUI","大模型","音频与音乐"], "firsthand": true, "editor_pick": false},{"id": "2026-08-11-anthropic-claude-text-watermark-eu", "url": "https://www.transwonder.top/insights/#2026-08-11-anthropic-claude-text-watermark-eu", "title": "Anthropic为Claude全平台输出嵌入隐形水印：EU AI Act合规，水印可跨编辑持续存在", "date": "2026-08-11", "category": "行业", "source": "The Decoder / TechTimes", "src": "https://the-decoder.com/anthropic-watermarks-all-claude-outputs-globally-with-marks-that-may-persist/", "sources": null, "summary": "Anthropic宣布为所有Claude输出嵌入不可见文本水印，水印在部分编辑后仍可持续存在，用于EU AI Act Article 50合规；标记证明「AI处理过」而非「AI完全创作」，全球用户即时生效。", "reason": null, "tags": ["Anthropic","Claude","水印","EU AI Act"], "topics": ["Anthropic","版权与政策"], "firsthand": false, "editor_pick": false},{"id": "2026-08-10-alibaba-wan30-public-launch", "url": "https://www.transwonder.top/insights/#2026-08-10-alibaba-wan30-public-launch", "title": "阿里万象Wan 3.0公测正式开放：单镜30秒+文档转视频+全模态输入，API ¥0.30–1.20/秒", "date": "2026-08-10", "category": "模型", "source": "TechNode Global / Alizila", "src": "https://technode.global/2026/08/10/chinas-alibaba-releases-wan3-0-ai-video-model-in-public-beta-with-30s-clips-multimodal-inputs/", "sources": null, "summary": "阿里通义实验室Wan 3.0于8月6日开公测、8月10日杭州发布活动正式上线：单镜30秒（Wan 2.7为15秒上限）；全模态输入——PDF、PPT、Excel、网页均可作为输入源，生成文档转视频；API计价¥0.30–1.20/秒（480P–1080P）。", "reason": null, "tags": ["Wan 3.0","阿里","通义","视频生成"], "topics": ["阿里巴巴","视频生成"], "firsthand": false, "editor_pick": false},{"id": "2026-08-08-adobe-firefly-interfaces-enterprise-batch-workflow", "url": "https://www.transwonder.top/insights/#2026-08-08-adobe-firefly-interfaces-enterprise-batch-workflow", "title": "Adobe Firefly「Interfaces」企业级批量工作流Beta：500资产一次性生成", "date": "2026-08-08", "category": "产品", "source": "Adobe官方", "src": "https://helpx.adobe.com/firefly/web/work-with-enterprise-features/creative-production/creative-production-overview.html", "sources": null, "summary": "Adobe推出Firefly Interfaces beta，提供企业级UI支持单次批量生成最多500个视觉资产，面向品牌营销和内容团队", "reason": null, "tags": ["Adobe","Firefly","企业工作流","批量生成","Interfaces"], "topics": ["Adobe","智能体与工作流"], "firsthand": true, "editor_pick": false},{"id": "2026-08-07-wan-animate-2-opensource", "url": "https://www.transwonder.top/insights/#2026-08-07-wan-animate-2-opensource", "title": "阿里通义 Wan-Animate-2 开源：Apache 2.0 角色动画框架，ComfyUI 原生支持，本地跑赢可灵 MotionControl", "date": "2026-08-07", "category": "模型", "source": "Open Source For You / ComfyUI Wiki", "src": "https://comfyui-wiki.com/en/news/2026-08-07-wan-animate-2", "sources": null, "summary": "阿里通义 Wan-Animate-2（14B Diffusion Transformer）8月7日以 Apache 2.0 协议开源：直接吃原始视频潜变量，省去骨架姿态提取，Lite 版 24fps 可实时运行；内置 ComfyUI 原生节点，开箱即用；盲测超越可灵 MotionControl，适用于时尚服装动态展示、叙事插画角色演绎、虚拟主播与 Avatar 动作捕捉迁移等场景。", "reason": null, "tags": ["Wan-Animate-2","通义","开源","ComfyUI"], "topics": ["阿里巴巴","快手","ComfyUI"], "firsthand": false, "editor_pick": false},{"id": "2026-08-07-grok-imagine-image2-multiref-region-editing-design", "url": "https://www.transwonder.top/insights/#2026-08-07-grok-imagine-image2-multiref-region-editing-design", "title": "Grok Imagine Image 2.0：最多5图参考联合输入+区域Magic Wand精准编辑+透明背景导出，Arena全球第二", "date": "2026-08-07", "category": "模型", "source": "xAI官方 / Unite.AI", "src": "https://x.ai/news/grok-imagine-image-2", "sources": null, "summary": "xAI于8月7日正式发布Grok Imagine Image 2.0，以设计工作流为核心：最多5张参考图联合输入（一步合成免手动拼接）、区域Magic Wand精准编辑（只修目标区域）、前/背景精确分离+透明背景PNG导出；同时优化密排文字、版式、图标、游戏资产等密集布局渲染；发布日在两项主流图像Arena中排名全球第二（文生图+图像编辑双榜）。", "reason": null, "tags": ["xAI","Grok Imagine","Image 2.0","多参考输入"], "topics": ["图像生成","大模型"], "firsthand": true, "editor_pick": false},{"id": "2026-08-06-chatgpt-gpt56-luna-free-unlimited-sol-smarter", "url": "https://www.transwonder.top/insights/#2026-08-06-chatgpt-gpt56-luna-free-unlimited-sol-smarter", "title": "ChatGPT 全量发布 GPT-5.6 Luna：免费层无限制使用，SOL 推理能力显著提升", "date": "2026-08-06", "category": "模型", "source": "OpenAI 官方", "src": "https://openai.com/index/improving-gpt-5-6-sol-in-chatgpt/", "sources": null, "summary": "OpenAI 全量发布 GPT-5.6 Luna，面向所有 ChatGPT 用户开放免费无限制使用，SOL 逻辑推理与数学问题求解能力较前版本显著提升。", "reason": null, "tags": ["ChatGPT","GPT-5.6","Luna","OpenAI","免费","SOL推理"], "topics": ["OpenAI"], "firsthand": true, "editor_pick": false},{"id": "2026-08-06-adobe-chatgpt-plugin-photoshop-firefly-70tools", "url": "https://www.transwonder.top/insights/#2026-08-06-adobe-chatgpt-plugin-photoshop-firefly-70tools", "title": "Adobe发布免费ChatGPT统一插件：Photoshop/Firefly/Premiere等70+工具，@Adobe对话中直接唤醒", "date": "2026-08-06", "category": "产品", "source": "Adobe官方 / 9to5Mac", "src": "https://blog.adobe.com/en/publish/2026/08/06/introducing-adobe-chatgpt-create-edit-get-work-done-all-in-chatgpt", "sources": null, "summary": "Adobe在ChatGPT内上线免费统一插件，整合Photoshop/Firefly/Express/Premiere/Lightroom/Illustrator/InDesign/Acrobat/Adobe Stock共70+工具；任意ChatGPT对话中@Adobe即可调用，无需Creative Cloud订阅（免费Adobe账号即可使用Firefly文生图/文生视频）；同步宣布接入Claude与Microsoft Copilot，Slack/Google Gemini列入计划。", "reason": null, "tags": ["Adobe","ChatGPT","Photoshop","Firefly"], "topics": ["OpenAI","Adobe"], "firsthand": true, "editor_pick": false},{"id": "2026-08-11-anthropic-claude-for-creative-work-adobe-blender", "url": "https://www.transwonder.top/insights/#2026-08-11-anthropic-claude-for-creative-work-adobe-blender", "title": "Anthropic发布「Claude for Creative Work」：接入Adobe/Canva/Blender/Autodesk等六款创意软件，三所艺术学校参与课堂测试", "date": "2026-08-05", "category": "产品", "source": "Anthropic官方 / No Film School", "src": "https://www.anthropic.com/news/claude-for-creative-work", "sources": null, "summary": "Anthropic发布「Claude for Creative Work」专项功能集，将Claude深度接入Adobe、Canva、Blender、Autodesk等六款以上创意软件，并与三所艺术学校开展课堂测试合作，标志着AI助理进入「内嵌于设计师日常工具」阶段。", "reason": null, "tags": ["Anthropic","Claude","Adobe","Blender"], "topics": ["Anthropic","Adobe"], "firsthand": true, "editor_pick": false},{"id": "2026-08-05-qwen-image-3-0-pro", "url": "https://www.transwonder.top/insights/#2026-08-05-qwen-image-3-0-pro", "title": "阿里 Qwen-Image-3.0-Pro 发布：10px 文字渲染精度大幅提升，全球图像生成排名第五", "date": "2026-08-05", "category": "模型", "source": "Alibaba Qwen 官方 X / Alibaba Cloud Model Studio", "src": "https://help.aliyun.com/en/model-studio/qwen-image-3-0-pro", "sources": null, "summary": "阿里在 Qwen-Image-3.0（7月21日发布，已入池）基础上推出 Pro 版本，重点强化 10px 文字渲染精度（毛孔/发丝/微表情细节），注入更丰富的世界知识（地标/品牌/符号可信度），Qwen 团队自评全球排名第五。", "reason": null, "tags": ["Qwen","阿里巴巴","图像生成","Pro"], "topics": ["阿里巴巴","图像生成"], "firsthand": true, "editor_pick": false},{"id": "2026-08-05-google-veo-3-1-120s-audio-character", "url": "https://www.transwonder.top/insights/#2026-08-05-google-veo-3-1-120s-audio-character", "title": "Google发布Veo 3.1：120秒4K视频+原生音频+三参考图一致性，Gemini API付费预览即开", "date": "2026-08-05", "category": "模型", "source": "Google Developers Blog / Gemini API官方", "src": "https://developers.googleblog.com/en/introducing-veo-3-1-and-new-creative-capabilities-in-the-gemini-api/", "sources": null, "summary": "Google Veo 3.1于2026年8月5日登陆Gemini API付费预览：视频时长从8秒升至120秒，支持4K分辨率、原生立体声音频全链路生成（对话/音效/环境音）、跨场景角色一致性、竖版9:16，最多3张参考图引导生图到视频；Veo 3.1 Lite专注高并发与快速迭代，同步上线。", "reason": null, "tags": ["Veo 3.1","Google","AI视频","120秒"], "topics": ["Google","视频生成","音频与音乐"], "firsthand": true, "editor_pick": false},{"id": "2026-08-05-flux3-video-ga-20sec-native-audio", "url": "https://www.transwonder.top/insights/#2026-08-05-flux3-video-ga-20sec-native-audio", "title": "Black Forest Labs FLUX 3 Video 正式 GA：20 秒原生音频，Draft 模式 $0.06/秒，1080p 首发", "date": "2026-08-05", "category": "模型", "source": "Black Forest Labs 官方 / VentureBeat", "src": "https://venturebeat.com/ai/black-forest-labs-launches-flux-3-video-to-production-20-second-native-audio-videos/", "sources": null, "summary": "Black Forest Labs 于 2026-08-05 将 FLUX 3 Video 推送至生产可用（GA）：单次最长 20 秒视频生成，内置原生音频联动输出，Draft 模式定价 $0.06/s（约 ¥0.44/s），首发 1080p（2K/4K 已在路线图）；依托 FLUX API 及授权合作平台接入，ComfyUI Partner Nodes 预计跟进。", "reason": null, "tags": ["FLUX 3 Video","Black Forest Labs","视频生成","原生音频"], "topics": ["Black Forest Labs","视频生成","音频与音乐"], "firsthand": true, "editor_pick": false},{"id": "2026-08-04-openai-dalle-gpt-retirement-chatgpt-images", "url": "https://www.transwonder.top/insights/#2026-08-04-openai-dalle-gpt-retirement-chatgpt-images", "title": "ChatGPT官方DALL-E GPT将于8/30退役，图像生成入口统一迁往ChatGPT Images", "date": "2026-08-04", "category": "产品", "source": "OpenAI帮助中心 + CryptoBriefing + ITdaily", "src": "https://help.openai.com/en/articles/11084440-images-in-chatgpt", "sources": null, "summary": "OpenAI 宣布 8月30日关闭 ChatGPT 内的官方 DALL-E GPT，所有图像生成入口统一迁往「ChatGPT Images」（基于 gpt-image-1 和 gpt-image-1-mini）。", "reason": null, "tags": ["OpenAI","DALL-E","ChatGPT Images","gpt-image-1"], "topics": ["OpenAI","图像生成"], "firsthand": true, "editor_pick": false},{"id": "2026-08-04-claude-design-figma-saas-disruption", "url": "https://www.transwonder.top/insights/#2026-08-04-claude-design-figma-saas-disruption", "title": "Claude Design对Figma的冲击：AI原生设计工具如何重构SaaS设计软件格局", "date": "2026-08-04", "category": "行业", "source": "多家科技媒体综合", "src": "https://www.anthropic.com/news/claude-design-anthropic-labs", "sources": null, "summary": "Claude Design等AI原生设计入口对以Figma为代表的传统SaaS设计工具形成竞争冲击，引发设计工具格局重构的行业讨论", "reason": null, "tags": ["Claude Design","Figma","AI原生设计","SaaS","设计工具","格局变化"], "topics": ["Anthropic","Figma","设计工具"], "firsthand": true, "editor_pick": false},{"id": "2026-08-03-sensetime-sensenova-u15-lite-4k-8b-mot", "url": "https://www.transwonder.top/insights/#2026-08-03-sensetime-sensenova-u15-lite-4k-8b-mot", "title": "商汤SenseNova U1.5-Lite-Preview：8B-MoT架构，4K精准「指哪改哪」局部编辑，已开源，中文风格化直出", "date": "2026-08-03", "category": "模型", "source": "商汤官方 / 量子位", "src": "https://www.sensetime.com/", "sources": null, "summary": "商汤发布SenseNova U1.5-Lite-Preview：8B参数MoT（Mixture of Transformers）架构，核心突破为精准局部编辑（框选区域只改该区域、边缘自然融合）+ 4K原生输出 + 中文风格化，权重已开源，S级5+源齐发。", "reason": null, "tags": ["商汤","SenseNova","精准编辑","4K"], "topics": ["大模型"], "firsthand": true, "editor_pick": false},{"id": "2026-08-03-comfyui-v030-minimax-h3-grok-video", "url": "https://www.transwonder.top/insights/#2026-08-03-comfyui-v030-minimax-h3-grok-video", "title": "ComfyUI v0.30.0发布：原生支持MiniMax H3开源权重+Grok Imagine Video，4新节点6工作流", "date": "2026-08-03", "category": "产品", "source": "ComfyUI官方/ComfyUI Wiki/Kombitz", "src": "https://github.com/Comfy-Org/ComfyUI/releases/tag/v0.30.0", "sources": null, "summary": "ComfyUI v0.30.0正式发布，原生支持MiniMax H3音视频联合生成（含4个新节点与6套官方工作流）以及xAI Grok Imagine Video 1.5合作节点，MiniMax H3开源权重同日可用。", "reason": null, "tags": ["ComfyUI","v0.30.0","MiniMax H3","Grok","工作流","开源","本地运行"], "topics": ["ComfyUI","大模型","智能体与工作流"], "firsthand": true, "editor_pick": false},{"id": "2026-08-02-eu-ai-act-article50-ai-video-watermark", "url": "https://www.transwonder.top/insights/#2026-08-02-eu-ai-act-article50-ai-video-watermark", "title": "EU AI Act Article 50正式生效：AI生成视频须机器可读水印，多层标记义务全面启动", "date": "2026-08-02", "category": "行业", "source": "Cloud Security Alliance/Hard2bit/ai-eu-act.eu", "src": "https://hard2bit.com/en/blog/ai-act-article-50-ai-transparency-chatbots-deepfakes/", "sources": null, "summary": "EU AI Act Article 50（透明度义务）于2026年8月2日正式生效，所有AI生成图像、视频与音频须嵌入机器可读标记；AI视频水印子义务设4个月宽限期至12月2日。", "reason": null, "tags": ["EU AI Act","Article 50","AI水印","合规","透明度","视频标记","政策"], "topics": ["版权与政策"], "firsthand": true, "editor_pick": false},{"id": "2026-08-01-openai-astra-next-major-model-math-proofs", "url": "https://www.transwonder.top/insights/#2026-08-01-openai-astra-next-major-model-math-proofs", "title": "OpenAI预告下一代旗舰Astra：多智能体协同攻克十道历史数学难题，深度研究助手形态浮现", "date": "2026-08-01", "category": "模型", "source": "OpenAI官博 + The Decoder + BleepingComputer", "src": "https://openai.com/index/ten-advances-in-mathematics/", "sources": null, "summary": "OpenAI 以非常规方式预告下一代旗舰模型 Astra：不发布产品，而是在数学博客第三段点名，随后将 Astra 内部版本解决的十道历史级数学难题（群论/算子代数/格密码/量子复杂度等）推上 GitHub，附249页技术报告及完整 Lean 4 机器可验证证明。", "reason": null, "tags": ["OpenAI","Astra","多智能体","大模型"], "topics": ["OpenAI","大模型","智能体与工作流"], "firsthand": true, "editor_pick": false},{"id": "2026-08-01-liveedit-realtime-video-editing-eccv2026", "url": "https://www.transwonder.top/insights/#2026-08-01-liveedit-realtime-video-editing-eccv2026", "title": "LiveEdit（清华×港科大）：4步去噪实现12.66 FPS实时逐帧视频编辑，ECCV 2026", "date": "2026-08-01", "category": "论文", "source": "机器之心 / ECCV 2026", "src": "https://arxiv.org/abs/2606.26740", "sources": null, "summary": "清华×港科大联合提出LiveEdit，通过三阶段渐进蒸馏将双向扩散Transformer的视频编辑能力迁移至因果流式编辑器，仅需4步去噪即可实现12.66FPS实时逐帧视频编辑；ECCV 2026接收。", "reason": null, "tags": ["LiveEdit","实时视频编辑","清华大学","香港科技大学"], "topics": ["研究论文"], "firsthand": true, "editor_pick": false},{"id": "2026-08-01-google-earth-nano-banana-2-rollback", "url": "https://www.transwonder.top/insights/#2026-08-01-google-earth-nano-banana-2-rollback", "title": "谷歌地球集成Nano Banana 2生图功能不到24小时因「灾难场景」紧急撤回", "date": "2026-08-01", "category": "产品", "source": "新智元 / 新浪科技", "src": "https://finance.sina.com.cn/tech/digi/2026-08-01/doc-inikufhr4107884.shtml", "sources": null, "summary": "谷歌地球嵌入Nano Banana 2模型，允许用户用文字指令修改真实地理影像（支持2K/4K输出+精准文字渲染），但因用户生成逼真「灾难场景」并在社媒传播，上线不到24小时紧急下架。", "reason": null, "tags": ["谷歌地球","Nano Banana 2","撤回","AI合规"], "topics": ["Google","图像生成"], "firsthand": false, "editor_pick": false},{"id": "2026-07-31-siggraph-2026-test-of-time-award-deep-learning-character-animation", "url": "https://www.transwonder.top/insights/#2026-07-31-siggraph-2026-test-of-time-award-deep-learning-character-animation", "title": "SIGGRAPH 2026时间检验奖：2016港大角色动画深度学习框架提前十年押中物理AI方向", "date": "2026-07-31", "category": "论文", "source": "量子位 / 智源社区", "src": "https://www.qbitai.com/2026/07/464328.html", "sources": null, "summary": "SIGGRAPH 2026时间检验奖授予港大Taku Komura团队2016年发表的角色运动合成与编辑深度学习框架；该论文是首次将深度学习引入角色动画合成的先驱工作，对当前AI角色动画/物理AI运动生成有持续影响，GitHub项目已达8000+Star。", "reason": null, "tags": ["SIGGRAPH","时间检验奖","角色动画","深度学习"], "topics": ["研究论文"], "firsthand": false, "editor_pick": false},{"id": "2026-07-31-seedance-2-5-official-launch", "url": "https://www.transwonder.top/insights/#2026-07-31-seedance-2-5-official-launch", "title": "Seedance 2.5 正式发布：单次30秒直出、50素材多模态参考、Maya/Blender插件", "date": "2026-07-31", "category": "模型", "source": "量子位 / 新浪科技", "src": "https://www.qbitai.com/2026/07/464329.html", "sources": null, "summary": "字节跳动Seedance 2.5于7月31日正式发布，单次AI视频生成时长从10秒提升至30秒，支持50个多模态参考素材（图像+视频+文字混合），新增局部编辑，发布Maya/Blender插件打通影视工业流程；已全量上线即梦AI与豆包专业版。", "reason": null, "tags": ["Seedance 2.5","字节跳动","视频生成","30秒直出"], "topics": ["字节跳动","视频生成"], "firsthand": false, "editor_pick": false},{"id": "2026-07-31-minimax-h3-api-open-weights-launch", "url": "https://www.transwonder.top/insights/#2026-07-31-minimax-h3-api-open-weights-launch", "title": "MiniMax H3正式发布：全模态2K/15秒视频+原生立体声，开源权重8月首发ModelScope", "date": "2026-07-31", "category": "模型", "source": "MarkTechPost/ComfyUI官方/Civitai", "src": "https://www.marktechpost.com/2026/08/01/minimax-releases-minimax-h3-an-omni-modal-video-model-that-generates-15-second-2k-clips-with-native-stereo-audio/", "sources": null, "summary": "MiniMax发布H3（Hailuo 3.0），支持全模态输入生成最高2K分辨率、15秒、原生立体声视频，API当日可调用，开源权重首发ModelScope，ComfyUI同步支持。", "reason": null, "tags": ["MiniMax","H3","Hailuo","2K视频","原生立体声","开源","视频生成"], "topics": ["视频生成","大模型"], "firsthand": true, "editor_pick": false},{"id": "2026-07-31-a2-edit-reference-image-editing", "url": "https://www.transwonder.top/insights/#2026-07-31-a2-edit-reference-image-editing", "title": "A²-Edit（上交大×上海创智学院）：突破掩码精度限制，实现任意物体精准参考图引导局部编辑", "date": "2026-07-31", "category": "论文", "source": "机器之心 / arXiv", "src": "https://arxiv.org/html/2603.10685", "sources": null, "summary": "上交大×上海创智学院提出A²-Edit，通过混合Transformer专家路由与掩码退火训练策略，支持任意物体类别+任意精度掩码的参考图引导局部编辑；无论用精细掩码还是粗略框选都能保持参考图细节，适用于电商视觉合成、虚拟试穿、品牌局部替换等场景。", "reason": null, "tags": ["A²-Edit","参考图编辑","局部编辑","掩码"], "topics": ["研究论文"], "firsthand": true, "editor_pick": false},{"id": "2026-07-29-wonder-video-world-model-adobe-jhu", "url": "https://www.transwonder.top/insights/#2026-07-29-wonder-video-world-model-adobe-jhu", "title": "Wonder（Adobe Research+JHU）：解决控制漂移/记忆衰减/延迟膨胀的视频世界模型，单图→可游走3D一致性世界", "date": "2026-07-29", "category": "论文", "source": "arXiv 2607.26037 + alphaXiv", "src": "https://arxiv.org/html/2607.26037", "sources": null, "summary": "Adobe Research与Johns Hopkins大学提出Wonder视频世界模型：将相机运动转化为像素空间视觉线索（绕过神经隐式重建）；稀疏检索全保真历史帧解决记忆衰减；实时蒸馏学生模型解决延迟膨胀；可从单张图像或视频生成可交互、3D一致性、分钟级时长的沉浸式世界；6方向实时导航@16fps。", "reason": null, "tags": ["Adobe Research","Johns Hopkins","视频世界模型","3D一致性"], "topics": ["Adobe","3D 与世界模型","研究论文"], "firsthand": true, "editor_pick": false},{"id": "2026-07-29-google-lyria-35-flow-music-image-to-music", "url": "https://www.transwonder.top/insights/#2026-07-29-google-lyria-35-flow-music-image-to-music", "title": "Google Lyria 3.5发布：Flow工具新增图片转音乐功能，上传图像即可生成情绪匹配原创音乐", "date": "2026-07-29", "category": "模型", "source": "The Unsigned Network/TechTimes/musicnews", "src": "https://blog.google/innovation-and-ai/models-and-research/google-labs/lyria-3-5/", "sources": null, "summary": "Google发布Lyria 3.5音乐AI模型，在Google Labs Flow工具中集成「图片转音乐」功能，上传一张图像即可生成与其情绪、色彩、主题匹配的原创音乐，音质与音乐性显著提升。", "reason": null, "tags": ["Google","Lyria","AI音乐","图片转音乐","跨模态","Flow","音乐生成"], "topics": ["Google","音频与音乐"], "firsthand": true, "editor_pick": false},{"id": "2026-07-29-comfyui-v029-joyimageedit-mageflow-minimax-h3", "url": "https://www.transwonder.top/insights/#2026-07-29-comfyui-v029-joyimageedit-mageflow-minimax-h3", "title": "ComfyUI v0.29.0–0.29.2：JoyImageEdit/MageFlow/Uni3C ControlNet/MiniMax H3 Partner Nodes一次到位", "date": "2026-07-29", "category": "产品", "source": "GitHub Comfy-Org / docs.comfy.org", "src": "https://github.com/Comfy-Org/ComfyUI/releases/tag/v0.29.0", "sources": null, "summary": "ComfyUI v0.29.0–0.29.2原生集成JoyImageEdit图像指令编辑、MageFlow 4B、Uni3C ControlNet（Wan系列视频精准控制），v0.29.2追加MiniMax H3与Ideogram Partner Nodes。", "reason": null, "tags": ["ComfyUI","JoyImageEdit","MageFlow","Uni3C","ControlNet","Wan","MiniMax","工作流","视频控制"], "topics": ["ComfyUI","图像生成","大模型"], "firsthand": true, "editor_pick": false},{"id": "2026-07-28-lottie-creator-20-ai-motion-copilot-browser", "url": "https://www.transwonder.top/insights/#2026-07-28-lottie-creator-20-ai-motion-copilot-browser", "title": "Lottie Creator 2.0发布：浏览器内「After Effects」，Motion Copilot AI描述即生成动画，支持State Machine交互", "date": "2026-07-28", "category": "产品", "source": "LottieFiles官方X + ProductHunt", "src": "https://x.com/LottieFiles/status/2079748705868259728", "sources": null, "summary": "LottieFiles发布Lottie Creator 2.0：浏览器内完整运动设计工具，Motion Copilot AI可从自然语言描述直接生成关键帧动画；新增图形编辑器、State Machine可交互性系统；支持导入AI/EPS/PDF/PSD/SVG/AEP/FLA等16种格式；无需安装，导出生产级Lottie动画；定价$19.99/月。", "reason": null, "tags": ["Lottie","动效设计","AI动画","Motion Copilot"], "topics": ["视频生成","设计工具"], "firsthand": true, "editor_pick": false},{"id": "2026-07-28-jarvishub-canvas-native-creative-agent", "url": "https://www.transwonder.top/insights/#2026-07-28-jarvishub-canvas-native-creative-agent", "title": "JarvisHub：首个画布原生多模态创作Agent框架，可编辑画布同时作工作区+外部记忆+动作空间", "date": "2026-07-28", "category": "论文", "source": "机器之心 / arXiv:2607.23588", "src": "https://arxiv.org/abs/2607.23588", "sources": null, "summary": "JarvisHub（arXiv:2607.23588，7/26提交，7/28开源）是首个以「可编辑画布」同时作为Agent工作空间、外部记忆和动作空间的多模态创作框架；三层架构（画布状态→协议桥接→Agent运行时）支持持续规划、迭代生成、版本管理；用户可随时介入、引导、检查Agent创作过程，突破孤立工具调用限制，迈向可持续人机协同创作自动化。", "reason": null, "tags": ["JarvisHub","画布Agent","多模态创作","开源"], "topics": ["智能体与工作流","研究论文"], "firsthand": true, "editor_pick": false},{"id": "2026-07-26-andrew-ng-open-source-desktop-agent", "url": "https://www.transwonder.top/insights/#2026-07-26-andrew-ng-open-source-desktop-agent", "title": "吴恩达发布100%开源个人桌面Agent：本地运行、隐私保护、跨应用操作", "date": "2026-07-26", "category": "产品", "source": "量子位/机器之心", "src": "https://www.qbitai.com/2026/07/460892.html", "sources": null, "summary": "吴恩达团队发布完全开源的个人桌面Agent框架，支持本地运行、隐私保护与跨浏览器及桌面应用操作，量子位、机器之心同步报道。", "reason": null, "tags": ["吴恩达","开源","桌面Agent","本地运行","隐私","自动化"], "topics": ["智能体与工作流"], "firsthand": false, "editor_pick": false},{"id": "2026-07-26-ai-mona-lisa-four-models-iteration", "url": "https://www.transwonder.top/insights/#2026-07-26-ai-mona-lisa-four-models-iteration", "title": "四大 AI 8 轮迭代临摹蒙娜丽莎：谁最接近原作？过度修正如何发生", "date": "2026-07-26", "category": "案例", "source": "AIERA / 量子位", "src": "https://aiera.com.cn/2026/07/26/other/admin/105826/", "sources": null, "summary": "测试者用 GPT-4o/Midjourney/SD/Flux 各自完成 8 轮蒙娜丽莎临摹迭代，记录过度修正与风格漂移全过程", "reason": null, "tags": ["AI绘画","蒙娜丽莎","模型对比","迭代修正","Midjourney","DALL-E","Stable Diffusion","图像生成"], "topics": ["Midjourney","Stability AI","图像生成"], "firsthand": false, "editor_pick": false},{"id": "2026-07-25-vivix-a1-w1-realtime-interactive-model", "url": "https://www.transwonder.top/insights/#2026-07-25-vivix-a1-w1-realtime-interactive-model", "title": "Vivix灵动时刻发布首个实时互动多模态模型A1/W1，全双工全身表演+单卡10000 video tokens/s", "date": "2026-07-25", "category": "模型", "source": "量子位/极客公园/网易/36氪", "src": "https://www.qbitai.com/2026/07/460174.html", "sources": null, "summary": "Vivix灵动时刻A1实现AI角色全身表演与全双工实时交互，TTFF低于0.6秒；W1支持统一流式交互叙事，单卡突破10000 video tokens/s，成本0.2美元/小时。", "reason": null, "tags": ["Vivix","实时互动","多模态","全双工","AI角色","视频生成"], "topics": ["视频生成","大模型"], "firsthand": false, "editor_pick": false},{"id": "2026-07-25-qinghua-automia-cvpr26-mirror-illusion-3d", "url": "https://www.transwonder.top/insights/#2026-07-25-qinghua-automia-cvpr26-mirror-illusion-3d", "title": "清华 AutoMIA 入选 CVPR'26：输入两张图片 76 秒生成镜像错觉 3D 艺术品，支持 3D 打印", "date": "2026-07-25", "category": "论文", "source": "新智元 / 量子位 / 36氪 / CSDN / 51CTO", "src": "https://aiera.com.cn/2026/07/25/other/admin/105716/", "sources": null, "summary": "清华 AutoMIA 系统可将任意两张图片自动生成镜像错觉 3D 物体，76 秒完成，支持直接 3D 打印输出，已入选 CVPR 2026", "reason": null, "tags": ["清华","AutoMIA","CVPR 2026","3D错觉","镜像艺术","3D打印","AI装置艺术"], "topics": ["3D 与世界模型","研究论文"], "firsthand": false, "editor_pick": false},{"id": "2026-07-25-claude-code-80pct-prompt-removal-context-windows", "url": "https://www.transwonder.top/insights/#2026-07-25-claude-code-80pct-prompt-removal-context-windows", "title": "Claude Code 删去 80% 系统提示词：Anthropic 工程师解释新模型为何不再需要旧规则", "date": "2026-07-25", "category": "观点", "source": "Anthropic工程师", "src": "https://x.com/trq212/status/2080710971228918066", "sources": null, "summary": "Anthropic工程师Thariq Shihipar详解Claude Code为Opus 5/Fable 5删除80%系统提示词的原因：旧规则是为旧模型设计的，新模型已能根据代码自主判断", "reason": null, "tags": ["Claude","Code","上下文工程","系统提示词"], "topics": ["Anthropic","智能体与工作流"], "firsthand": false, "editor_pick": false},{"id": "2026-07-24-wenxiaobi5pro-long-content-30k-no-drift", "url": "https://www.transwonder.top/insights/#2026-07-24-wenxiaobi5pro-long-content-30k-no-drift", "title": "问小白5 Pro：文件沙箱+结构化记忆，跨三天写三万字不漂移", "date": "2026-07-24", "category": "产品", "source": "量子位/IT之家", "src": "https://www.qbitai.com/2026/07/458687.html", "sources": null, "summary": "元石科技发布问小白5 Pro长内容生成引擎，通过文件沙箱与结构化记忆管理解决AI长任务前后矛盾问题，支持跨三天持续写作三万字而不发生主线漂移。", "reason": null, "tags": ["长内容生成","问小白","结构化记忆","AI写作","学术辅助","元石科技"], "topics": ["大模型"], "firsthand": false, "editor_pick": false},{"id": "2026-07-24-midjourney-v82-default-aesthetics", "url": "https://www.transwonder.top/insights/#2026-07-24-midjourney-v82-default-aesthetics", "title": "Midjourney V8.2 正式成为默认版本：美学与个性化升级，sref 风格板精准度提升，新增 Big Batch Draft Mode", "date": "2026-07-24", "category": "模型", "source": "Midjourney 官方更新页", "src": "https://docs.midjourney.com/hc/en-us/articles/32199405667853-Version", "sources": null, "summary": "Midjourney V8.2 于 7 月 24 日正式成为默认版本，美学表现和 sref 风格板个性化一致性较 V8.1 显著提升，新增 Big Batch Draft Mode 加速风格探索。", "reason": null, "tags": ["Midjourney","V8.2","图像生成","美学升级","个性化","sref","风格一致性"], "topics": ["Midjourney","图像生成"], "firsthand": true, "editor_pick": false},{"id": "2026-07-24-midjourney-costar-acquisition-design-app", "url": "https://www.transwonder.top/insights/#2026-07-24-midjourney-costar-acquisition-design-app", "title": "Midjourney 收购天文 App Co-Star：全部 24 名员工加入，创始人 Banu Guler 出任设计总监，独立图像 App 建设提速", "date": "2026-07-24", "category": "行业", "source": "Bloomberg / TechCrunch / Gizmodo / Engadget", "src": "https://techcrunch.com/2026/07/24/midjourney-acquired-the-astrology-app-co-star/", "sources": null, "summary": "Midjourney 全盘收购天文 App Co-Star，全部 24 名员工加入，创始人 Banu Guler 出任设计总监，信号 Midjourney 独立消费端 App 建设明显提速。", "reason": null, "tags": ["Midjourney","Co-Star","收购","独立App","设计总监","移动端","图像生成"], "topics": ["Midjourney","图像生成","行业与融资"], "firsthand": false, "editor_pick": false},{"id": "2026-07-24-inworldview-worldscore-benchmark-top", "url": "https://www.transwonder.top/insights/#2026-07-24-inworldview-worldscore-benchmark-top", "title": "兔展智能UniWorld-View登顶WorldScore：单图生成任意视角视频，全开源适配昇腾", "date": "2026-07-24", "category": "论文", "source": "量子位 / 凤凰科技 / ITBear", "src": "https://www.qbitai.com/2026/07/460041.html", "sources": null, "summary": "兔展智能联合北京大学提出UniWorld-View，在李飞飞团队WorldScore评测中登顶（国产首次），双流架构支持单图生成任意视角视频，代码权重全开源，已适配国产昇腾算力。", "reason": null, "tags": ["世界模型","UniWorld","3D视角","开源","国产AI","兔展智能","WorldScore"], "topics": ["3D 与世界模型","研究论文"], "firsthand": false, "editor_pick": false},{"id": "2026-07-24-claude-opus-5-half-price-fable", "url": "https://www.transwonder.top/insights/#2026-07-24-claude-opus-5-half-price-fable", "title": "Anthropic Claude Opus 5 正式发布：$5/$25 每百万 Token（Fable 5 半价），1M 上下文，接近旗舰级性能", "date": "2026-07-24", "category": "模型", "source": "Axios / Anthropic 官方", "src": "https://www.anthropic.com/news/claude-opus-5", "sources": null, "summary": "Anthropic 正式发布 Claude Opus 5，输入 $5 / 输出 $25 每百万 Token（约为 Fable 5 半价），1M token 上下文，性能接近旗舰级。", "reason": null, "tags": ["Anthropic","Claude Opus 5","大模型","长上下文","AI工具","性价比"], "topics": ["Anthropic","大模型"], "firsthand": true, "editor_pick": false},{"id": "2026-07-23-runway-media-router-generative", "url": "https://www.transwonder.top/insights/#2026-07-23-runway-media-router-generative", "title": "Runway Media Router 发布：首个生成式媒体模型路由器，按质量 / 速度 / 成本需求自动调度最优 AI 模型", "date": "2026-07-23", "category": "产品", "source": "TechCrunch / Runway 官方", "src": "https://runwayml.com/changelog", "sources": null, "summary": "Runway 发布 Media Router，首个生成式媒体模型路由器，创作者设定优先级后系统自动在 Gen-4.5 / Aleph 2.0 等模型间选择最优路径。", "reason": null, "tags": ["Runway","Media Router","模型路由","生成式媒体","AI工具链","Gen-4.5","Aleph"], "topics": ["Runway"], "firsthand": true, "editor_pick": false},{"id": "2026-07-23-meitu-hatch-catch-ai-image-builder", "url": "https://www.transwonder.top/insights/#2026-07-23-meitu-hatch-catch-ai-image-builder", "title": "美图发起 Hatch Catch AI 影像 Builder 挑战赛：1 亿元总奖金，寻找已上线 AI 原生影像产品，单项最高 500 万投资", "date": "2026-07-23", "category": "案例", "source": "量子位", "src": "https://www.qbitai.com/2026/07/457223.html", "sources": null, "summary": "美图发起 Meitu Hatch Catch 产品挑战赛，总规模 1 亿元，面向已上线且有种子用户的 AI 原生影像产品，聚焦 AI 影像赛道，单项最高投资 500 万元。", "reason": null, "tags": ["美图","Hatch Catch","AI影像","AI创业","投资挑战赛","B2C","影像AI"], "topics": ["图像生成"], "firsthand": false, "editor_pick": false},{"id": "2026-07-23-flux-3-multimodal-image-video-audio", "url": "https://www.transwonder.top/insights/#2026-07-23-flux-3-multimodal-image-video-audio", "title": "Black Forest Labs FLUX 3 发布：首个统一图像/视频/音频的多模态前沿模型，20 秒视频生成，Canva/Krea/Magnific 等伙伴首批接入", "date": "2026-07-23", "category": "模型", "source": "Black Forest Labs 官方博客/VentureBeat/GlobeNewswire/TechTimes（多源）", "src": "https://bfl.ai/blog/flux-3", "sources": null, "summary": "Black Forest Labs 发布 FLUX 3，首个统一图像生成、20 秒视频与原生音频输出的多模态流模型，Canva 等首批生态伙伴同步接入。", "reason": null, "tags": ["FLUX 3","Black Forest Labs","多模态生成","图像生成","视频生成","音频","设计工具生态"], "topics": ["Black Forest Labs","图像生成","视频生成"], "firsthand": true, "editor_pick": false},{"id": "2026-07-23-beijing-future-ai-art-immersive-exhibition", "url": "https://www.transwonder.top/insights/#2026-07-23-beijing-future-ai-art-immersive-exhibition", "title": "「未来之境」AI艺术沉浸大展北京启幕：情绪感知引擎+XR影像，近20项AI体验展至12月", "date": "2026-07-23", "category": "赛展", "source": "中新网/腾讯/京报网/新京报", "src": "https://www.chinanews.com.cn/cul/2026/07-24/10665978.shtml", "sources": null, "summary": "第五届仲夏艺术季旗舰展览「未来之境」于北京启幕，融合情绪感知智能引擎与XR虚拟影像，近20项AI+艺术沉浸体验，展至2026年12月。", "reason": null, "tags": ["AI艺术展览","沉浸式","XR","情绪感知","北京","装置艺术","仲夏艺术季"], "topics": ["艺术与展览"], "firsthand": true, "editor_pick": false},{"id": "2026-07-22-waic-2026-agent-era-design-tool-trend", "url": "https://www.transwonder.top/insights/#2026-07-22-waic-2026-agent-era-design-tool-trend", "title": "量子位 WAIC 2026 十大趋势：Agent 演示取代大模型发布，设计工具 Agent 化成明确风向标", "date": "2026-07-22", "category": "观点", "source": "量子位 / 51CTO", "src": "https://www.qbitai.com/2026/07/456974.html", "sources": null, "summary": "量子位发布 WAIC 2026 十大趋势：83 家参展主体聚焦 AI Agent，设计工具 Agent 化从个别案例变为行业风向标，从「对话式 AI」进入「交付式 Agent」。", "reason": null, "tags": ["WAIC 2026","AI趋势","Agent化","设计工具","量子位","行业洞察"], "topics": ["设计工具","大模型","智能体与工作流"], "firsthand": false, "editor_pick": false},{"id": "2026-07-22-videochat3-nanjing-unified-video-understanding", "url": "https://www.transwonder.top/insights/#2026-07-22-videochat3-nanjing-unified-video-understanding", "title": "南京大学 VideoChat3 全栈开源：4B 参数统一短 / 长 / 流式视频理解，登顶 HuggingFace 趋势榜", "date": "2026-07-22", "category": "论文", "source": "南京大学 / 上海AI Lab / 机器之心 / 新无尽 / alphaxiv", "src": "https://www.xinfinite.net/t/topic/21076", "sources": null, "summary": "南京大学联合多校发布 VideoChat3，全栈开源 4B 参数视频理解模型，统一处理短视频 / 长视频 / 流式视频三类任务，登顶 HuggingFace 趋势榜。", "reason": null, "tags": ["南京大学","VideoChat3","视频理解","开源模型","HuggingFace","多模态","短/长/流式视频"], "topics": ["研究论文"], "firsthand": true, "editor_pick": false},{"id": "2026-07-22-tripo-ai-siggraph-keynote-3d-intelligence", "url": "https://www.transwonder.top/insights/#2026-07-22-tripo-ai-siggraph-keynote-3d-intelligence", "title": "Tripo AI SIGGRAPH 2026主题演讲：茶壶测试从「能展示」到「能制造」，3D生成AI迈向交互式世界智能", "date": "2026-07-22", "category": "观点", "source": "SIGGRAPH 2026官方日程 / Tripo AI官方X", "src": "https://s2026.conference-schedule.org/presentation/?id=keysp_103", "sources": null, "summary": "Tripo AI在SIGGRAPH 2026发表题为「茶壶测试：从展示走向制造」的主题演讲，阐述如何从孤立静态资产生成迈向动态环境/智能体系统/世界建模，展示Tripo全栈生成生态在游戏/3D打印/机器人仿真等工业管线中的应用。", "reason": null, "tags": ["Tripo AI","SIGGRAPH 2026","3D生成","世界模型","游戏资产","交互式AI","3D智能"], "topics": ["3D 与世界模型","研究论文"], "firsthand": true, "editor_pick": false},{"id": "2026-07-22-tencent-miora-design-agent-launch", "url": "https://www.transwonder.top/insights/#2026-07-22-tencent-miora-design-agent-launch", "title": "腾讯 Miora 国际版全量上线：生图 / 视频 / 3D / UI 四子 Agent 协同，登顶 Product Hunt 日榜，新用户赠 1000 积分", "date": "2026-07-22", "category": "产品", "source": "量子位 / 腾讯新闻 / 虎嗅 / 新浪财经 / 数字生命卡兹克 / 机器之心", "src": "https://news.qq.com/rain/a/20260722A08LY600", "sources": null, "summary": "腾讯首个设计 Agent 平台 Miora 国际版全量上线，生图 / 视频 / 3D / UI 四子 Agent 同一画布协同交付，首日登顶 Product Hunt 日榜第一。", "reason": null, "tags": ["腾讯","Miora","设计Agent","多Agent协同","品牌视觉","Product Hunt","生成式AI"], "topics": ["腾讯","图像生成","3D 与世界模型"], "firsthand": false, "editor_pick": false},{"id": "2026-07-22-siggraph2026-art-paper-award-resonance", "url": "https://www.transwonder.top/insights/#2026-07-22-siggraph2026-art-paper-award-resonance", "title": "SIGGRAPH 2026最佳艺术论文：Resonance——AI生成艺术中共鸣感的形式化研究", "date": "2026-07-22", "category": "论文", "source": "SIGGRAPH 2026官方", "src": "https://s2026.conference-schedule.org/presentation/?id=artpl_169&sess=sess226", "sources": null, "summary": "SIGGRAPH 2026最佳艺术论文奖授予论文Resonance，探索AI生成艺术中「共鸣感」的形式化定义与计算方法", "reason": null, "tags": ["SIGGRAPH 2026","最佳艺术论文","Resonance","AI艺术","生成艺术","学术"], "topics": ["研究论文"], "firsthand": true, "editor_pick": false},{"id": "2026-07-22-microsoft-mage-flow-4b-native-resolution", "url": "https://www.transwonder.top/insights/#2026-07-22-microsoft-mage-flow-4b-native-resolution", "title": "微软开源 Mage-Flow：4B 参数原生任意分辨率图像生成与编辑，GenEval 0.90 开源最高，MIT 许可可商用", "date": "2026-07-22", "category": "模型", "source": "机器之心/ComfyUI Wiki/arXiv/GitHub/HuggingFace（多源）", "src": "https://comfyui-wiki.com/zh/news/2026-07-22-mage-flow-microsoft", "sources": null, "summary": "Mage-Flow 4B 参数原生支持 512–2048 任意宽高比文生图与指令式编辑，GenEval 0.90 为开源最优，Turbo 版 0.59 秒/张。", "reason": null, "tags": ["微软","Mage-Flow","开源","图像生成","图像编辑","原生分辨率","ComfyUI","MIT许可"], "topics": ["ComfyUI","图像生成"], "firsthand": false, "editor_pick": false},{"id": "2026-07-22-figma-ceo-dylan-field-ai-design-tailwind", "url": "https://www.transwonder.top/insights/#2026-07-22-figma-ceo-dylan-field-ai-design-tailwind", "title": "Figma CEO Dylan Field：AI是设计顺风，设计师需成为Out-of-Distribution力量，Q1营收+46%", "date": "2026-07-22", "category": "观点", "source": "Stratechery / Lenny's Newsletter / Yahoo Finance / TechCrunch", "src": "https://stratechery.com/2026/an-interview-with-figma-ceo-dylan-field-about-design-and-ai/", "sources": null, "summary": "Figma CEO Dylan Field在Stratechery等多个访谈中阐明：AI处理分布内均值，设计师需以品味和文化洞察成为Out-of-Distribution力量；Figma Q1营收+46%至3.334亿美元印证AI扩大了设计工具市场。", "reason": null, "tags": ["Figma","Dylan Field","设计价值","AI顺风","Out-of-Distribution","设计教育","英文源"], "topics": ["Figma","设计教育","行业与融资"], "firsthand": false, "editor_pick": false},{"id": "2026-07-22-cafa-ai-traditional-culture-digital", "url": "https://www.transwonder.top/insights/#2026-07-22-cafa-ai-traditional-culture-digital", "title": "央美教授吴小虎策展：AI 数字技术让千年壁画走出高墙，联合国教科文组织报告背书", "date": "2026-07-22", "category": "案例", "source": "央广网 / 新华社", "src": "https://news.cnr.cn/native/gd/20260722/t20260722_527723412.shtml", "sources": null, "summary": "中央美术学院教授吴小虎策划展览，借助 AI 数字技术让传统壁画突破展馆局限面向大众，项目获联合国教科文组织报告背书。", "reason": null, "tags": ["中央美术学院","AI传统文化","壁画数字化","非遗AI","文化遗产","联合国教科文组织","艺术展览"], "topics": ["艺术与展览","设计教育"], "firsthand": false, "editor_pick": false},{"id": "2026-07-21-waic-ai-creators-night-human-irreplaceable", "url": "https://www.transwonder.top/insights/#2026-07-21-waic-ai-creators-night-human-irreplaceable", "title": "WAIC 2026「AI 创作者之夜」：郭靖宇 / 陈楸帆等讨论 AI 时代创作边界，「真实、情感与在场」是不可替代核心", "date": "2026-07-21", "category": "观点", "source": "腾讯新闻 / 哈佛商业评论", "src": "https://news.qq.com/rain/a/20260721A02JNH00", "sources": null, "summary": "郭靖宇、陈楸帆等创作者在 WAIC 2026 圆桌讨论认为，真实体验、情感共鸣与在场感是 AI 时代人类不可替代的核心价值。", "reason": null, "tags": ["WAIC 2026","AI创作边界","人类不可替代","版权保护","郭靖宇","陈楸帆","科幻作家","创作者圆桌"], "topics": ["版权与政策"], "firsthand": false, "editor_pick": false},{"id": "2026-07-21-qwen-image-30-infographic-text-render", "url": "https://www.transwonder.top/insights/#2026-07-21-qwen-image-30-infographic-text-render", "title": "阿里通义 Qwen-Image-3.0 发布：4.5K 超长指令、10px 文字清晰渲染、12 国语言+100 余种艺术风格，图像生成从「画质比拼」转向「生产力交付」", "date": "2026-07-21", "category": "模型", "source": "AIHub/量子位/网易/无界AI/aitop100/机器之心（多源）", "src": "https://www.aihub.cn/news/qwen-image-3-0-release/", "sources": null, "summary": "Qwen-Image-3.0 文字渲染精度达 10px 可辨级别，4.5K 超长指令支持多区域版面控制，12 语言覆盖多语言出版需求。", "reason": null, "tags": ["阿里","Qwen-Image","图像生成","文字渲染","信息图表","多语言","艺术风格","通义万象"], "topics": ["阿里巴巴","图像生成"], "firsthand": true, "editor_pick": false},{"id": "2026-07-21-nvidia-dlss5-neural-rendering-siggraph", "url": "https://www.transwonder.top/insights/#2026-07-21-nvidia-dlss5-neural-rendering-siggraph", "title": "NVIDIA DLSS 5 SIGGRAPH 2026揭秘：三路AI模型分工神经渲染，让艺术家「主导最终画面」，秋季上线", "date": "2026-07-21", "category": "模型", "source": "WCCFTech / guru3d / TechPowerUp（多源）", "src": "https://wccftech.com/nvidia-dlss-5-hands-over-full-control-to-artists-to-direct-the-final-frame/", "sources": null, "summary": "NVIDIA在SIGGRAPH 2026披露DLSS 5技术细节：三路独立AI模型（A/B/C）分别负责全局光照/次表面散射/材质细节等渲染环节，开发者可对场景/角色/关卡单独分配不同模型；紧凑扩散Transformer架构，单GPU实现4K/60fps/亚16ms延迟；DLSS 5秋季正式上线游戏。", "reason": null, "tags": ["NVIDIA","DLSS 5","神经渲染","AI渲染","SIGGRAPH 2026","GPU","3D设计"], "topics": ["3D 与世界模型","研究论文"], "firsthand": false, "editor_pick": false},{"id": "2026-07-21-gemini-36-flash-three-models", "url": "https://www.transwonder.top/insights/#2026-07-21-gemini-36-flash-three-models", "title": "Google三模型齐发：Gemini 3.6 Flash降本17%、3.5 Flash-Lite最廉价、3.5 Flash Cyber安全专项，同步宣告Gemini 4预训练开启", "date": "2026-07-21", "category": "模型", "source": "Google DeepMind官方Blog / TechCrunch / 9to5Google（多源）", "src": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-6-flash-3-5-flash-lite-3-5-flash-cyber/", "sources": null, "summary": "Google DeepMind同日发布三款新Flash模型：Gemini 3.6 Flash是主力workhorse、token成本降低17%，3.5 Flash-Lite为最低价轻量版（$0.30/百万输入），3.5 Flash Cyber为安全漏洞专用精调版限政府伙伴；同步宣布Gemini 4已进入预训练阶段，Gemini 3.5 Pro仍在伙伴测试中。", "reason": null, "tags": ["Google","Gemini","Flash模型","Gemini 4","多模态","设计工具生态","降本"], "topics": ["Google","设计工具"], "firsthand": true, "editor_pick": false},{"id": "2026-07-21-autodesk-siggraph-3dsmax-gaussian-splat-maya-motionmaker", "url": "https://www.transwonder.top/insights/#2026-07-21-autodesk-siggraph-3dsmax-gaussian-splat-maya-motionmaker", "title": "Autodesk SIGGRAPH 2026：3ds Max支持可编辑3D高斯溅射，Maya MotionMaker支持「带自己的数据」定制生成动作模型", "date": "2026-07-21", "category": "产品", "source": "Autodesk官方新闻室 / Creative Bloq（多源）", "src": "https://adsknews.autodesk.com/en/news/autodesk-siggraph-2026-animation-vfx-tools/", "sources": null, "summary": "Autodesk在SIGGRAPH 2026宣布：3ds Max将3D Gaussian Splats变为完全可编辑资产（支持建模/动画/场景构建）；Maya MotionMaker新增BYOD（Bring Your Own Data），团队可用自家动捕/手绑动作数据训练自定义生成动作模型；Flow Production Tracking引入实时媒体审阅与协同标注。", "reason": null, "tags": ["Autodesk","3ds Max","Maya","Gaussian Splat","MotionMaker","3D动画","VFX","SIGGRAPH 2026","自定义动作模型"], "topics": ["3D 与世界模型","研究论文"], "firsthand": true, "editor_pick": false},{"id": "2026-07-20-wondershare-waic-ai-video-matrix", "url": "https://www.transwonder.top/insights/#2026-07-20-wondershare-waic-ai-video-matrix", "title": "万兴科技WAIC 2026三箭齐发：万兴剧厂无限画布+Agent助手、Filmora.AI专业画布工作台、Virbo AI视频带货全球首秀", "date": "2026-07-20", "category": "产品", "source": "新浪科技 / 搜狐 / 腾讯新闻（多源）", "src": "https://finance.sina.com.cn/jjxw/2026-07-17/doc-iniiccuy4225769.shtml", "sources": null, "summary": "万兴科技在WAIC 2026展示三条AI视频产品线：万兴剧厂推出无限画布节点式影视工作流和专属AI Agent助手，Filmora.AI提供专业TVC画布级工作台，Virbo AI定位短视频电商带货场景，CEO宣布将推动AI影视规模化商业化。", "reason": null, "tags": ["万兴科技","Filmora","Virbo","万兴剧厂","WAIC 2026","AI视频矩阵","无限画布","内容创作"], "topics": ["视频生成","智能体与工作流"], "firsthand": false, "editor_pick": false},{"id": "2026-07-20-springer-genai-architecture-course-china", "url": "https://www.transwonder.top/insights/#2026-07-20-springer-genai-architecture-course-china", "title": "Springer实证研究：Midjourney融入高校建筑PBL课程，弱基础学生受益显著高于强基础组", "date": "2026-07-20", "category": "论文", "source": "Springer / 学术期刊", "src": "https://link.springer.com/", "sources": null, "summary": "Springer发表实证研究，分析Midjourney融入高校建筑设计PBL课程的效果，发现弱基础学生在AI辅助下的学习收益显著高于强基础学生，两组均报告正向体验。", "reason": null, "tags": ["Springer","Midjourney","建筑设计","PBL","高校教育","实证研究","生成式AI教育"], "topics": ["Midjourney","设计教育","研究论文"], "firsthand": true, "editor_pick": false},{"id": "2026-07-20-qwen-audio-30-tts", "url": "https://www.transwonder.top/insights/#2026-07-20-qwen-audio-30-tts", "title": "阿里Qwen-Audio-3.0-TTS发布：Flash版首包延迟300ms级实时交互，Plus版登顶Artificial Analysis全球榜单，覆盖20种中文方言", "date": "2026-07-20", "category": "模型", "source": "新浪科技 / IT之家 / 量子位 / 机器之心（多源）", "src": "https://finance.sina.com.cn/roll/2026-07-20/doc-iniinfpx2969101.shtml", "sources": null, "summary": "阿里发布语音合成大模型Qwen-Audio-3.0-TTS，含实时交互Flash版（首包延迟300ms级）和高质量生成Plus版；支持细粒度标签控制、freestyle指令遵循、20种中文方言及多语种，Plus版登顶全球权威榜单Artificial Analysis。", "reason": null, "tags": ["阿里","Qwen-Audio","语音合成","TTS","实时交互","方言支持","多模态"], "topics": ["阿里巴巴","音频与音乐"], "firsthand": false, "editor_pick": false},{"id": "2026-07-20-nvidia-synthetic-video-detector-nim", "url": "https://www.transwonder.top/insights/#2026-07-20-nvidia-synthetic-video-detector-nim", "title": "NVIDIA SIGGRAPH 2026发布Synthetic Video Detector NIM：22ms检测AI生成视频，92%精准度，API即用", "date": "2026-07-20", "category": "产品", "source": "NVIDIA官方Blog/WCCFTech/Digital Trends（多源）", "src": "https://blogs.nvidia.com/blog/siggraph-news-2026/", "sources": null, "summary": "NVIDIA在SIGGRAPH 2026发布Synthetic Video Detector NIM，22ms/帧检测速度、92%精准率，通过标准API调用，填补AI生成视频内容真实性验证空白，与Adobe Content Credentials、C2PA标准形成生态配合。", "reason": null, "tags": ["NVIDIA","AI视频检测","SIGGRAPH 2026","内容真实性"], "topics": ["视频生成","研究论文"], "firsthand": true, "editor_pick": false},{"id": "2026-07-20-nvidia-siggraph-mcp-creative-tools-cosmos-edge", "url": "https://www.transwonder.top/insights/#2026-07-20-nvidia-siggraph-mcp-creative-tools-cosmos-edge", "title": "NVIDIA SIGGRAPH 2026：MCP接入Adobe/Blender/Houdini/Unreal等全链路创作工具，同步发布Cosmos 3 Edge 4B边缘世界模型", "date": "2026-07-20", "category": "产品", "source": "NVIDIA官方Blog / Claypier / 掘金 / OmniTools（多源）", "src": "https://blogs.nvidia.com/blog/siggraph-news-2026/", "sources": null, "summary": "NVIDIA在SIGGRAPH 2026宣布Adobe Firefly/Creative Cloud、Canva/Affinity Designer、Blender、SideFX Houdini 22、Unreal Editor等主流创作软件全面支持MCP协议，AI Agent可直接在DCC工具内执行场景检查/材质标记/版本导出等操作；同步开源Cosmos 3 Edge（4B参数，VANTAGE-Bench同参数同类第一），可在Jetson Thor/RTX PRO/GeForce RTX上本地运行。", "reason": null, "tags": ["NVIDIA","MCP","Adobe Firefly","Blender","Houdini","Unreal","Cosmos 3 Edge","边缘世界模型","SIGGRAPH 2026","AI工具链"], "topics": ["Adobe","3D 与世界模型","智能体与工作流"], "firsthand": true, "editor_pick": false},{"id": "2026-07-20-nvidia-siggraph-21-papers-motionbricks-artifixer", "url": "https://www.transwonder.top/insights/#2026-07-20-nvidia-siggraph-21-papers-motionbricks-artifixer", "title": "NVIDIA SIGGRAPH 2026发表21篇论文：MotionBricks实时动作AI同时驱动数字角色与人形机器人，ArtiFixer将杂乱3D扫描转为完整场景", "date": "2026-07-20", "category": "论文", "source": "NVIDIA Research / SIGGRAPH 2026（多源）", "src": "https://blogs.nvidia.com/blog/siggraph-news-2026/", "sources": null, "summary": "NVIDIA在SIGGRAPH 2026发表21篇研究论文：MotionBricks基于35万段动作片段训练的实时动作AI可同时驱动屏幕数字角色与Unitree G1人形机器人；ArtiFixer将杂乱现实3D扫描自动转为可用虚拟场景；新雪/沙/弹性体物理求解器一并开源。", "reason": null, "tags": ["NVIDIA","MotionBricks","ArtiFixer","SIGGRAPH 2026","动作AI","3D场景重建","开源研究","物理模拟"], "topics": ["3D 与世界模型","研究论文"], "firsthand": true, "editor_pick": false},{"id": "2026-07-20-hkust-surreality-mr-ai-exhibition", "url": "https://www.transwonder.top/insights/#2026-07-20-hkust-surreality-mr-ai-exhibition", "title": "港科大SURREALITY：全球首个大规模MR×AI数字艺术跨城展，70+国际艺术家50件作品", "date": "2026-07-20", "category": "赛展", "source": "HKUST官方/Yahoo Finance", "src": "https://hkust.edu.hk/news/hkust-and-hkustgz-present-surreality-mr-ai-digital-art-cross-city-exhibition", "sources": null, "summary": "香港科技大学与HKUST(GZ)联合呈现「SURREALITY」MR×AI数字艺术跨城展，汇集70+国际艺术家50件作品，同步在香港与广州两城举行，展期至7月31日。", "reason": null, "tags": ["港科大","SURREALITY","MR","混合现实","AI数字艺术展","国际展览","香港","广州","跨城"], "topics": ["艺术与展览"], "firsthand": true, "editor_pick": false},{"id": "2026-07-20-hidream-vivago-r1-unlimited-duration-video-agent", "url": "https://www.transwonder.top/insights/#2026-07-20-hidream-vivago-r1-unlimited-duration-video-agent", "title": "智象未来vivago R1：全球首个无限时长多模态创作智能体，打破AI视频15-30秒上限，商用内容可用率85%", "date": "2026-07-20", "category": "产品", "source": "量子位 / 环球网 / 凤凰科技 / 雷锋网 / Chinaz（多源）", "src": "https://www.qbitai.com/2026/07/455434.html", "sources": null, "summary": "智象未来（HiDream.ai）在WAIC 2026发布基于自研HD-AgentOS的vivago R1，突破行业15-30秒视频时长限制，支持任意时长视频连续连贯生成，叙事逻辑/画面风格/角色IP全程统一，商业可用率达85%，适配短剧/专题片/品牌宣传片等长周期创作场景。", "reason": null, "tags": ["智象未来","HiDream","vivago R1","AI视频生成","无限时长","HD-AgentOS","WAIC 2026","商业内容"], "topics": ["视频生成","智能体与工作流"], "firsthand": false, "editor_pick": false},{"id": "2026-07-20-gmi-happyhorse-wujiecraft-waic", "url": "https://www.transwonder.top/insights/#2026-07-20-gmi-happyhorse-wujiecraft-waic", "title": "GMI Cloud × 阿里HappyHorse「无界造物节」WAIC 2026决赛收官：AI Coding+AI Video双赛道，MaaS平台支撑全云端AI创作", "date": "2026-07-20", "category": "案例", "source": "量子位 / 36氪 / 新浪（多源）", "src": "https://www.qbitai.com/2026/07/455593.html", "sources": null, "summary": "GMI Cloud联合阿里巴巴HappyHorse在WAIC 2026举办「无界造物节」全民AI创作赛事决赛收官；AI Coding赛道产出智能运营Agent和低代码工具，AI Video赛道依托HappyHorse全链路能力完成云端脚本到成片，展示MaaS平台支撑非技术用户实现完整AI视频创作的落地路径。", "reason": null, "tags": ["GMI Cloud","HappyHorse","无界造物节","WAIC 2026","AI Coding","AI Video","MaaS","内容创作"], "topics": ["阿里巴巴"], "firsthand": false, "editor_pick": false},{"id": "2026-07-19-siggraph2026-gimmbo-best-paper", "url": "https://www.transwonder.top/insights/#2026-07-19-siggraph2026-gimmbo-best-paper", "title": "SIGGRAPH 2026最佳论文：GimmBO——贝叶斯优化引导的交互式生成模型权重融合框架（多伦多大学/CMU）", "date": "2026-07-19", "category": "论文", "source": "ACM SIGGRAPH Blog / s2026.siggraph.org", "src": "https://blog.siggraph.org/2026/05/siggraph-2026-technical-papers-awards-best-papers-honorable-mentions-and-test-of-time.html", "sources": null, "summary": "SIGGRAPH 2026最佳论文（洛杉矶7月19–23日，1120+投稿）：GimmBO（多伦多大学/CMU/Vector Institute）——实时交互式框架，将两个或多个生成图像模型的权重集融合，通过贝叶斯优化在融合空间导航，由用户偏好点击引导。", "reason": null, "tags": ["SIGGRAPH 2026","模型融合","贝叶斯优化","生成式图像"], "topics": ["研究论文"], "firsthand": true, "editor_pick": false},{"id": "2026-07-19-nvidia-dlss45-blender-siggraph2026", "url": "https://www.transwonder.top/insights/#2026-07-19-nvidia-dlss45-blender-siggraph2026", "title": "NVIDIA DLSS 4.5 进驻 Blender：SIGGRAPH 2026 宣布，随 Blender 5.3「今秋」发布", "date": "2026-07-19", "category": "产品", "source": "NVIDIA 官方 / Blender Foundation", "src": "https://www.nvidia.com/en-us/events/siggraph/", "sources": null, "summary": "NVIDIA 在 SIGGRAPH 2026 开幕首日宣布 DLSS 4.5（含多帧生成与 AI 超分辨率）将随 Blender 5.3「今秋」正式发布，为开源 3D 工具引入 AI 加速渲染。", "reason": null, "tags": ["NVIDIA","DLSS 4.5","Blender","SIGGRAPH 2026"], "topics": ["研究论文"], "firsthand": true, "editor_pick": false},{"id": "2026-07-19-kunlun-waic-world-model-era", "url": "https://www.transwonder.top/insights/#2026-07-19-kunlun-waic-world-model-era", "title": "昆仑万维WAIC宣布「2026世界模型元年」：Matrix-Game 3.5单卡实时生成，Riemann-1.0机器人世界模型同步发布", "date": "2026-07-19", "category": "行业", "source": "机器之心 / 新浪财经 / Chinaz / 同花顺（多源）", "src": "https://finance.sina.com.cn/stock/t/2026-07-20/doc-iniimzfz3033851.shtml", "sources": null, "summary": "昆仑万维董事长方汉在WAIC 2026世界模型专场论坛宣布2026为「世界模型元年」；发布Matrix-Game 3.5（Patch级记忆注入、单卡实时生成）和Riemann-1.0机器人世界模型，同步升级Mureka v9.5与O3音乐模型，预判AI音乐与视频将成大众创作工具。", "reason": null, "tags": ["昆仑万维","世界模型","Matrix-Game","Riemann","WAIC 2026","AI音乐","Mureka","多模态"], "topics": ["3D 与世界模型","音频与音乐"], "firsthand": false, "editor_pick": false},{"id": "2026-07-19-flux2-rtx-nvfp4-fp8-siggraph-optimization", "url": "https://www.transwonder.top/insights/#2026-07-19-flux2-rtx-nvfp4-fp8-siggraph-optimization", "title": "NVIDIA × Black Forest Labs FLUX.2 RTX优化：NVFP4量化2.7×加速+55%显存减少，SIGGRAPH同步推进ComfyUI集成", "date": "2026-07-19", "category": "产品", "source": "NVIDIA官方Blog", "src": "https://blogs.nvidia.com/blog/siggraph-news-2026/", "sources": null, "summary": "NVIDIA与Black Forest Labs在SIGGRAPH 2026发布FLUX.2 RTX专项优化：FP8量化VRAM减少40%/性能提升40%，NVFP4量化进一步实现2.7×加速和55%显存减少，ComfyUI深度集成同步推进，让消费级RTX显卡可流畅运行FLUX.2工作流。", "reason": null, "tags": ["FLUX.2","NVIDIA","RTX","NVFP4","FP8量化","ComfyUI","Black Forest Labs","SIGGRAPH 2026","图像生成","本地部署"], "topics": ["Black Forest Labs","ComfyUI","图像生成"], "firsthand": true, "editor_pick": false},{"id": "2026-07-19-figma-make-github-push-integration", "url": "https://www.transwonder.top/insights/#2026-07-19-figma-make-github-push-integration", "title": "Figma Make正式连接GitHub仓库：设计师可直接从画布提交界面代码修改，打通设计到生产「最后一公里」", "date": "2026-07-19", "category": "产品", "source": "Figma官方帮助文档", "src": "https://help.figma.com/hc/en-us/articles/35463818346647-Push-from-Figma-Make-to-GitHub", "sources": null, "summary": "Figma Make新增GitHub双向集成，设计师可在画布中直接导入Git仓库、修改界面并提交代码变更，将视觉编辑器转变为可连接真实代码仓库的开发端口。", "reason": null, "tags": ["Figma Make","GitHub","代码交付","设计工作流"], "topics": ["Figma","智能体与工作流"], "firsthand": true, "editor_pick": false},{"id": "2026-07-19-autodesk-flow-studio-ai-rigging-wonder3d", "url": "https://www.transwonder.top/insights/#2026-07-19-autodesk-flow-studio-ai-rigging-wonder3d", "title": "Autodesk SIGGRAPH 2026：Flow Studio新增AI自动绑骨架+神经层，Wonder 3D文本/图片一步生成3D角色，Maya AI运动全线升级", "date": "2026-07-19", "category": "产品", "source": "Autodesk官方新闻室 / 3DVF.com / Creative Bloq（多源）", "src": "https://www.autodesk.com/campaigns/siggraph-2026", "sources": null, "summary": "Autodesk在SIGGRAPH 2026展示Flow Studio重大更新：AI Rigging移除角色制作最复杂技术环节自动绑骨架，Wonder 3D支持文字或参考图一步生成完整3D角色，Maya新增AI辅助四足动物运动生成。", "reason": null, "tags": ["Autodesk","Flow Studio","AI Rigging","Wonder 3D","Maya","SIGGRAPH 2026","3D生成","角色动画"], "topics": ["3D 与世界模型","研究论文"], "firsthand": true, "editor_pick": false},{"id": "2026-07-18-vggrpo-geometry-video-generation", "url": "https://www.transwonder.top/insights/#2026-07-18-vggrpo-geometry-video-generation", "title": "VGGRPO（ECCV 2026 Google）：4D 隐空间几何奖励强化学习，解决视频扩散模型几何漂移", "date": "2026-07-18", "category": "论文", "source": "机器之心 / QQ新闻", "src": "https://news.qq.com/rain/a/20260717A0B3CG00", "sources": null, "summary": "Google 团队提出 VGGRPO，通过隐空间 4D 几何奖励进行强化学习后训练，解决视频扩散模型几何漂移与摄像机轨迹不稳定问题，已被 ECCV 2026 收录。", "reason": null, "tags": ["VGGRPO","ECCV 2026","视频生成","几何一致性"], "topics": ["Google","视频生成","研究论文"], "firsthand": false, "editor_pick": false},{"id": "2026-07-18-sensenova-u1-pro-waic-8k-delivery", "url": "https://www.transwonder.top/insights/#2026-07-18-sensenova-u1-pro-waic-8k-delivery", "title": "商汤SenseNova U1 Pro @ WAIC 2026：首个「交付级」多模态创作模型，原生8K直出，200位美院评审盲测六成图可直接交付客户", "date": "2026-07-18", "category": "模型", "source": "新智元/新浪财经/36氪（多源 6+源）", "src": "https://aiera.com.cn/2026/07/19/other/admin/104701/", "sources": null, "summary": "商汤在WAIC 2026发布SenseNova U1 Pro，基于NEO-unify统一架构，原生8K输出，200位美院学生与设计师盲评六成作品「可直接交付客户」，在文字精准度与东方美学维度对标GPT Image 2并部分胜出。", "reason": null, "tags": ["商汤SenseNova","U1 Pro","WAIC 2026","交付级设计"], "topics": ["大模型","设计教育"], "firsthand": false, "editor_pick": false},{"id": "2026-07-18-claude-code-artifacts-mcp-connectors", "url": "https://www.transwonder.top/insights/#2026-07-18-claude-code-artifacts-mcp-connectors", "title": "Claude Code Artifacts 支持 MCP 连接器：生成的看板可接实时数据，不同权限看到不同内容", "date": "2026-07-18", "category": "产品", "source": "新智元", "src": "https://aiera.com.cn/2026/07/18/other/admin/104408/", "sources": null, "summary": "Anthropic 为 Claude Code Artifacts 新增 MCP 连接器调用能力，生成的看板/仪表盘页面可连接实时数据，不同查看者按各自权限拉取不同内容，7 月 16 日上线。", "reason": null, "tags": ["Claude Code","Artifacts","MCP连接器","实时数据"], "topics": ["Anthropic","智能体与工作流"], "firsthand": false, "editor_pick": false},{"id": "2026-07-17-wan-streamer-v02-550ms-realtime", "url": "https://www.transwonder.top/insights/#2026-07-17-wan-streamer-v02-550ms-realtime", "title": "阿里通义 Wan-Streamer v0.2：端到端 AI 视频通话延迟降至 550ms，听/看/说/演统一进单一 Transformer", "date": "2026-07-17", "category": "模型", "source": "IT之家 / 虎嗅 / 同花顺", "src": "https://www.ithome.com/0/978/125.htm", "sources": null, "summary": "阿里通义实验室发布 Wan-Streamer v0.2，将听、看、说、演统一进单个 Transformer，端到端交互延迟约 550ms，较 v0.1 大幅改善。", "reason": null, "tags": ["阿里通义","Wan-Streamer","实时交互","多模态"], "topics": ["阿里巴巴","视频生成"], "firsthand": false, "editor_pick": false},{"id": "2026-07-17-siggraph-2026-preview-diffusion-tv", "url": "https://www.transwonder.top/insights/#2026-07-17-siggraph-2026-preview-diffusion-tv", "title": "SIGGRAPH 2026 开幕预告：「Diffusion TV」扩散模型沉浸装置与 AI 创作伙伴工作坊", "date": "2026-07-17", "category": "赛展", "source": "PR Newswire（SIGGRAPH 官方）", "src": "https://www.prnewswire.com/news-releases/siggraph-2026-puts-ai-to-work-as-a-creative-partner-across-research-art-and-industry-302828862.html", "sources": null, "summary": "SIGGRAPH 2026 会前公告披露两大 AI 亮点：将扩散模型实时可视化的沉浸装置「Diffusion TV」，以及「视频生成视觉语言」专题研讨工作坊，大会 7 月 19—23 日于洛杉矶举行。", "reason": null, "tags": ["SIGGRAPH 2026","AI艺术装置","扩散模型","学术会议"], "topics": ["研究论文","艺术与展览"], "firsthand": true, "editor_pick": false},{"id": "2026-07-17-minimax-h3-waic-unified-multimodal", "url": "https://www.transwonder.top/insights/#2026-07-17-minimax-h3-waic-unified-multimodal", "title": "MiniMax H3 @ WAIC 2026：下一代多模态生成模型，统一图像/视频/声音生成", "date": "2026-07-17", "category": "模型", "source": "每日经济新闻 / 新浪财经", "src": "https://finance.sina.com.cn/roll/2026-07-17/doc-iniiaxpa4241254.shtml", "sources": null, "summary": "MiniMax 在 WAIC 2026 首次公开 H3 模型，定位为将图像、视频、声音生成统一进单一架构的下一代多模态生成模型，现场标注「Coming Soon」。", "reason": null, "tags": ["MiniMax","H3","WAIC 2026","多模态生成"], "topics": ["大模型"], "firsthand": false, "editor_pick": false},{"id": "2026-07-17-mentalthink-svg-spatial-reasoning", "url": "https://www.transwonder.top/insights/#2026-07-17-mentalthink-svg-spatial-reasoning", "title": "MentalThink：让AI用SVG「脑补草图」做空间推理，MindCube三维测试76分超越GPT-5的56分", "date": "2026-07-17", "category": "论文", "source": "科技行者", "src": "https://www.techwalker.com/2026/0717/3193553.shtml", "sources": null, "summary": "北邮×中科院×StepFun提出MentalThink，让多模态大模型在推理中生成SVG草图「看图思考」，MindCube三维旋转测试76分超越GPT-5的56分，仅用7B参数模型。", "reason": null, "tags": ["SVG推理","空间推理","北邮","arXiv"], "topics": ["OpenAI","3D 与世界模型","研究论文"], "firsthand": false, "editor_pick": false},{"id": "2026-07-17-kimi-k3-28t-open-weight-multimodal", "url": "https://www.transwonder.top/insights/#2026-07-17-kimi-k3-28t-open-weight-multimodal", "title": "月之暗面Kimi K3：2.8万亿参数全球最大开源多模态模型，原生文字/图像/视频三模态，权重7/27全量开源", "date": "2026-07-17", "category": "模型", "source": "Moonshot官方/新华网（多源）", "src": "https://www.kimi.com/blog/kimi-k3", "sources": null, "summary": "月之暗面发布Kimi K3，2.8万亿参数MoE架构，原生三模态（文字/图像/视频）统一，100万Token超长上下文，权重7月27日全量开源，Apache 2.0协议允许商业使用。", "reason": null, "tags": ["Kimi K3","月之暗面","开源多模态","MoE架构"], "topics": ["大模型"], "firsthand": true, "editor_pick": false},{"id": "2026-07-17-gemini-35-pro-delay-confirmed", "url": "https://www.transwonder.top/insights/#2026-07-17-gemini-35-pro-delay-confirmed", "title": "Gemini 3.5 Pro 确认延期：前端 UI 像素级一次生成能力仍在打磨", "date": "2026-07-17", "category": "模型", "source": "新浪财经 / 36氪", "src": "https://finance.sina.com.cn/stock/usstock/c/2026-07-17/doc-inihzvzs1344436.shtml", "sources": null, "summary": "谷歌 Gemini 3.5 Pro 原定 7 月 17 日发布已确认延期，主因是编程能力仍在打磨，10 名现任和前任员工参与确认。", "reason": null, "tags": ["Google","Gemini","前端生成","模型动态"], "topics": ["Google"], "firsthand": false, "editor_pick": false},{"id": "2026-07-16-tiangong-skyproduction-ai-drama-director-workflow", "url": "https://www.transwonder.top/insights/#2026-07-16-tiangong-skyproduction-ai-drama-director-workflow", "title": "天工短剧工作台 SkyProduction：Agent 智能分镜先规划站位再生成视频，3 部 AI 短剧 7 天均破百万美元", "date": "2026-07-16", "category": "产品", "source": "中新网 / 36氪 / AIbase（多源）", "src": "https://www.chinanews.com/cj/2026/07-16/10660511.shtml", "sources": null, "summary": "昆仑万维发布天工短剧工作台 SkyProduction，Agent 智能分镜可在视频生成前规划角色站位与机位；DramaWave 平台 3 部 AI 短剧 7 天内均破百万美元流水。", "reason": null, "tags": ["天工AI","AI短剧","分镜Agent","视频生成"], "topics": ["视频生成","智能体与工作流"], "firsthand": true, "editor_pick": false},{"id": "2026-07-16-moworld-3d-huaweicloud", "url": "https://www.transwonder.top/insights/#2026-07-16-moworld-3d-huaweicloud", "title": "魔芯科技MoWorld 3D世界模型正式上线华为云：单张照片生成完整3D场景，全栈国产NPU（昇腾），向全行业开放", "date": "2026-07-16", "category": "模型", "source": "量子位 / 人民网 / China Daily（多源）", "src": "https://www.qbitai.com/2026/07/455517.html", "sources": null, "summary": "国家具身智能应用中试基地联合华为云、魔芯科技发布MoWorld 3D世界模型，支持从单张照片生成完整3D场景，全栈基于国产NPU（昇腾）构建，正式上线华为云向全行业开放。", "reason": null, "tags": ["魔芯科技","MoWorld","3D生成","华为云","国产NPU","昇腾","3D场景"], "topics": ["3D 与世界模型"], "firsthand": false, "editor_pick": false},{"id": "2026-07-16-google-vids-gemini-omni-personal-avatar", "url": "https://www.transwonder.top/insights/#2026-07-16-google-vids-gemini-omni-personal-avatar", "title": "Google Vids 引入 Gemini Omni 视频生成与个人 AI 数字人", "date": "2026-07-16", "category": "产品", "source": "Google Blog / TechCrunch", "src": "https://blog.google/products-and-platforms/products/workspace/gemini-omni-personal-avatars/", "sources": null, "summary": "Google Workspace 正式推出个人 AI 数字人，用户录制一次自拍和声音样本，即可在 Vids 中以本人形象出镜任意视频，并内嵌 SynthID 隐形水印。", "reason": null, "tags": ["Google","Gemini Omni","AI数字人","视频生成"], "topics": ["Google","视频生成"], "firsthand": true, "editor_pick": false},{"id": "2026-07-16-canva-google-gemini-connected-app", "url": "https://www.transwonder.top/insights/#2026-07-16-canva-google-gemini-connected-app", "title": "Canva深度接入Google Gemini：@Canva对话即生成品牌视觉，Magic Layers AI图层分离可独立编辑", "date": "2026-07-16", "category": "产品", "source": "Canva官方新闻室/量子位/优设（多源）", "src": "https://www.canva.com/newsroom/news/google-gemini/", "sources": null, "summary": "Google将Canva作为Connected App引入Workspace，@Canva指令即可调用Gemini生成品牌视觉并直接插入文档；Canva同步推出Magic Layers AI图层分离功能，每层可独立编辑。", "reason": null, "tags": ["Canva","Google Gemini","Magic Layers","品牌视觉"], "topics": ["Google"], "firsthand": true, "editor_pick": false},{"id": "2026-07-15-xmax-ai-x2-real-time-interactive-video", "url": "https://www.transwonder.top/insights/#2026-07-15-xmax-ai-x2-real-time-interactive-video", "title": "Xmax AI X2.0发布：端侧实时交互视频生成，延迟低于300ms", "date": "2026-07-15", "category": "模型", "source": "量子位/36氪/凤凰科技等", "src": "https://www.qbitai.com/2026/07/450130.html", "sources": null, "summary": "Xmax AI发布X2.0模型，实现端侧实时交互视频生成，延迟低于300ms，是端侧推理能力在视频生成场景的重要突破。", "reason": null, "tags": ["实时视频生成","端侧推理","Xmax","交互视频"], "topics": ["视频生成"], "firsthand": false, "editor_pick": false},{"id": "2026-07-15-pixverse-game-ai-interactive-engine", "url": "https://www.transwonder.top/insights/#2026-07-15-pixverse-game-ai-interactive-engine", "title": "PixVerse进军交互娱乐：Game AI引擎支持实时视频驱动游戏场景", "date": "2026-07-15", "category": "产品", "source": "PixVerse官方博客/Yahoo Finance等", "src": "https://pixverse.ai/en/blog/pixverse-closes-series-c-extension-and-announces-expansion-into-interactive-entertainment", "sources": null, "summary": "PixVerse宣布完成C轮融资延伸轮并进军交互娱乐，发布Game AI引擎，支持实时视频生成驱动游戏场景与角色行为。", "reason": null, "tags": ["PixVerse","游戏AI","交互娱乐","视频生成"], "topics": ["视频生成"], "firsthand": true, "editor_pick": false},{"id": "2026-07-15-jenny-wen-cursor", "url": "https://www.transwonder.top/insights/#2026-07-15-jenny-wen-cursor", "title": "Anthropic 核心设计负责人 Jenny Wen 跳槽 Cursor 出任设计主管", "date": "2026-07-15", "category": "行业", "source": "新智元", "src": "https://aiera.com.cn/2026/07/15/other/admin/103918/", "sources": null, "summary": "Anthropic 核心设计负责人 Jenny Wen 官宣跳槽，出任 AI 编程工具 Cursor 设计主管；她此前在 Figma 任设计总监，在 Anthropic 期间主导 Claude 产品设计与 Cowork 智能体愿景。Cursor 母公司 Anysphere 此前已被 SpaceXAI 收购。", "reason": null, "tags": ["Cursor","Anthropic","设计领导力","AI编程工具"], "topics": ["Anthropic","行业与融资"], "firsthand": false, "editor_pick": false},{"id": "2026-07-15-comfyui-v0280", "url": "https://www.transwonder.top/insights/#2026-07-15-comfyui-v0280", "title": "ComfyUI v0.28.0发布：原生视频节点与性能优化", "date": "2026-07-15", "category": "产品", "source": "ComfyUI官方GitHub", "src": "https://github.com/comfyanonymous/ComfyUI/releases/tag/v0.28.0", "sources": null, "summary": "ComfyUI发布v0.28.0版本，新增原生视频处理节点并优化整体运行性能，进一步强化其作为AI创作工作流核心枢纽的地位。", "reason": null, "tags": ["ComfyUI","工作流","视频生成","开源"], "topics": ["ComfyUI","视频生成","智能体与工作流"], "firsthand": true, "editor_pick": false},{"id": "2026-07-15-ant-lingbot-vision-boundary-pretraining", "url": "https://www.transwonder.top/insights/#2026-07-15-ant-lingbot-vision-boundary-pretraining", "title": "蚂蚁 LingBot-Vision：边界驱动视觉预训练新范式，10 亿参数超越 70 亿参数 DINOv3，权重开源", "date": "2026-07-15", "category": "论文", "source": "蚂蚁集团 / arXiv / 腾讯新闻", "src": "https://news.qq.com/rain/a/20260715A03MP800", "sources": null, "summary": "蚂蚁集团提出「掩码边界建模」预训练范式，10 亿参数 ViT-g 在深度估计基准上超越 70 亿参数 DINOv3，透明物体深度估计达 RMSE 0.010，权重已开源。", "reason": null, "tags": ["视觉预训练","深度估计","蚂蚁集团","开源模型","arXiv","ECCV"], "topics": ["研究论文"], "firsthand": false, "editor_pick": false},{"id": "2026-07-14-canva-code-2-ai-website-builder-free", "url": "https://www.transwonder.top/insights/#2026-07-14-canva-code-2-ai-website-builder-free", "title": "Canva Code 2.0：AI 建网站能力向 2.65 亿月活用户全量免费开放，生成速度提升 75%", "date": "2026-07-14", "category": "产品", "source": "VentureBeat", "src": "https://venturebeat.com/technology/canva-launches-code-2-0-offering-ai-website-building-to-every-user-including-free-accounts", "sources": null, "summary": "Canva 将 AI 代码生成功能向全部 2.65 亿月活用户（含免费账户）开放，生成速度提升 75%，从提示到上线时间压缩 30%。", "reason": null, "tags": ["Canva","AI建站","设计工具","免费功能","原型制作"], "topics": ["设计工具"], "firsthand": false, "editor_pick": false},{"id": "2026-07-13-tongji-tencent-skill-contest", "url": "https://www.transwonder.top/insights/#2026-07-13-tongji-tencent-skill-contest", "title": "同济×腾讯云TT设计学院Skill创新应用大赛收官：全国26所高校近300名选手将设计痛点封装为AI智能体工具", "date": "2026-07-13", "category": "赛展", "source": "中国日报/新浪科技/腾讯新闻", "src": "https://cn.chinadaily.com.cn/a/202607/13/WS6a54a7f5a310d709c2fbd41f.html", "sources": null, "summary": "同济大学与腾讯云联合举办的TT设计学院Skill创新应用大赛于2026年7月13日收官，全国26所高校近300名选手参赛，选手将真实设计流程痛点封装为AI智能体（Skill）工具，7支战队获奖，AI Agent正式进入高校设计教育的实践评估体系。", "reason": null, "tags": ["AI教育","同济大学","设计竞赛","AI智能体"], "topics": ["腾讯","智能体与工作流","设计教育"], "firsthand": true, "editor_pick": false},{"id": "2026-07-13-sjtu-world-narrative", "url": "https://www.transwonder.top/insights/#2026-07-13-sjtu-world-narrative", "title": "上海交大发布WNM世界叙事模型：创作者与AI协同构建可编辑、可交互的叙事时空，针对AI视频生成「不可控」瓶颈", "date": "2026-07-13", "category": "论文", "source": "人民网/上海交大", "src": "https://sh.people.com.cn/n2/2026/0713/c134768-41637589.html", "sources": null, "summary": "上海交通大学于2026年7月13日发布WNM（World Narrative Model）世界叙事模型，支持创作者与AI协同构建可编辑、可交互的叙事时空，专门针对现有AI视频生成在叙事连贯性和创作可控性上的瓶颈，面向影视级视频媒体制作场景。", "reason": null, "tags": ["世界模型","叙事生成","AI视频","上海交大"], "topics": ["视频生成","3D 与世界模型","研究论文"], "firsthand": false, "editor_pick": false},{"id": "2026-07-13-sensetime-sensenova-vision", "url": "https://www.transwonder.top/insights/#2026-07-13-sensetime-sensenova-vision", "title": "商汤开源SenseNova-Vision统一视觉大模型：目标检测、图像分割、深度预测、3D重建四大任务原生统一单一架构，超越Vision Banana", "date": "2026-07-13", "category": "模型", "source": "IT之家/虎嗅/新智元/腾讯新闻/TechNode/GitHub/Hugging Face", "src": "https://github.com/OpenSenseNova/SenseNova-Vision", "sources": null, "summary": "商汤科技于2026年7月13日开源SenseNova-Vision，将检测、分割、深度预测、3D重建四大视觉任务原生统一到单一多模态架构，性能超越Vision Banana，以CC BY-NC 4.0协议开放，教学环境可自由使用。", "reason": null, "tags": ["商汤","视觉大模型","开源","图像理解"], "topics": ["大模型","3D 与世界模型"], "firsthand": true, "editor_pick": false},{"id": "2026-07-13-pku-pra-image-gen", "url": "https://www.transwonder.top/insights/#2026-07-13-pku-pra-image-gen", "title": "北大+深势科技提出PRA自回归图像生成框架：16维低维中间状态+并行解码，135M参数在像素空间超越1.9B基线", "date": "2026-07-13", "category": "论文", "source": "机器之心/深势科技", "src": "https://www.jiqizhixin.com/articles/2026-07-13", "sources": null, "summary": "北京大学联合深势科技提出PRA自回归图像生成框架，采用16维低维中间状态与并行解码机制，135M参数规模在像素空间的表现超过1.9B参数基线，绕开传统视觉tokenizer，目前处于理论突破阶段，工程化程度待评估。", "reason": null, "tags": ["自回归","图像生成","北京大学","深度学习"], "topics": ["图像生成","研究论文"], "firsthand": false, "editor_pick": false},{"id": "2026-07-13-deepmind-genception-video-general-visual-learner", "url": "https://www.transwonder.top/insights/#2026-07-13-deepmind-genception-video-general-visual-learner", "title": "DeepMind发布GenCeption：统一视觉理解与生成的视频通用模型", "date": "2026-07-13", "category": "论文", "source": "genception.github.io/TechTimes/机器之心等", "src": "https://genception.github.io/", "sources": null, "summary": "DeepMind发布GenCeption，是首个在单一模型中统一视频理解与生成能力的视觉通用学习器，在多项基准上取得领先结果。", "reason": null, "tags": ["DeepMind","视频理解","视觉生成","多模态"], "topics": ["Google","研究论文"], "firsthand": true, "editor_pick": false},{"id": "2026-07-12-claude-code-effort", "url": "https://www.transwonder.top/insights/#2026-07-12-claude-code-effort", "title": "Anthropic官方回应Claude Code「变笨」争议：默认Effort降档非模型退化，ultracode可恢复xhigh火力", "date": "2026-07-12", "category": "产品", "source": "Anthropic/36氪/新智元/火星财经", "src": "https://m.36kr.com/p/3892222176574211", "sources": null, "summary": "Anthropic官方确认Claude Code近期「响应变浅」源于默认Effort从high静默调至medium，非模型退化；ultracode模式启用xhigh档位与自主子智能体调度可恢复深度推理能力。", "reason": null, "tags": ["Claude Code","Effort","AI辅助编码","Anthropic"], "topics": ["Anthropic"], "firsthand": false, "editor_pick": false},{"id": "2026-07-11-ghost-font-ai-crack", "url": "https://www.transwonder.top/insights/#2026-07-11-ghost-font-ai-crack", "title": "Ghost Font幽灵字体72小时攻防：字母编码为运动轨迹，人眼秒懂而GPT-5.6等主流大模型全部失读，72小时后凭一句方向提示GPT-5.6完整破解", "date": "2026-07-11", "category": "案例", "source": "新智元/36氪/Hacker News/Tom's Guide/ghostfont.org", "src": "https://ghostfont.org", "sources": null, "summary": "设计师Eric Lu与Mixfont将字母编码为移动轨迹，发布24小时内达14.7M播放；GPT-5.6、Gemini、Claude均无法读取；72小时后一句方向提示使GPT-5.6完整破解，完成设计与AI认知对抗的首个完整回合制案例。", "reason": null, "tags": ["字体设计","AI认知","大模型局限","人机对抗"], "topics": ["OpenAI","大模型","设计工具"], "firsthand": true, "editor_pick": false},{"id": "2026-07-10-siggraph-experience-hall", "url": "https://www.transwonder.top/insights/#2026-07-10-siggraph-experience-hall", "title": "SIGGRAPH 2026 Experience Hall全面公布：五大板块，Art Gallery「In-Betweens」含4件AI艺术", "date": "2026-07-10", "category": "赛展", "source": "SIGGRAPH官方", "src": "https://s2026.siggraph.org/siggraph-2026-experience-hall-invites-attendees-to-step-inside-the-future-of-interaction-and-storytelling/", "sources": null, "summary": "SIGGRAPH 2026 Experience Hall公布五大板块完整阵容，Art Gallery「In-Betweens」12件入选作品中含4件AI艺术，探索AI与记忆、身份及人机关系，7月19-23日洛杉矶举行。", "reason": null, "tags": ["SIGGRAPH","AI艺术","展览","计算机图形"], "topics": ["研究论文","艺术与展览"], "firsthand": true, "editor_pick": false},{"id": "2026-07-10-meta-muse-image-pulled", "url": "https://www.transwonder.top/insights/#2026-07-10-meta-muse-image-pulled", "title": "Meta Muse Image上线不足一周因opt-out默认机制遭SAG-AFTRA/CAA联合声讨后下架", "date": "2026-07-10", "category": "行业", "source": "财联社/Deadline/Hollywood Reporter/TechWeez/Forbes/MacRumors/Variety", "src": "https://www.hollywoodreporter.com/business/digital/meta-pulls-opt-out-ai-tool-hollywood-outrage-1236644312/", "sources": null, "summary": "Meta Muse Image的「@提及公开账号生成」功能因采用opt-out默认机制允许生成公开IG账号AI图像，遭SAG-AFTRA与CAA联合声讨，Meta承认「missed the mark」后将该功能下架。", "reason": null, "tags": ["Meta","AI肖像权","版权","opt-out"], "topics": ["Meta","图像生成","版权与政策"], "firsthand": true, "editor_pick": false},{"id": "2026-07-10-limssr", "url": "https://www.transwonder.top/insights/#2026-07-10-limssr", "title": "LIMSSR（ICML 2026 Spotlight）：不完全多模态学习的条件序列推理新范式", "date": "2026-07-10", "category": "论文", "source": "机器之心", "src": "https://www.jiqizhixin.com/articles/2026-07-10-8", "sources": null, "summary": "北京大学与福州大学联合提出LIMSSR，将不完全多模态学习问题重定义为条件序列推理任务，通过LLM语义推理实现上下文感知模态插补，ICML 2026 Spotlight，不依赖完整训练数据即超越现有方法。", "reason": null, "tags": ["多模态学习","ICML","北大","AI研究"], "topics": ["研究论文"], "firsthand": false, "editor_pick": false},{"id": "2026-07-10-hoi-edit-scpe", "url": "https://www.transwonder.top/insights/#2026-07-10-hoi-edit-scpe", "title": "北大HOI-Edit+SCPE：AI图像编辑进入「交互理解」时代（ICML 2026）", "date": "2026-07-10", "category": "论文", "source": "机器之心", "src": "https://www.163.com/dy/article/L1GGB6U30511AQHO.html", "sources": null, "summary": "北大王选所等团队发布HOI-Edit基准+SCPE自纠错框架（ICML 2026），将AI图像编辑从「改像素」推进到「交互理解」三层认知评测，L1交互分数提升22%，数据集与代码已开源。", "reason": null, "tags": ["图像编辑","AI研究","ICML","北大"], "topics": ["图像生成","研究论文"], "firsthand": false, "editor_pick": false},{"id": "2026-07-10-gege-music", "url": "https://www.transwonder.top/insights/#2026-07-10-gege-music", "title": "歌歌AI音乐大模型全面上线：单卡10秒生成华语歌曲，打通字节七大平台", "date": "2026-07-10", "category": "模型", "source": "量子位", "src": "https://www.qbitai.com/2026/07/447602.html", "sources": null, "summary": "歌歌AI音乐大模型7月10日全面上线，十亿级参数华语端到端扩散+歌声合成混合架构，单卡10秒生成完整华语歌曲，中文咬字还原率较通用模型提升67%，打通字节七大平台，同期启动非遗民乐数据采集计划。", "reason": null, "tags": ["AI音乐","歌曲生成","字节","音频创作"], "topics": ["字节跳动","大模型","音频与音乐"], "firsthand": false, "editor_pick": false},{"id": "2026-07-09-reve-2-1", "url": "https://www.transwonder.top/insights/#2026-07-09-reve-2-1", "title": "Reve 2.1 发布：图像质量与风格一致性双升级", "date": "2026-07-09", "category": "模型", "source": "Reve官方", "src": "https://blog.reve.com/posts/launching-reve-2.1/", "sources": null, "summary": "Reve发布2.1版本，重点提升图像质量与风格一致性，是2.0基础上的稳定性与精度迭代。", "reason": null, "tags": ["图像生成","Reve","模型更新","AI绘画"], "topics": ["图像生成"], "firsthand": true, "editor_pick": false},{"id": "2026-07-09-pai-2-hollywood", "url": "https://www.transwonder.top/insights/#2026-07-09-pai-2-hollywood", "title": "PAI 2.0进入好莱坞：独立导演实拍级AI短片破百万播放，60天内ARR达1100万美元", "date": "2026-07-09", "category": "案例", "source": "Variety/BusinessWire/新智元/51CTO", "src": "https://variety.com/2026/digital/tech/ai-utopai-studios-pai-2-generative-video-platform-1236764434/", "sources": null, "summary": "好莱坞独立导演PJ Ace使用PAI 2.0生成的3分钟AI短片播放量破百万，Utopai Studios平台在60天内年化营收达1100万美元，PAI Pro Skill矩阵开源并已接入Claude Code与Cursor。", "reason": null, "tags": ["AI视频","好莱坞","PAI","商业化"], "topics": ["视频生成","行业与融资"], "firsthand": false, "editor_pick": false},{"id": "2026-07-09-lingbot-world-2", "url": "https://www.transwonder.top/insights/#2026-07-09-lingbot-world-2", "title": "LingBot-World 2.0开源：全球首个小时级连续运行世界模型，720p/60fps实时生成", "date": "2026-07-09", "category": "论文", "source": "新智元/搜狐/知乎", "src": "https://www.163.com/dy/article/L1DA18770511ABV6.html", "sources": null, "summary": "Ant Group旗下Robbyant开源LingBot-World 2.0，实现全球首个小时级连续运行世界模型，支持720p/60fps实时生成，一张图片可延展为无限AI世界。", "reason": null, "tags": ["世界模型","开源","视频生成","Robbyant"], "topics": ["视频生成","3D 与世界模型","研究论文"], "firsthand": false, "editor_pick": false},{"id": "2026-07-09-gpt56-full-release", "url": "https://www.transwonder.top/insights/#2026-07-09-gpt56-full-release", "title": "GPT-5.6三款模型全量公开发布，结束国家安全审查有限预览期", "date": "2026-07-09", "category": "模型", "source": "OpenAI官方", "src": "https://openai.com/index/previewing-gpt-5-6-sol/", "sources": null, "summary": "GPT-5.6 Sol旗舰、Terra均衡、Luna轻量三款模型于7月9日全量公开，结束约两周国家安全审查，在ChatGPT与API中全球同步上线。", "reason": null, "tags": ["OpenAI","GPT-5","大模型发布","AI基础设施"], "topics": ["OpenAI","大模型"], "firsthand": true, "editor_pick": false},{"id": "2026-07-09-figma-make-gpt56", "url": "https://www.transwonder.top/insights/#2026-07-09-figma-make-gpt56", "title": "Figma Make正式接入GPT-5.6：从提示词、草图到可运行交互原型，支持自我修复", "date": "2026-07-09", "category": "产品", "source": "Figma官方Blog/优设情报", "src": "https://www.figma.com/blog/gpt-5-6-is-now-available-in-figma-make/", "sources": null, "summary": "Figma Make正式接入GPT-5.6，支持从提示词、草图或既有Figma文件生成可运行交互原型，首轮生成质量提升，遇错误可自我修复。", "reason": null, "tags": ["Figma","GPT-5.6","原型设计","设计工具"], "topics": ["OpenAI","Figma","设计工具"], "firsthand": true, "editor_pick": false},{"id": "2026-07-08-siggraph-games-summit", "url": "https://www.transwonder.top/insights/#2026-07-08-siggraph-games-summit", "title": "SIGGRAPH 2026首设Games Summit：全天聚焦AI在游戏开发中的应用", "date": "2026-07-08", "category": "赛展", "source": "SIGGRAPH官方/PRNewswire", "src": "https://s2026.siggraph.org/siggraph-2026-introduces-the-games-summit-a-new-one-day-home-for-game-developers/", "sources": null, "summary": "SIGGRAPH 2026首次设立独立的Games Summit，以全天专题形式聚焦AI在游戏开发中的动效、无障碍、VFX、音频与管线应用，于7月19日在洛杉矶举办。", "reason": null, "tags": ["SIGGRAPH","游戏","AI开发","动效"], "topics": ["设计工具","研究论文","艺术与展览"], "firsthand": true, "editor_pick": false},{"id": "2026-07-08-seedream-5-pro", "url": "https://www.transwonder.top/insights/#2026-07-08-seedream-5-pro", "title": "字节Seedream 5.0 Pro全球发布：信息可视化、图层分离、10+语言文字渲染", "date": "2026-07-08", "category": "模型", "source": "字节Seed官方", "src": "https://seed.bytedance.com/en/blog/beyond-generation-it-understands-design-introducing-seedream-5-0-pro", "sources": null, "summary": "字节Seedream 5.0 Pro正式发布，带来高密度信息可视化、交互式精准编辑（圈选/图层分离）、真实影像质感及10+语言原生文字生成四项核心升级。", "reason": null, "tags": ["图像生成","Seedream","字节","设计工具"], "topics": ["字节跳动","图像生成","设计工具"], "firsthand": true, "editor_pick": false},{"id": "2026-07-08-runway-dev", "url": "https://www.transwonder.top/insights/#2026-07-08-runway-dev", "title": "Runway Dev发布：单一API接入Gen-4.5/Aleph 2.0等，Adobe/Figma Weave已接入", "date": "2026-07-08", "category": "产品", "source": "Runway官方", "src": "https://runwayml.com/news/introducing-runway-dev", "sources": null, "summary": "Runway推出一站式AI媒体开发者平台Runway Dev，通过单一API调用接入Gen-4.5、Aleph 2.0、Act-Two、Seedance等多模型，Adobe、Figma Weave、Shutterstock等已接入，通过SOC 2 Type II合规认证。", "reason": null, "tags": ["Runway","API","开发者平台","设计工具集成"], "topics": ["Adobe","Figma","Runway"], "firsthand": true, "editor_pick": false},{"id": "2026-07-08-grok-4-5", "url": "https://www.transwonder.top/insights/#2026-07-08-grok-4-5", "title": "Grok 4.5发布：SpaceXAI+Cursor联训1.5T V9架构，$2/$6每百万token", "date": "2026-07-08", "category": "模型", "source": "xAI官方", "src": "https://x.ai/news/grok-4-5", "sources": null, "summary": "xAI发布Grok 4.5，SpaceXAI+Cursor联合训练的1.5T参数V9架构，定位「Opus级」编程智能体，$2/$6每百万token，较同级模型便宜60%以上，AI分析指数排名第四。", "reason": null, "tags": ["Grok","大语言模型","AI编程","多模态"], "topics": ["大模型"], "firsthand": true, "editor_pick": false},{"id": "2026-07-08-gpt-live-voice", "url": "https://www.transwonder.top/insights/#2026-07-08-gpt-live-voice", "title": "OpenAI发布GPT-Live-1/mini：全双工实时语音模型支持打断与实时翻译", "date": "2026-07-08", "category": "模型", "source": "OpenAI官方", "src": "https://openai.com/index/introducing-gpt-live/", "sources": null, "summary": "OpenAI发布GPT-Live-1与mini两款全双工实时语音模型，支持边听边说、主动打断及实时多语言翻译，已向ChatGPT用户和开发者开放。", "reason": null, "tags": ["OpenAI","实时语音","多语言","AI交互"], "topics": ["OpenAI","音频与音乐"], "firsthand": true, "editor_pick": false},{"id": "2026-07-08-google-photos-video-remix", "url": "https://www.transwonder.top/insights/#2026-07-08-google-photos-video-remix", "title": "Google Photos推出Video Remix：Gemini Omni驱动水彩/素描/油画风格AI视频变换", "date": "2026-07-08", "category": "产品", "source": "Google官方Blog", "src": "https://blog.google/products-and-platforms/products/photos/video-remix/", "sources": null, "summary": "Google Photos上线Video Remix功能，由Gemini Omni驱动，支持将用户视频转换为水彩、素描、油画艺术风格及电影级色彩重映，已在14国上线。", "reason": null, "tags": ["Google Photos","风格迁移","AI视频","Gemini"], "topics": ["Google","视频生成"], "firsthand": true, "editor_pick": false},{"id": "2026-07-08-cowart-canvas-plugin", "url": "https://www.transwonder.top/insights/#2026-07-08-cowart-canvas-plugin", "title": "Cowart开源：tldraw无限画布内「指哪改哪」Codex AI改图插件", "date": "2026-07-08", "category": "产品", "source": "量子位", "src": "https://github.com/zhongerxin/Cowart", "sources": null, "summary": "Cowart开源插件将Codex与tldraw无限画布结合，通过视觉标注加截图的方式解决AI改图中空间关系描述难的问题。", "reason": null, "tags": ["Codex","开源工具","无限画布","AI改图"], "topics": ["OpenAI"], "firsthand": true, "editor_pick": false},{"id": "2026-07-08-comfyui-v0271", "url": "https://www.transwonder.top/insights/#2026-07-08-comfyui-v0271", "title": "ComfyUI v0.27.1：接入Seedream 5.0 Pro与Seed Audio 1.0节点", "date": "2026-07-08", "category": "产品", "source": "ComfyUI官方GitHub", "src": "https://github.com/Comfy-Org/ComfyUI/releases", "sources": null, "summary": "ComfyUI v0.27.1通过Partner Nodes接入Seedream 5.0 Pro图像生成与Seed Audio 1.0多模态音频节点，同时停用Ideogram旧版并移除StabilityAI合作节点。", "reason": null, "tags": ["ComfyUI","工作流","Seedream","音频生成"], "topics": ["字节跳动","ComfyUI","音频与音乐"], "firsthand": true, "editor_pick": false},{"id": "2026-07-07-tryoncrafter-4d-tryon", "url": "https://www.transwonder.top/insights/#2026-07-07-tryoncrafter-4d-tryon", "title": "TryOnCrafter：首个自由视角视频虚拟试衣 DiT 框架（ECCV 2026）", "date": "2026-07-07", "category": "论文", "source": "机器之心 / arXiv", "src": "https://www.jiqizhixin.com/articles/2026-07-07-tryoncrafter", "sources": null, "summary": "TryOnCrafter提出首个支持自由视角的视频虚拟试衣DiT框架，引入4D试衣代理机制，入选ECCV 2026，机器之心与arXiv双源确认。", "reason": null, "tags": ["虚拟试衣","AI视频","计算机视觉","ECCV 2026"], "topics": ["视频生成","研究论文"], "firsthand": false, "editor_pick": true},{"id": "2026-07-07-tongji-chi-2026", "url": "https://www.transwonder.top/insights/#2026-07-07-tongji-chi-2026", "title": "同济设创11项成果亮相CHI 2026，2篇获最佳论文提名", "date": "2026-07-07", "category": "论文", "source": "同济DF", "src": "https://tjdi.tongji.edu.cn/NewsDetail.do?ID=5990&lang=", "sources": null, "summary": "同济大学设计创意学院11项研究成果入选CHI 2026，其中2篇获最佳论文提名，研究聚焦AI与交互设计、可及性及教育技术的交叉应用。", "reason": null, "tags": ["CHI 2026","交互设计","同济大学","AI教育"], "topics": ["研究论文"], "firsthand": true, "editor_pick": true},{"id": "2026-07-07-openjiuwen-skill-omni", "url": "https://www.transwonder.top/insights/#2026-07-07-openjiuwen-skill-omni", "title": "openJiuwen Skill-Omni：AI Agent经验从文字升级为视觉多模态", "date": "2026-07-07", "category": "产品", "source": "量子位", "src": "https://www.qbitai.com/2026/07/445229.html", "sources": null, "summary": "openJiuwen Skill-Omni将AI Agent经验存储从纯文字升级为视觉+文字形式，支持从网页和视频自动提取截图生成可复用经验库。", "reason": null, "tags": ["AI Agent","多模态","Skill","视觉理解"], "topics": ["智能体与工作流"], "firsthand": false, "editor_pick": false},{"id": "2026-07-07-meta-muse-image", "url": "https://www.transwonder.top/insights/#2026-07-07-meta-muse-image", "title": "Meta Muse Image 正式发布：Superintelligence Labs 首个自研图像模型，Agent 式架构上线 Instagram/WhatsApp", "date": "2026-07-07", "category": "模型", "source": "Meta 官方 / TechCrunch / CNBC / Bloomberg / Axios", "src": "https://about.fb.com/news/2026/07/introducing-muse-image-meta-ai/", "sources": null, "summary": "Meta Superintelligence Labs 发布首个自研图像生成模型 Muse Image，采用 Agent 式架构（调用搜索与代码工具自我修正），已上线 Meta AI 应用、Instagram Stories 及 WhatsApp，并将接入广告系统 Advantage+，多指标超越 Google Nano Banana 2", "reason": null, "tags": ["Meta","图像生成","Agent","社交媒体"], "topics": ["Meta","图像生成","智能体与工作流"], "firsthand": true, "editor_pick": true},{"id": "2026-07-07-gemini-35-pro-leak", "url": "https://www.transwonder.top/insights/#2026-07-07-gemini-35-pro-leak", "title": "Gemini 3.5 Pro 传闻预告：完全重新训练，主打前端代码像素级精准与 200 万上下文（未经官方确认）", "date": "2026-07-07", "category": "模型", "source": "新智元 / 36氪 / 凤凰科技 / 多源", "src": "https://aiera.com.cn/2026/07/07/other/admin/102702/", "sources": null, "summary": "多源传言 Google DeepMind 将于 7 月 17 日发布 Gemini 3.5 Pro，称完全重新预训练，主打前端代码像素级精准、200 万上下文窗口及 SVG 与 3D 建模能力，尚未经官方确认", "reason": null, "tags": ["Gemini","Google","模型发布","前端设计"], "topics": ["Google","设计工具"], "firsthand": false, "editor_pick": false},{"id": "2026-07-07-flash-world-model", "url": "https://www.transwonder.top/insights/#2026-07-07-flash-world-model", "title": "魔芯Flash World Model：视频世界模型训练成本降70%，50FPS实时交互", "date": "2026-07-07", "category": "模型", "source": "量子位", "src": "https://www.qbitai.com/2026/07/446411.html", "sources": null, "summary": "魔芯发布Flash World Model，视频世界模型训练成本较同类降低70%，实现50FPS实时交互世界生成。", "reason": null, "tags": ["世界模型","视频生成","实时交互","开源"], "topics": ["视频生成","3D 与世界模型"], "firsthand": false, "editor_pick": false},{"id": "2026-07-07-fable5-63-3d-worlds", "url": "https://www.transwonder.top/insights/#2026-07-07-fable5-63-3d-worlds", "title": "1600行代码，Claude Fable 5一次生成63个完整3D世界", "date": "2026-07-07", "category": "案例", "source": "新智元 / Arena.ai", "src": "https://aiera.com.cn/2026/07/07/fable-5-3d-world", "sources": null, "summary": "研究人员通过1600行代码提示Claude Fable 5，一次性生成63个完整3D世界场景（含水下曼哈顿等），此案例由新智元与Arena.ai报道，引发Karpathy等人公开评论。", "reason": null, "tags": ["Claude","3D生成","代码生成","AI能力"], "topics": ["Anthropic","3D 与世界模型"], "firsthand": false, "editor_pick": false},{"id": "2026-07-07-claude-cowork-mobile", "url": "https://www.transwonder.top/insights/#2026-07-07-claude-cowork-mobile", "title": "Claude Cowork 扩展至 web/iOS/Android：后台持续运行，90% 使用集中于内容创作", "date": "2026-07-07", "category": "产品", "source": "Anthropic 官方 / TechCrunch / 9to5Mac", "src": "https://claude.com/blog/cowork-web-mobile", "sources": null, "summary": "Anthropic宣布Claude Cowork扩展至web、iOS与Android端，支持后台持续运行与跨设备协作，官方数据显示90%的使用场景集中于内容创作等知识型工作而非编程。", "reason": null, "tags": ["Claude","AI工具","跨设备","知识工作"], "topics": ["Anthropic"], "firsthand": true, "editor_pick": false},{"id": "2026-07-07-ai-wrong-first-step", "url": "https://www.transwonder.top/insights/#2026-07-07-ai-wrong-first-step", "title": "用AI第一步就错了：「问对问题」比「给对指令」更重要", "date": "2026-07-07", "category": "教程", "source": "爱范儿", "src": "https://www.ifanr.com/1671120", "sources": null, "summary": "爱范儿观点文章指出，大多数人在使用AI时最常犯的错误是一开始就“给指令”而非“提问题”，降低预期与承认无知是用好AI工具的前置条件。", "reason": null, "tags": ["AI使用方法","提示工程","AI工具","思维框架"], "topics": ["智能体与工作流"], "firsthand": false, "editor_pick": false},{"id": "2026-07-06-tongji-di-newline-2026", "url": "https://www.transwonder.top/insights/#2026-07-06-tongji-di-newline-2026", "title": "同济D&I 2026毕业展「换行 NewLine」：近260件作品，AI辅助设计方向成常态", "date": "2026-07-06", "category": "赛展", "source": "同济DF", "src": "https://tjdi.tongji.edu.cn/NewsDetail.do?ID=5989&lang=", "sources": null, "summary": "同济大学设计创意学院2026届毕业展「换行 NewLine」于7月6日开幕，近260件作品涵盖AI辅助设计方向，展现设计专业毕业生对AI工具融入创作实践的探索。", "reason": null, "tags": ["毕业展","同济大学","AI设计","设计教育"], "topics": ["设计教育","艺术与展览"], "firsthand": true, "editor_pick": false},{"id": "2026-07-06-siggraph-2026-keynotes", "url": "https://www.transwonder.top/insights/#2026-07-06-siggraph-2026-keynotes", "title": "SIGGRAPH 2026 主旨演讲公布：NVIDIA 实时神经渲染 + TripoAI 3D 生成双线并行", "date": "2026-07-06", "category": "赛展", "source": "SIGGRAPH 官方 / PRNewswire", "src": "https://s2026.siggraph.org/program/keynote-presentations/", "sources": null, "summary": "SIGGRAPH 2026 官方公布主旨演讲阵容：NVIDIA 主讲实时神经渲染技术方向，TripoAI 代表 3D 生成新范式参与演讲，大会 7 月 19–23 日于洛杉矶举行", "reason": null, "tags": ["SIGGRAPH","NVIDIA","3D生成"], "topics": ["3D 与世界模型","研究论文","艺术与展览"], "firsthand": true, "editor_pick": false},{"id": "2026-07-06-seedance-hollywood-filmmakers", "url": "https://www.transwonder.top/insights/#2026-07-06-seedance-hollywood-filmmakers", "title": "字节Seedance正在占领好莱坞：独立电影人以AI视频$9/分钟成本替代专业摄制", "date": "2026-07-06", "category": "案例", "source": "量子位 / TechTimes / AI Weekly / Variety", "src": "https://www.qbitai.com/2026/07/443665.html", "sources": null, "summary": "独立电影人群体正悄然转向字节Seedance 2.5，AI视频生成成本约$9/分钟较传统摄制$24/分钟大幅压低，量子位、Variety等多源报道证实这一趋势正在专业创作圈扩散。", "reason": null, "tags": ["AI视频","Seedance","好莱坞","独立电影"], "topics": ["字节跳动","视频生成"], "firsthand": false, "editor_pick": true},{"id": "2026-07-06-seedance-2-5-public-beta", "url": "https://www.transwonder.top/insights/#2026-07-06-seedance-2-5-public-beta", "title": "即梦 Seedance 2.5 公布：多镜头叙事一致性 + 运镜控制，公测尚未正式开放", "date": "2026-07-06", "category": "模型", "source": "IT之家 / 量子位 / 36氪 / 雷锋网 / 新浪科技", "src": "https://www.ithome.com/0/972/458.htm", "sources": null, "summary": "字节跳动即梦旗下 Seedance 2.5 视频生成模型公布：主打多镜头跨场景人物一致性与专业运镜控制；公测尚未正式面向全体用户开放，具体上线时间以官方为准", "reason": null, "tags": ["Seedance 2.5","即梦","AI视频生成"], "topics": ["字节跳动","视频生成"], "firsthand": false, "editor_pick": true},{"id": "2026-07-06-icrdag-region-drag", "url": "https://www.transwonder.top/insights/#2026-07-06-icrdag-region-drag", "title": "ICRDrag：上海交大提出首个上下文区域拖拽图像编辑模型，精准变形代替点选", "date": "2026-07-06", "category": "论文", "source": "机器之心 / 新浪科技 / arXiv:2606.25907", "src": "https://arxiv.org/pdf/2606.25907", "sources": null, "summary": "上海交大 ICRDrag 提出首个基于 DiT 上下文学习的区域拖拽图像编辑模型，用掩码替代点选实现源区域→任意目标区域精准变形，代码与模型已开源（ECCV 2026）", "reason": null, "tags": ["图像编辑","扩散模型","arXiv","上海交大"], "topics": ["图像生成","研究论文"], "firsthand": true, "editor_pick": true},{"id": "2026-07-06-gemini-omni-video-arena", "url": "https://www.transwonder.top/insights/#2026-07-06-gemini-omni-video-arena", "title": "Gemini Omni Flash 以 1404 Elo 登顶 Video Arena 盲测榜，Google 占前十中 5 席", "date": "2026-07-06", "category": "模型", "source": "优设情报 / IT之家", "src": "https://www.ithome.com/0/972/253.htm", "sources": null, "summary": "Gemini Omni Flash 在 Video Arena 盲测中以 1404 Elo 分领先全部 AI 视频模型，字节 Seedance 2.0 Mini 以 1303 Elo 居次，前十名 Google 占 5 席", "reason": null, "tags": ["Gemini","AI视频","评测基准","Google"], "topics": ["Google","视频生成"], "firsthand": false, "editor_pick": true},{"id": "2026-07-06-cafa-hetero-resonance", "url": "https://www.transwonder.top/insights/#2026-07-06-cafa-hetero-resonance", "title": "央美「异质共振」展：AI石狮与陶瓷、数字光影共探传统工艺与AI共生", "date": "2026-07-06", "category": "赛展", "source": "央美EAST", "src": "https://so.html5.qq.com/page/real/search_news?docid=70000021_2156a4cd84216652", "sources": null, "summary": "央美「异质共振」展览于7月6日开幕，以陶瓷、数字光影及AI生成石狮等作品探讨传统手工艺与AI技术之间的共生与张力关系。", "reason": null, "tags": ["展览","央美","AI艺术","传统工艺"], "topics": ["艺术与展览","设计教育"], "firsthand": false, "editor_pick": true},{"id": "2026-07-06-anthropic-global-workspace", "url": "https://www.transwonder.top/insights/#2026-07-06-anthropic-global-workspace", "title": "Anthropic《语言模型中的全局工作空间》：Claude内部发现J-space意识结构", "date": "2026-07-06", "category": "论文", "source": "Anthropic官方", "src": "https://www.anthropic.com/research/global-workspace", "sources": null, "summary": "Anthropic研究发现Claude内部自发形成J-space全局工作空间结构，J-lens解释工具已开源，Claude能感知自身处于测试状态。", "reason": null, "tags": ["Anthropic","可解释性","意识研究","大模型"], "topics": ["Anthropic","大模型","研究论文"], "firsthand": true, "editor_pick": false},{"id": "2026-07-06-icml-2026-outstanding-paper", "url": "https://www.transwonder.top/insights/#2026-07-06-icml-2026-outstanding-paper", "title": "ICML 2026 杰出论文揭晓：清华扩散语言模型与 MIT/耶鲁采样算法同获大奖", "date": "2026-07-05", "category": "论文", "source": "ICML 官方 / 机器之心 / 量子位", "src": "https://blog.icml.cc/2026/07/05/announcing-the-icml-2026-awards/", "sources": null, "summary": "ICML 2026 两篇扩散模型论文同获杰出论文奖：清华黄高团队 JustGRPO（GSM8K 89.1%）与 MIT/耶鲁高精度采样算法；英伟达 Motive 获荣誉提名", "reason": null, "tags": ["ICML","扩散模型","机器学习","学术奖项"], "topics": ["研究论文"], "firsthand": true, "editor_pick": true},{"id": "2026-07-04-wan-streamer-v01", "url": "https://www.transwonder.top/insights/#2026-07-04-wan-streamer-v01", "title": "Wan-Streamer v0.1：首个端到端实时音视频流数字人框架，arXiv 提前公开", "date": "2026-07-04", "category": "模型", "source": "arXiv / 优设", "src": "https://arxiv.org/abs/2606.25041", "sources": null, "summary": "Wan-Streamer v0.1 提出端到端实时音视频流数字人生成框架，论文 arXiv:2606.25041 提前公开，覆盖音频驱动唇形同步与实时视频流输出全链路", "reason": null, "tags": ["数字人","实时流","AI视频"], "topics": ["阿里巴巴","视频生成","研究论文"], "firsthand": true, "editor_pick": false},{"id": "2026-07-04-siggraph-2026-design-directions", "url": "https://www.transwonder.top/insights/#2026-07-04-siggraph-2026-design-directions", "title": "SIGGRAPH 2026 技术论文 5 大设计方向：3D 感知/身体交互/AI 产品化/虚实融合/设计工具", "date": "2026-07-04", "category": "论文", "source": "搜狐 / 一只梨 / SIGGRAPH 2026 官方", "src": "https://www.sohu.com/a/1045842105_121124300", "sources": null, "summary": "从 SIGGRAPH 2026 技术论文提炼 5 大设计前沿方向：3D 空间感知、身体与情境交互、AI 产品化、虚实融合、设计工具，覆盖从感知到工具的完整链路", "reason": null, "tags": ["SIGGRAPH","设计研究","视觉计算","AI前沿"], "topics": ["3D 与世界模型","设计工具","研究论文"], "firsthand": true, "editor_pick": false},{"id": "2026-07-04-midjourney-copyright-discovery", "url": "https://www.transwonder.top/insights/#2026-07-04-midjourney-copyright-discovery", "title": "Midjourney 申请强制迪士尼等披露 AI 使用，AI 版权诉讼「发现权」博弈推上台面", "date": "2026-07-04", "category": "行业", "source": "TechCrunch / Engadget / Variety / ArtNet News", "src": "https://techcrunch.com/2026/07/04/midjourney-wants-hollywood-studios-to-reveal-the-details-of-their-ai-usage/", "sources": null, "summary": "Midjourney 向法院申请强制迪士尼/环球/华纳兄弟披露其自身 AI 使用情况，以『双重标准』为由反制版权诉讼，发现权博弈将影响 AI 艺术版权判例走向", "reason": null, "tags": ["Midjourney","AI版权","版权诉讼","好莱坞"], "topics": ["Midjourney","版权与政策"], "firsthand": false, "editor_pick": true},{"id": "2026-07-03-vidu-s1-realtime-video", "url": "https://www.transwonder.top/insights/#2026-07-03-vidu-s1-realtime-video", "title": "Vidu S1：一张照片 + 实时语音生成可持续互动的 AI 数字角色，AI 视频进入实时对话范式", "date": "2026-07-03", "category": "模型", "source": "PR Newswire / 雷锋网 / IT之家 / 量子位", "src": "https://www.prnewswire.com/news-releases/shengshu-technology-unveils-vidu-s1-bringing-real-time-interactive-generation-to-ai-video-302817626.html", "sources": null, "summary": "生数科技发布 Vidu S1：一张照片 + 实时语音即可生成可持续互动的 AI 数字角色，540P/25FPS 全双工交互，AR+扩散混合架构，模型侧延迟约 200ms", "reason": null, "tags": ["Vidu","实时视频生成","数字角色","生数科技"], "topics": ["视频生成","音频与音乐"], "firsthand": true, "editor_pick": true},{"id": "2026-07-03-kling-ai-funding", "url": "https://www.transwonder.top/insights/#2026-07-03-kling-ai-funding", "title": "可灵 AI 独立融资近 30 亿美元：腾讯/阿里/百度联合领投，全球 AI 视频最大单笔融资纪录", "date": "2026-07-03", "category": "行业", "source": "新浪财经 / TechNode / CNBC / 量子位", "src": "https://technode.com/2026/07/03/kuaishous-kling-ai-raises-nearly-3-billion-in-funding/", "sources": null, "summary": "快手旗下可灵 AI 完成独立融资近 30 亿美元（约 196 亿元），腾讯/阿里/百度历史性联合领投，刷新全球 AI 视频模型最大单笔融资纪录", "reason": null, "tags": ["可灵AI","AI视频","融资","快手"], "topics": ["阿里巴巴","腾讯","快手"], "firsthand": false, "editor_pick": true},{"id": "2026-07-03-generation-consistency-survey", "url": "https://www.transwonder.top/insights/#2026-07-03-generation-consistency-survey", "title": "中科大等联合综述 500+ 篇文献：扩散模型的「生成一致性危机」三类框架与开源评测", "date": "2026-07-03", "category": "论文", "source": "新智元", "src": "https://aiera.com.cn/2026/07/03/other/admin/102188/", "sources": null, "summary": "中科大+清华+剑桥等机构联合综述 500+ 篇论文，系统分析扩散模型『生成一致性危机』，提出外部/内部/规范三类一致性框架", "reason": null, "tags": ["扩散模型","生成一致性","综述研究","开源评测"], "topics": ["研究论文"], "firsthand": false, "editor_pick": false},{"id": "2026-07-02-vast-ai-tripo-funding", "url": "https://www.transwonder.top/insights/#2026-07-02-vast-ai-tripo-funding", "title": "VAST AI Tripo 完成超 10 亿元 A3 轮融资，4399/贪玩/巨人网络等产业方参投", "date": "2026-07-02", "category": "行业", "source": "投资界 / 新浪财经 / SiliconAngle", "src": "https://news.pedaily.cn/202607/565779.shtml", "sources": null, "summary": "VAST AI 完成超 10 亿元 A3 轮融资，游戏产业方参投，Tripo 3D 大模型加速游戏/影视/工业设计商业化落地", "reason": null, "tags": ["VAST AI","Tripo","3D生成","融资"], "topics": ["3D 与世界模型","行业与融资"], "firsthand": false, "editor_pick": false},{"id": "2026-07-02-siggraph-caf-google-deepmind", "url": "https://www.transwonder.top/insights/#2026-07-02-siggraph-caf-google-deepmind", "title": "SIGGRAPH 2026 计算机动画节：Google DeepMind AI 辅助油画风动画短片《Dear Upstairs Neighbors》入选", "date": "2026-07-02", "category": "案例", "source": "SIGGRAPH 2026 官方 / PRNewswire", "src": "https://s2026.siggraph.org/siggraph-2026-computer-animation-festival-celebrates-global-storytelling-alongside-a-blockbuster-production-sessions-lineup/", "sources": null, "summary": "SIGGRAPH 2026 计算机动画节入选 Google DeepMind 团队 AI 辅助油画风动画短片，艺术家全程主导 + AI 协同创作模式获学院奖认证节展认可", "reason": null, "tags": ["SIGGRAPH","AI动画","Google DeepMind","计算机动画节"], "topics": ["Google","视频生成","研究论文"], "firsthand": true, "editor_pick": false},{"id": "2026-07-02-runway-agent-skills", "url": "https://www.transwonder.top/insights/#2026-07-02-runway-agent-skills", "title": "Runway 推出 Agent Skills：命令驱动 AI 代理接管完整广告活动与商业视频全流程", "date": "2026-07-02", "category": "产品", "source": "Runway 官方 changelog", "src": "https://runwayml.com/changelog", "sources": null, "summary": "Runway 推出 Agent Skills（7/2），支持用简单命令构建完整广告活动与商业视频，AI 代理接管从概念到成片的全流程", "reason": null, "tags": ["Runway","AI视频","Agent自动化","商业视频"], "topics": ["Runway","视频生成","智能体与工作流"], "firsthand": true, "editor_pick": false},{"id": "2026-07-01-google-veo-3-1", "url": "https://www.transwonder.top/insights/#2026-07-01-google-veo-3-1", "title": "Google Veo 3.1 重大更新：原生 9:16 竖版输出、1080p-4K 超分、跨场景角色一致性提升", "date": "2026-07-01", "category": "模型", "source": "Google 官方 / CineD / NoFilmSchool / Gadgetbridge", "src": "https://blog.google/innovation-and-ai/technology/ai/veo-3-1-ingredients-to-video/", "sources": null, "summary": "Google Veo 3.1 新增原生 9:16 竖版视频输出、1080p-4K 超分辨率及跨场景角色一致性能力，已上线 Gemini App、YouTube Shorts、Gemini API 等平台", "reason": null, "tags": ["Google Veo","AI视频生成","竖版视频","角色一致性"], "topics": ["Google","视频生成"], "firsthand": true, "editor_pick": false},{"id": "2026-07-01-genevolve-image-agent", "url": "https://www.transwonder.top/insights/#2026-07-01-genevolve-image-agent", "title": "GenEvolve：自进化图像生成 Agent 开源，工具编排轨迹替代单步提示词", "date": "2026-07-01", "category": "论文", "source": "机器之心 / 腾讯新闻 / arXiv:2605.21605 / GitHub（MeiGen-AI）", "src": "https://arxiv.org/abs/2605.21605", "sources": null, "summary": "GenEvolve 将图像生成从『一句话提示词』升级为 Agent 自主编排工具调用轨迹，港科广+美团+港科大+新加坡国立联合开源", "reason": null, "tags": ["GenEvolve","AI Agent","图像生成","开源"], "topics": ["图像生成","智能体与工作流","研究论文"], "firsthand": true, "editor_pick": true},{"id": "2026-07-01-claude-fable-5", "url": "https://www.transwonder.top/insights/#2026-07-01-claude-fable-5", "title": "Claude Fable 5 全球重新上线：19 天出口管制停摆后美国解除限制", "date": "2026-07-01", "category": "模型", "source": "Anthropic / VentureBeat / MacRumors / 9to5mac", "src": "https://www.anthropic.com/news/claude-fable-5-mythos-5", "sources": null, "summary": "Claude Fable 5 与 Mythos 5 于 2026-07-01 在 Claude.ai、Claude Code 及 API 全球恢复访问，此前因美国出口管制中断 19 天（6/12–7/1）；配套新安全分类器拦截率 >99%，7/7 前 Pro/Max/Team 用户可用 50% 额度。", "reason": null, "tags": ["Claude","Fable 5","AI大模型","政策合规"], "topics": ["Anthropic","大模型","版权与政策"], "firsthand": true, "editor_pick": false},{"id": "2026-06-30-siggraph-2026-best-paper", "url": "https://www.transwonder.top/insights/#2026-06-30-siggraph-2026-best-paper", "title": "SIGGRAPH 2026 技术论文最佳奖：CMU 平面映射算法 3D 矢量化速度提升数量级，7 月洛杉矶举行", "date": "2026-06-30", "category": "论文", "source": "SIGGRAPH 2026 官方 / PRNewswire / 新浪财经", "src": "https://www.prnewswire.com/news-releases/siggraph-2026-technical-papers-showcase-the-research-making-visual-computing-faster-more-reliable-and-accessible-302813822.html", "sources": null, "summary": "SIGGRAPH 2026 于 2026-06-30 公布技术论文最佳奖：CMU 平面映射算法获最佳论文（3D 矢量化速度提升数个数量级），慕尼黑工大时空流体仿真获荣誉提名；大会 7 月 19–23 日在洛杉矶举行（第 53 届）。", "reason": null, "tags": ["SIGGRAPH","技术论文","3D矢量化","视觉计算"], "topics": ["3D 与世界模型","设计工具","研究论文"], "firsthand": true, "editor_pick": false},{"id": "2026-06-30-google-nano-banana-2", "url": "https://www.transwonder.top/insights/#2026-06-30-google-nano-banana-2", "title": "Google Nano Banana 2 Lite + Gemini Omni Flash：4 秒出图、质量反超 Pro 档，静态图一键带运镜", "date": "2026-06-30", "category": "模型", "source": "Google DeepMind / IT之家 / 新浪财经 / ITBear / Winbuzzer", "src": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-omni-flash-nano-banana-2-lite/", "sources": null, "summary": "Google DeepMind 于 2026-06-30 同步发布 Nano Banana 2 Lite 图像模型（约 4 秒出图、每张 1K 分辨率图约 $0.034）与 Gemini Omni Flash 静态图转带运镜视频模型；两款均已接入 Adobe Firefly 与 Google Stitch 生态。", "reason": null, "tags": ["Google","Nano Banana 2","图像生成","Gemini"], "topics": ["Google","图像生成"], "firsthand": true, "editor_pick": true},{"id": "2026-06-30-comfyui-v0270", "url": "https://www.transwonder.top/insights/#2026-06-30-comfyui-v0270", "title": "ComfyUI v0.27.0：Int8 量化模型支持，VRAM 降低约 40%、推理速度提升约 40%", "date": "2026-06-30", "category": "产品", "source": "GitHub（Comfy-Org）/ CSDN / 网易科技", "src": "https://github.com/Comfy-Org/ComfyUI/releases", "sources": null, "summary": "ComfyUI v0.27.0 正式支持 Int8 量化模型，VRAM 需求降低约 40%、推理速度提升约 40%，适用于 Turing 及以上 GPU 架构", "reason": null, "tags": ["ComfyUI","量化加速","显存优化","AI生成工具"], "topics": ["ComfyUI"], "firsthand": true, "editor_pick": false},{"id": "2026-06-30-claude-sonnet-5", "url": "https://www.transwonder.top/insights/#2026-06-30-claude-sonnet-5", "title": "Anthropic 发布 Claude Sonnet 5：Agentic 场景最强 Sonnet，成为 Free/Pro 默认模型", "date": "2026-06-30", "category": "模型", "source": "Anthropic 官方 / TechCrunch / VentureBeat", "src": "https://www.anthropic.com/news/claude-sonnet-5", "sources": null, "summary": "Anthropic 于 6 月 30 日发布 Claude Sonnet 5，定位 Agentic 场景，性能接近旗舰 Opus 级，成为 Claude Free/Pro 用户默认模型；入门定价 $2/$10 每百万 Token（至 2026 年 8 月底），与 GPT-5.6 同日形成正面竞争格局。", "reason": null, "tags": ["Claude","Anthropic","大模型","Agentic"], "topics": ["Anthropic","大模型","智能体与工作流"], "firsthand": true, "editor_pick": true},{"id": "2026-06-30-apple-creator-studio", "url": "https://www.transwonder.top/insights/#2026-06-30-apple-creator-studio", "title": "Apple Creator Studio 重大更新：Final Cut Pro 端侧 AI 自动遮罩与字幕，Pixelmator Pro 跨应用打通", "date": "2026-06-30", "category": "产品", "source": "Apple Newsroom / PetaPixel / IT之家 / iClarified", "src": "https://www.apple.com/newsroom/2026/06/apple-creator-studio-gets-smarter-faster-and-more-connected/", "sources": null, "summary": "Apple 于 2026-06-30 免费更新 Creator Studio：Final Cut Pro 新增端侧 AI 的 Auto Mask（自动分离皮肤/头发/天空等）、Generate Subtitles（自动字幕转录）、Edit Detection（从成品还原原始片段）；Pixelmator Pro 与 FCP/Keynote/Pages/Numbers 跨应用实时同步编辑。", "reason": null, "tags": ["Apple","Final Cut Pro","AI视频剪辑","Pixelmator Pro"], "topics": ["视频生成"], "firsthand": true, "editor_pick": true},{"id": "2026-06-30-agnes-pavo", "url": "https://www.transwonder.top/insights/#2026-06-30-agnes-pavo", "title": "Agnes AI 发布 Pavo 创作平台：一句话拆剧本到成片，全模态自研模型 API 永久免费", "date": "2026-06-30", "category": "产品", "source": "量子位 / 新智元 / 甲子光年", "src": "https://www.qbitai.com/2026/06/439750.html", "sources": null, "summary": "Agnes AI 发布 PC 端创作平台 Pavo，用智能路由把「一句话」自动拆成剧本、角色设计、分镜直至成片；其自研文本/图像/视频全模态模型 API 自 6 月起永久免费、不限量，在烧钱的 AI 短剧赛道以「免费」切入。", "reason": null, "tags": ["AgnesAI","Pavo","AI短剧","AI视频","全模态"], "topics": ["视频生成"], "firsthand": false, "editor_pick": true},{"id": "2026-06-29-radio-tv-ai-microdrama-standard", "url": "https://www.transwonder.top/insights/#2026-06-29-radio-tv-ai-microdrama-standard", "title": "广电总局微短剧分类分层标准 7 月 1 日施行：AI 生成微短剧须每集加标识", "date": "2026-06-29", "category": "行业", "source": "央视网 / 人民日报 / 澎湃", "src": "https://news.cctv.cn/2026/06/26/ARTIENiZ1cMOJy4Jdwfctm3U260625.shtml", "sources": null, "summary": "广电总局发布微短剧分类分层管理标准，按投资额与题材分三类实行备案与发行许可，并要求 AI 生成、制作的微短剧在每集明显位置添加提示标识，自 2026 年 7 月 1 日起施行。", "reason": null, "tags": ["广电总局","微短剧","AIGC标识","政策"], "topics": ["视频生成","版权与政策","行业与融资"], "firsthand": false, "editor_pick": false},{"id": "2026-06-29-meituan-poster-aigc", "url": "https://www.transwonder.top/insights/#2026-06-29-meituan-poster-aigc", "title": "美团开源 AIGC 海报体系 PosterCraft/PosterOmni/PosterReward：「生成—编辑—评判」闭环", "date": "2026-06-29", "category": "模型", "source": "美团技术团队 / 机器之心 / MeiGen-AI", "src": "https://tech.meituan.com/2026/06/18/AIGC-poster.html", "sources": null, "summary": "美团开源覆盖生成、编辑、评判全链路的 AIGC 海报技术体系，含 PosterCraft、PosterOmni、PosterReward 三项工作（分别入选 ICLR/CVPR 2026），并已在真实业务落地。", "reason": null, "tags": ["美团","AIGC海报","开源","平面设计"], "topics": ["设计工具"], "firsthand": true, "editor_pick": false},{"id": "2026-06-29-google-design-md", "url": "https://www.transwonder.top/insights/#2026-06-29-google-design-md", "title": "Google Labs 开源 DESIGN.md：给 AI 编码智能体一份「看得懂」的设计系统说明", "date": "2026-06-29", "category": "产品", "source": "Google Labs / GitHub / 腾讯云开发者", "src": "https://github.com/google-labs-code/design.md", "sources": null, "summary": "Google Labs 开源 DESIGN.md 格式规范，用 YAML 设计 token + Markdown 说明的双层结构，让 Claude Code、Cursor 等 AI 编码智能体持久、结构化地理解并遵循一套设计系统。", "reason": null, "tags": ["Google","DESIGN.md","设计系统","AI编码"], "topics": ["Google","设计工具","智能体与工作流"], "firsthand": true, "editor_pick": false},{"id": "2026-06-29-deepseek-v4", "url": "https://www.transwonder.top/insights/#2026-06-29-deepseek-v4", "title": "DeepSeek V4 官宣 7 月中旬上线，首次引入峰谷动态定价机制", "date": "2026-06-29", "category": "模型", "source": "界面新闻 / 每日经济新闻", "src": "https://www.jiemian.com/article/14667473.html", "sources": null, "summary": "DeepSeek 宣布 V4 版本 7 月中旬上线，将引入峰谷定价机制，高峰时段 API 价格翻倍", "reason": null, "tags": ["DeepSeek","大模型","API定价"], "topics": ["大模型"], "firsthand": false, "editor_pick": false},{"id": "2026-06-28-midjourney-v82-preview", "url": "https://www.transwonder.top/insights/#2026-06-28-midjourney-v82-preview", "title": "Midjourney V8.2 预览版开放：强化非 V7 审美路线，Sref 一致性提升", "date": "2026-06-28", "category": "模型", "source": "Midjourney 官方 / Pexo", "src": "https://docs.midjourney.com/hc/en-us/articles/32199405667853-Version", "sources": null, "summary": "Midjourney V8.2 预览版可通过 --preview 参数提前体验，主要强化非 V7 审美风格路线、提升 Sref 风格板一致性；目前尚在开发阶段，表现不保证稳定，与 V8.1 Draft Mode 并行可用。", "reason": null, "tags": ["Midjourney","V8.2","文生图","Sref"], "topics": ["Midjourney","图像生成"], "firsthand": true, "editor_pick": false},{"id": "2026-06-28-loop-engineering", "url": "https://www.transwonder.top/insights/#2026-06-28-loop-engineering", "title": "「Prompt 已死」与 Loop Engineering 崛起：从写提示词到设计 AI 驱动循环", "date": "2026-06-28", "category": "观点", "source": "凤凰科技 / 新智元 / SlashDot / HackerNoon / X 多源", "src": "https://tech.ifeng.com/c/8uITKAT4QLO", "sources": null, "summary": "Peter Steinberger、Addy Osmani（Google Cloud）命名，黄仁勋、Karpathy、Claude Code 作者 Boris Cherny 等多人呼应——「Loop Engineering」主张从调单次提示词转向设计以 AI 为引擎的持续循环工作结构，被认为是 Prompt Engineering 的范式接班。", "reason": null, "tags": ["Loop Engineering","AI工作流","范式转移","Prompt工程"], "topics": ["智能体与工作流"], "firsthand": false, "editor_pick": false},{"id": "2026-06-26-trae-work-design", "url": "https://www.transwonder.top/insights/#2026-06-26-trae-work-design", "title": "字节 TRAE Work Design 模式：AI 识别设计系统组件库，自动生成 UI 稿并转代码", "date": "2026-06-26", "category": "产品", "source": "量子位", "src": "https://www.qbitai.com/2026/06/438750.html", "sources": null, "summary": "字节 TRAE 推出 Work Design 模式，可精准识别设计系统组件库，输入需求后自动生成 UI 设计稿并同步转代码，覆盖从设计到开发的完整工作流。", "reason": null, "tags": ["字节","TRAE","设计到代码","AI工作流"], "topics": ["字节跳动","设计工具","智能体与工作流"], "firsthand": false, "editor_pick": false},{"id": "2026-06-26-taste-skill-design-framework", "url": "https://www.transwonder.top/insights/#2026-06-26-taste-skill-design-framework", "title": "Taste Skill 开源「反平庸前端框架」：为 AI 生成界面注入设计品味", "date": "2026-06-26", "category": "产品", "source": "机器之心", "src": "https://www.jiqizhixin.com/", "sources": null, "summary": "Taste Skill 在 GitHub 获 51K 星，通过布局、排版、动效规则库为 AI 编程工具提供设计规范约束，旨在解决 Vibe Coding 同质化「AI 味」问题，提升 AI 辅助生成界面的视觉品质。", "reason": null, "tags": ["前端设计","AI编码","开源工具","界面品质"], "topics": ["设计工具","智能体与工作流"], "firsthand": false, "editor_pick": false},{"id": "2026-06-26-openai-gpt56-preview", "url": "https://www.transwonder.top/insights/#2026-06-26-openai-gpt56-preview", "title": "OpenAI GPT-5.6 有限预览发布：Sol 旗舰 / Terra 均衡 / Luna 轻量三档模型", "date": "2026-06-26", "category": "模型", "source": "OpenAI 官方 / VentureBeat / Axios / 量子位", "src": "https://openai.com/index/previewing-gpt-5-6-sol/", "sources": null, "summary": "OpenAI 推出 GPT-5.6 三款模型的有限预览：Sol（旗舰）、Terra（均衡）、Luna（轻量），约 20 个组织受邀，Sol 已登顶 LMSYS 排行，预计数周内全量开放。", "reason": null, "tags": ["OpenAI","GPT-5.6","多模态","大模型生态"], "topics": ["OpenAI","大模型"], "firsthand": true, "editor_pick": false},{"id": "2026-06-26-milan-design-week-ai-edu-forum", "url": "https://www.transwonder.top/insights/#2026-06-26-milan-design-week-ai-edu-forum", "title": "第十届米兰设计周中国高校展颁奖+「AI + 艺术设计教育」教学创新论坛举办", "date": "2026-06-26", "category": "赛展", "source": "米兰设计周官网", "src": "https://milan-aap.org.cn/article/4502", "sources": null, "summary": "第十届米兰设计周中国高校设计学科师生优秀作品展颁奖典礼与「AI + 艺术设计教育」教学创新论坛同期举办，聚焦 AI 赋能艺术设计教学改革，与高校教师受众直接相关。", "reason": null, "tags": ["米兰设计周","AI教育","高校","颁奖"], "topics": ["设计教育","艺术与展览"], "firsthand": true, "editor_pick": false},{"id": "2026-06-26-cn-ai-video-commercial-race", "url": "https://www.transwonder.top/insights/#2026-06-26-cn-ai-video-commercial-race", "title": "国内 AI 视频赛道从「技术炫技」进入「商业化验证」阶段", "date": "2026-06-26", "category": "观点", "source": "新浪财经", "src": "https://finance.sina.com.cn/jjxw/2026-06-26/doc-inieszry5247748.shtml", "sources": null, "summary": "字节 Seedance 2.5 与阿里 HappyHorse 1.1 同日亮相，可灵估值下调至 150 亿美元，Sand.ai 融资超亿美元——深度分析认为国内 AI 视频赛道正从「技术炫技」进入「商业化验证」阶段。", "reason": null, "tags": ["AI视频","商业化","即梦","HappyHorse"], "topics": ["字节跳动","视频生成"], "firsthand": false, "editor_pick": false},{"id": "2026-06-25-tencent-workreally-huolong", "url": "https://www.transwonder.top/insights/#2026-06-25-tencent-workreally-huolong", "title": "腾讯视频 WorkRally + 火龙漫剧：AI 全链路内容生产与「5–20 人创作小队」组织模式", "date": "2026-06-25", "category": "案例", "source": "DoNews / 新浪 / 腾讯新闻", "src": "https://finance.sina.com.cn/tech/roll/2026-06-25/doc-inierpqn0899151.shtml", "sources": null, "summary": "腾讯视频发布 WorkRally 全链路 AI 内容生产平台，并披露火龙漫剧平台运营数据：5 个月产出 7 万部，40 万成本作品获 2000 万分账；提出「5–20 人 AI 创作小队」为 AI 时代主流生产力单元。", "reason": null, "tags": ["AI影视","漫剧","内容生产","腾讯视频"], "topics": ["腾讯"], "firsthand": false, "editor_pick": false},{"id": "2026-06-25-siggraph-ai-relief", "url": "https://www.transwonder.top/insights/#2026-06-25-siggraph-ai-relief", "title": "SIGGRAPH 2026 投稿：AI 秒级生成高精度数字浮雕（港城大 / 斯坦福 / 康奈尔联合研究）", "date": "2026-06-25", "category": "论文", "source": "techwalker.com", "src": "https://www.techwalker.com/2026/0625/3191631.shtml", "sources": null, "summary": "香港城市大学、斯坦福、康奈尔科技学院、德州大学奥斯汀联合研究，实现 AI 在秒级时间内生成高精度数字浮雕，相关成果已投递 SIGGRAPH 2026；目前为单一来源报道。", "reason": null, "tags": ["SIGGRAPH","数字浮雕","3D生成","计算机图形"], "topics": ["3D 与世界模型","研究论文"], "firsthand": false, "editor_pick": false},{"id": "2026-06-25-qwen-image-rl-aesthetics", "url": "https://www.transwonder.top/insights/#2026-06-25-qwen-image-rl-aesthetics", "title": "Qwen-Image-2.0-RL：阿里用强化学习训练 AI「审美」，图像生成 Elo 提升 78 分", "date": "2026-06-25", "category": "论文", "source": "科技行者 / arXiv (2606.27608)", "src": "https://arxiv.org/abs/2606.27608", "sources": null, "summary": "阿里 Qwen 团队发布 Qwen-Image-2.0-RL（arXiv:2606.27608），通过 GRPO 强化学习与在线策略蒸馏设计图文对齐、美感、肖像三类奖励模型，文生图 Elo 从 1115 升至 1193（+78），图像编辑 Elo 从 1256 升至 1349（+93）。", "reason": null, "tags": ["Qwen","强化学习","AI审美","图像生成"], "topics": ["阿里巴巴","图像生成","研究论文"], "firsthand": true, "editor_pick": false},{"id": "2026-06-25-icml-aesthetic-alignment", "url": "https://www.transwonder.top/insights/#2026-06-25-icml-aesthetic-alignment", "title": "ICML 2026 Spotlight：AI 图像审美对齐机制正在削弱生成内容的艺术多样性", "date": "2026-06-25", "category": "论文", "source": "机器之心 / arXiv:2512.11883", "src": "https://arxiv.org/abs/2512.11883", "sources": null, "summary": "ICML 2026 Spotlight 论文指出，HPS 等 AI 图像审美评分系统的对齐机制正在削弱生成图像的艺术多样性，越符合评分标准的图像越缺乏表现力", "reason": null, "tags": ["ICML","审美对齐","图像生成研究","多样性"], "topics": ["图像生成","研究论文"], "firsthand": true, "editor_pick": false},{"id": "2026-06-25-hyper3d-rodin-gen25", "url": "https://www.transwonder.top/insights/#2026-06-25-hyper3d-rodin-gen25", "title": "影眸 Hyper3D Rodin Gen-2.5 发布：首个「先思考再生成」机制的原生 3D 大模型", "date": "2026-06-25", "category": "模型", "source": "新智元", "src": "https://hyper3d.ai/", "sources": null, "summary": "影眸科技发布 Hyper3D Rodin Gen-2.5 并完成数亿元融资，首创五档思考深度机制（约 4–80 秒），支持千万面级几何、12K 原生贴图、3D ControlNet、自然语言局部编辑与递归分件，Unity/Blender/Unreal 等 DCC 全面接入。", "reason": null, "tags": ["3D生成","Hyper3D","AI建模","影视后期"], "topics": ["大模型","3D 与世界模型"], "firsthand": true, "editor_pick": true},{"id": "2026-06-25-cn-kr-ai-art-exhibition", "url": "https://www.transwonder.top/insights/#2026-06-25-cn-kr-ai-art-exhibition", "title": "2026 中韩 AI 艺术作品国际邀请展在河北美术学院开幕：327 件作品，中韩战略合作协议同期签署", "date": "2026-06-25", "category": "赛展", "source": "河北美术学院官网", "src": "https://www.hbafa.com/info/1023/7068.htm", "sources": null, "summary": "2026 中韩 AI 艺术作品国际邀请展于 2026-06-25 在河北美术学院开幕，河北美院联合韩国空间设计学会、弘益大学 MR 媒体实验室主办，展出中韩等多国艺术家 327 件 AI 艺术作品，涵盖 AI 生成、沉浸式空间设计与数字媒体艺术。", "reason": null, "tags": ["中韩合作","AI艺术展","国际邀请展","高校美院"], "topics": ["设计教育","艺术与展览"], "firsthand": true, "editor_pick": false},{"id": "2026-06-25-adobe-acquires-topaz-labs", "url": "https://www.transwonder.top/insights/#2026-06-25-adobe-acquires-topaz-labs", "title": "Adobe 宣布收购 AI 图像与视频增强公司 Topaz Labs", "date": "2026-06-25", "category": "行业", "source": "Adobe 官方 / TechCrunch / Variety / PetaPixel", "src": "https://news.adobe.com/news/2026/06/adobe-to-acquire-topaz-labs", "sources": null, "summary": "Adobe 宣布收购曾获艾美奖的 AI 图像与视频增强公司 Topaz Labs，其旗舰模型 Astra（视频超分）与 Wonder（图像增强）将整合进 Firefly 及 CC 全系产品；Neurostream 技术可在消费级设备本地运行大型 AI 模型；交易预计 2026 年下半年完成，Topaz 现有工具同期保持独立可用。", "reason": null, "tags": ["Adobe","Topaz Labs","图像增强","视频超分"], "topics": ["Adobe","行业与融资"], "firsthand": true, "editor_pick": true},{"id": "2026-06-24-siggraph-2026-art-papers", "url": "https://www.transwonder.top/insights/#2026-06-24-siggraph-2026-art-papers", "title": "SIGGRAPH 2026 Art Papers 录用名单公布，主题「创作性翻译的复杂性」", "date": "2026-06-24", "category": "论文", "source": "SIGGRAPH 2026 官方", "src": "https://s2026.siggraph.org/program/art-papers/", "sources": null, "summary": "SIGGRAPH 2026 Art Papers 录用名单已公布，主题为「The Creative Complexities of Translation」，入选论文发表于 PACMCGIT 特刊，覆盖艺术×科学×技术交叉研究，有别于早先公布的 Art Gallery 入选；大会于 7 月 19–23 日在洛杉矶举行。", "reason": null, "tags": ["SIGGRAPH","艺术研究","学术会议",2026], "topics": ["研究论文"], "firsthand": true, "editor_pick": false},{"id": "2026-06-24-pqsg-video-physics-eval", "url": "https://www.transwonder.top/insights/#2026-06-24-pqsg-video-physics-eval", "title": "PQSG：首个系统评估 AI 视频物理一致性的基准，揭示 Sora 2 / Veo 3 的系统性短板", "date": "2026-06-24", "category": "论文", "source": "arXiv (2606.25306) / TechWalker", "src": "https://arxiv.org/abs/2606.25306", "sources": null, "summary": "北卡罗来纳大学、Johns Hopkins、AI2 与德克萨斯大学奥斯汀联合发布 PQSG 评估体系（arXiv:2606.25306），首次系统评估 AI 视频生成模型的物理一致性，覆盖刚体动力学、流体模拟、软体变形，揭示主流视频模型的系统性物理短板。", "reason": null, "tags": ["AI视频评估","物理一致性","基准测试","视频生成"], "topics": ["OpenAI","Google","视频生成"], "firsthand": true, "editor_pick": false},{"id": "2026-06-24-figma-config-2026", "url": "https://www.transwonder.top/insights/#2026-06-24-figma-config-2026", "title": "Figma Config 2026：Motion 原生动效时间轴、Code Layers 双向同步、Design Agent MCP Connectors", "date": "2026-06-24", "category": "产品", "source": "Figma Blog / CMSWire / 优设", "src": "https://www.figma.com/blog/introducing-figma-motion/", "sources": null, "summary": "Figma Config 2026 发布三项核心更新：画布内原生动效时间轴 Figma Motion（含 AI Agent 生成动效，当日 Open Beta）、设计与代码双向同步的 Code Layers，以及接入 Notion / Slack / GitHub 等 14+ 工具的 Design Agent MCP Connectors。", "reason": null, "tags": ["Figma","Motion","Code Layers","MCP","设计工具"], "topics": ["Figma","设计工具","智能体与工作流"], "firsthand": true, "editor_pick": true},{"id": "2026-06-24-fid-metric-reliability", "url": "https://www.transwonder.top/insights/#2026-06-24-fid-metric-reliability", "title": "Kyutai+UC Berkeley 预印本系统质疑 FID 指标：AI 图像生成评测可靠性存疑", "date": "2026-06-24", "category": "论文", "source": "Kyutai / UC Berkeley / 腾讯新闻", "src": "https://arxiv.org/abs/2606.20536", "sources": null, "summary": "Kyutai 与 UC Berkeley 联合预印本（arXiv:2606.20536）对 AI 图像生成常用评测指标 FID 的可靠性提出系统性质疑，指出不同评测环境下结论相互矛盾，「最强」排名可能依赖了一个有缺陷的指标。", "reason": null, "tags": ["FID","评测方法论","AI图像生成","学术研究"], "topics": ["图像生成","研究论文"], "firsthand": true, "editor_pick": false},{"id": "2026-06-24-fastwan-qad", "url": "https://www.transwonder.top/insights/#2026-06-24-fastwan-qad", "title": "FastWan-QAD 开源：单卡 RTX 5090 端到端 1.8 秒生成 5 秒 480P 视频", "date": "2026-06-24", "category": "模型", "source": "foxnan.com", "src": "https://www.foxnan.com/2026/06/3929.html", "sources": null, "summary": "Sky Computing Lab 开源 FastWan-QAD 量化感知蒸馏方案，单卡 RTX 5090 端到端 1.8 秒生成 5 秒 480P 视频，较标准版显著降低硬件门槛，模型与代码已完整开源。", "reason": null, "tags": ["视频生成","开源","量化蒸馏","硬件门槛"], "topics": ["阿里巴巴","视频生成"], "firsthand": true, "editor_pick": false},{"id": "2026-06-24-comfyui-v0260", "url": "https://www.transwonder.top/insights/#2026-06-24-comfyui-v0260", "title": "ComfyUI v0.26.0 发布：一次接入 Krea 2、Boogu-Image、Qwen3-VL 三大开源模型", "date": "2026-06-24", "category": "产品", "source": "ComfyUI 官方", "src": "https://docs.comfy.org/zh/changelog", "sources": null, "summary": "ComfyUI v0.26.0 在单次版本更新中同时接入 Krea 2（12B DiT 开源图像模型，含 Raw/Turbo 两版）、Boogu-Image（Qwen3-VL-8B 编码器）、Qwen3-VL（4B/8B），并新增 Load3DAdvanced 3D 节点与 LTX2 视频上下文采样支持，工作流模板升至 0.9.26。", "reason": null, "tags": ["ComfyUI","Krea2","开源模型","工作流"], "topics": ["阿里巴巴","ComfyUI","图像生成"], "firsthand": true, "editor_pick": true},{"id": "2026-06-24-adobe-firefly-partner-models", "url": "https://www.transwonder.top/insights/#2026-06-24-adobe-firefly-partner-models", "title": "Adobe Firefly Partner Models 全面开放：CC 全系内置 30+ 第三方行业模型", "date": "2026-06-24", "category": "产品", "source": "Adobe 官方 / Creative Bloq / AWN", "src": "https://www.adobe.com/products/firefly/partner-models.html", "sources": null, "summary": "Adobe 将 30+ 第三方行业模型内置进 Creative Cloud 全系（Firefly / Express / Photoshop / Illustrator 等），含 Kling 3.0、Veo 3.1、FLUX.2、Runway Gen-4.5、Luma、ElevenLabs 等，现有 CC 订阅即可调用、无需额外付费。", "reason": null, "tags": ["Adobe","Firefly","合作模型","多模型","AIGC"], "topics": ["Adobe"], "firsthand": true, "editor_pick": true},{"id": "2026-06-23-seeddream-5-pro", "url": "https://www.transwonder.top/insights/#2026-06-23-seeddream-5-pro", "title": "字节 Seeddream 5.0 Pro：交互式精准编辑与多图层分离输出，对接 PSD 工作流", "date": "2026-06-23", "category": "模型", "source": "ChooseAI", "src": "https://www.chooseai.net/news/4583/", "sources": null, "summary": "字节跳动于 FORCE 大会（6 月 23 日）发布 Seeddream 5.0 Pro，新增箭头/圈选直接标注的交互式精准编辑、多图层分离输出（可直接对接 PSD 工作流）及 10 余种语言原生文字生成，与 Seedance 2.5 构成图像→视频完整工业化链路。", "reason": null, "tags": ["字节","图像生成","图层分离","多语种"], "topics": ["字节跳动","图像生成","智能体与工作流"], "firsthand": true, "editor_pick": false},{"id": "2026-06-23-seedance-2-5", "url": "https://www.transwonder.top/insights/#2026-06-23-seedance-2-5", "title": "豆包 Seedance 2.5 发布：30 秒原生直出、50 素材联合生成、局部一致性编辑", "date": "2026-06-23", "category": "模型", "source": "火山引擎 / 机器之心 / 量子位 / 36氪 / AIbase", "src": "https://news.aibase.com/zh/news/29087", "sources": null, "summary": "火山引擎在 FORCE 大会发布 Seedance 2.5，首次实现 30 秒原生单段直出和 50 个全模态素材联合生成，同步发布图像模型 Seedream 5.0 Pro 和音频模型 Seed-Audio 1.0。", "reason": null, "tags": ["加更","豆包","Seedance","视频生成","ByteDance"], "topics": ["字节跳动","视频生成"], "firsthand": true, "editor_pick": true},{"id": "2026-06-23-seed-audio-1-0", "url": "https://www.transwonder.top/insights/#2026-06-23-seed-audio-1-0", "title": "字节 Seed-Audio 1.0：零样本多模态参考生成人声、音效与 BGM 全要素", "date": "2026-06-23", "category": "模型", "source": "新华网", "src": "https://www.xinhuanet.com/tech/20260623/acd6f2f27fc34459a7d1684c03278431/c.html", "sources": null, "summary": "字节跳动于 FORCE 大会（6 月 23 日）发布 Seed-Audio 1.0，支持零样本多模态参考生成人声、音效与 BGM 全要素，可输出多角色对白与拟音，单次时长达 2 分钟并可延伸至数十分钟。", "reason": null, "tags": ["字节","音频生成","多模态","影视后期"], "topics": ["字节跳动","音频与音乐"], "firsthand": false, "editor_pick": false},{"id": "2026-06-23-krea-2-open-weights", "url": "https://www.transwonder.top/insights/#2026-06-23-krea-2-open-weights", "title": "Krea 2 技术报告：开源 12B DiT 图像模型，明确不使用 AI 生成图像预训练", "date": "2026-06-23", "category": "模型", "source": "Krea AI 官方", "src": "https://www.krea.ai/blog/krea-2-technical-report", "sources": null, "summary": "Krea 发布 Krea 2 技术报告，推出 Raw（LoRA 微调基座）与 Turbo（8 步推理，约 2 秒生成）两个开放权重变体，明确训练数据中不使用 AI 生成图像，采用商业友好许可，ComfyUI v0.26.0 于发布当日接入。", "reason": null, "tags": ["开源模型","Krea","图像生成","LoRA"], "topics": ["图像生成"], "firsthand": true, "editor_pick": false},{"id": "2026-06-23-doubao-2-1-pro", "url": "https://www.transwonder.top/insights/#2026-06-23-doubao-2-1-pro", "title": "字节跳动豆包 2.1 Pro 发布：Coding/Agent 能力提升，多模态视觉理解评测领先", "date": "2026-06-23", "category": "模型", "source": "新华网 / 量子位 / 多源", "src": "https://www.xinhuanet.com/tech/20260623/acd6f2f27fc34459a7d1684c03278431/c.html", "sources": null, "summary": "字节跳动发布豆包 2.1 Pro，Coding 与 Agent 能力较前代显著提升、对标旗舰级模型，并在多模态视觉理解评测集上保持领先，可作为设计与创作场景的多模态工作底座。", "reason": null, "tags": ["字节跳动","豆包","多模态","视觉理解","大模型"], "topics": ["字节跳动","大模型","智能体与工作流"], "firsthand": false, "editor_pick": false},{"id": "2026-06-23-cocacola-ai-tvc", "url": "https://www.transwonder.top/insights/#2026-06-23-cocacola-ai-tvc", "title": "可口可乐中国世界杯营销全面 AI 化：5 人 5 城 TVC 全由 Prompt 生成，范志毅数字人 2 秒实时互动", "date": "2026-06-23", "category": "案例", "source": "量子位 / 优设", "src": "https://www.qbitai.com/2026/06/437381.html", "sources": null, "summary": "可口可乐中国 2026 世界杯限定版 TVC 5 人 5 城场景全由 Prompt 生成（操盘方百度一镜），范志毅数字人实现 2 秒端到端实时互动，是顶级国际品牌世界杯广告预算被 AI 替代的直接落地案例。", "reason": null, "tags": ["加更","可口可乐","AI广告","数字人","品牌营销"], "topics": ["视频生成"], "firsthand": false, "editor_pick": true},{"id": "2026-06-23-adobe-ps-flux-kontext", "url": "https://www.transwonder.top/insights/#2026-06-23-adobe-ps-flux-kontext", "title": "Adobe Photoshop 生成填充引入 FLUX Kontext 与 Google Gemini 作为伴侣模型", "date": "2026-06-23", "category": "产品", "source": "RedSharkNews / PRONEWS / CreativeBloq", "src": "https://www.redsharknews.com/adobe-creative-cloud-june-2026-update", "sources": null, "summary": "Adobe CC 6 月 23 日周次更新：Photoshop 生成填充新增 FLUX Kontext 和 Google Gemini 作为可选伴侣模型，Generative Upscale 新增 Topaz Gigapixel 选项，After Effects AI 转描工具同步重构。", "reason": null, "tags": ["Adobe","Photoshop","FLUX","生成填充"], "topics": ["Google","Adobe","Black Forest Labs"], "firsthand": false, "editor_pick": true},{"id": "2026-06-22-waic-2026-sail-award", "url": "https://www.transwonder.top/insights/#2026-06-22-waic-2026-sail-award", "title": "WAIC 2026 SAIL 奖 TOP30 及青年优秀论文 TOP20 发布", "date": "2026-06-22", "category": "论文", "source": "机器之心 / 新浪 / WAIC 官网", "src": "https://lewz.cn/pjss2026sjrgzndhsail.html", "sources": null, "summary": "WAIC 2026（7 月 17–20 日，上海）发布 SAIL 奖 TOP30 及青年优秀论文 TOP20，计算机视觉、具身智能、AI 创作方向均有入围，首次自举办顶级学术分会 WAIC Academic。", "reason": null, "tags": ["WAIC","SAIL奖","AI研究","学术会议"], "topics": ["研究论文"], "firsthand": true, "editor_pick": false},{"id": "2026-06-22-utopai-journey-to-west", "url": "https://www.transwonder.top/insights/#2026-06-22-utopai-journey-to-west", "title": "全球首部工业级全 AI 中国经典 IP 动画剧集：华策×Utopai《西游记：失落的五百年》", "date": "2026-06-22", "category": "案例", "source": "Variety / Deadline / Business Wire / IT之家", "src": "https://variety.com/2026/tv/news/fully-ai-generated-journey-to-the-west-utopai-studios-huace-1236787171/", "sources": null, "summary": "华策影视与 Utopai Studios 联合宣布《西游记：失落的五百年》，采用 Utopai 第二代 PAI 平台全流程制作，被定位为全球首部工业级全 AI 中国经典 IP 动画剧集，第一季定向全球流媒体与广播发行。", "reason": null, "tags": ["AI动画","全AI生成","西游记","影视工业化"], "topics": ["视频生成"], "firsthand": false, "editor_pick": true},{"id": "2026-06-22-tongji-art-ai-dual-degree", "url": "https://www.transwonder.top/insights/#2026-06-22-tongji-art-ai-dual-degree", "title": "同济大学新增艺术设计学—人工智能双学位复合型人才培养项目", "date": "2026-06-22", "category": "行业", "source": "康石石", "src": "https://www.kangshishi.com/", "sources": null, "summary": "同济大学正式新增艺术设计学与人工智能双学位复合型人才培养项目，是国内顶尖工科高校在 AI 与艺术设计交叉培养方向的明确战略信号。", "reason": null, "tags": ["同济大学","双学位","AI艺术设计","教育政策"], "topics": ["版权与政策"], "firsthand": true, "editor_pick": false},{"id": "2026-06-22-happyhorse-1-1", "url": "https://www.transwonder.top/insights/#2026-06-22-happyhorse-1-1", "title": "阿里 HappyHorse 1.1 升级五维度，同步开启 Horsepower AI 影像大赛", "date": "2026-06-22", "category": "模型", "source": "量子位", "src": "https://www.qbitai.com/2026/06/437317.html", "sources": null, "summary": "阿里 HappyHorse 1.1 版本升级动态表现力、9 图角色一致性、复杂场景指令、视觉质感和音频口播五个维度，同步宣布启动 Horsepower AI 影像大赛（张纪中评委）。", "reason": null, "tags": ["阿里","HappyHorse","视频生成","AI影像"], "topics": ["阿里巴巴","视频生成","图像生成"], "firsthand": false, "editor_pick": false},{"id": "2026-06-22-deepmind-a24-ai-filmmaking", "url": "https://www.transwonder.top/insights/#2026-06-22-deepmind-a24-ai-filmmaking", "title": "Google DeepMind 与 A24 建立 AI 研究合作并投资约 7500 万美元，Google 首次持股电影公司", "date": "2026-06-22", "category": "案例", "source": "Google DeepMind Blog / Variety / IndieWire / 多源", "src": "https://blog.google/innovation-and-ai/models-and-research/google-deepmind/deepmind-a24-research-partnership/", "sources": null, "summary": "Google DeepMind 宣布与独立电影公司 A24 建立 AI 研究合作并投资约 7500 万美元，DeepMind 研究员将与 A24 主创并肩开发 AI 分镜与工作流工具，这是 Google 首次持股电影公司。", "reason": null, "tags": ["加更","Google DeepMind","A24","AI分镜","电影制作"], "topics": ["Google"], "firsthand": true, "editor_pick": true},{"id": "2026-06-22-cafa-grad-show-2026", "url": "https://www.transwonder.top/insights/#2026-06-22-cafa-grad-show-2026", "title": "央美 2026 本科毕业展收官：8 院系作品涵盖数码媒体与 AI 辅助创作", "date": "2026-06-22", "category": "案例", "source": "央美美术馆", "src": "https://www.cafamuseum.org/exhibit/newsdetail/4107", "sources": null, "summary": "中央美术学院 2026 年本科毕业展于 6 月 22 日收官，8 个院系作品涵盖数码媒体、空间装置等方向，AI 辅助创作引发广泛关注。", "reason": null, "tags": ["央美","毕业展","AI艺术","数码媒体"], "topics": ["艺术与展览","设计教育"], "firsthand": true, "editor_pick": false},{"id": "2026-06-22-ai-design-tools-collection", "url": "https://www.transwonder.top/insights/#2026-06-22-ai-design-tools-collection", "title": "2026 AI 设计工具全收录：按使用场景分类整理", "date": "2026-06-22", "category": "教程", "source": "AI工坊pro", "src": "https://mp.weixin.qq.com/", "sources": null, "summary": "AI工坊pro 发布 2026 年 AI 设计工具全收录，按图像、视频、3D、UI 等使用场景分类整理，覆盖当前主流工具生态。", "reason": null, "tags": ["AI设计工具","工具合集","图像生成","工作流"], "topics": ["图像生成","设计工具","智能体与工作流"], "firsthand": false, "editor_pick": false},{"id": "2026-06-21-siggraph-2026-ai-preview", "url": "https://www.transwonder.top/insights/#2026-06-21-siggraph-2026-ai-preview", "title": "SIGGRAPH 2026 前沿预热：多篇中文解读密集放出，AI「全面入侵」大会", "date": "2026-06-21", "category": "论文", "source": "内核观察 / 哲思哲言 / 深圳大学", "src": "https://s2026.siggraph.org/", "sources": null, "summary": "SIGGRAPH 2026 开幕前夕，国内多个技术媒体和高校研究团队集中放出中文深度解读，覆盖从芯片到像素的 AI 全链路渗透格局。", "reason": null, "tags": ["SIGGRAPH","计算机图形学","AI研究","学术会议"], "topics": ["研究论文"], "firsthand": true, "editor_pick": false},{"id": "2026-06-21-palmier-pro-editor", "url": "https://www.transwonder.top/insights/#2026-06-21-palmier-pro-editor", "title": "Palmier Pro 开源 macOS 视频编辑器：内置 Seedance / Kling 等，支持 MCP 与 AI Agent 协作", "date": "2026-06-21", "category": "产品", "source": "腾讯云 / 机器之心 / oschina / CSDN", "src": "https://cloud.tencent.com/developer/article/2694546", "sources": null, "summary": "YC S24 团队开源 Palmier Pro macOS 视频编辑器，时间线内集成 Seedance、Kling、Nano Banana，并通过 MCP 服务器与 Claude、Codex、Cursor 等 AI Agent 无缝协作。", "reason": null, "tags": ["Palmier","视频编辑","MCP","AI工作流"], "topics": ["字节跳动","快手","智能体与工作流"], "firsthand": true, "editor_pick": false},{"id": "2026-06-21-gduf-ai-art-minor", "url": "https://www.transwonder.top/insights/#2026-06-21-gduf-ai-art-minor", "title": "广东财经大学发布 2026 年 AI 艺术设计微专业招生方案", "date": "2026-06-21", "category": "行业", "source": "广东财经大学", "src": "https://www.gduf.edu.cn/", "sources": null, "summary": "广东财经大学正式发布 2026 年 AI 艺术设计微专业招生方案，是国内高校在 AI 与艺术设计交叉方向设置微专业的又一落地案例。", "reason": null, "tags": ["AI艺术设计","微专业","高校招生","艺术设计教育"], "topics": ["设计教育"], "firsthand": true, "editor_pick": false},{"id": "2026-06-20-grok-for-office", "url": "https://www.transwonder.top/insights/#2026-06-20-grok-for-office", "title": "xAI 推出 Grok for Office 插件，进入 Word / PowerPoint / Excel", "date": "2026-06-20", "category": "产品", "source": "xAI News", "src": "https://x.ai/news", "sources": null, "summary": "xAI 推出免费 Grok for Office 插件，进入 Word/PPT/Excel。", "reason": null, "tags": ["xAI","Grok","办公工具"], "topics": ["大模型"], "firsthand": true, "editor_pick": false},{"id": "2026-06-20-dataland", "url": "https://www.transwonder.top/insights/#2026-06-20-dataland", "title": "Dataland：全球首座 AI 艺术博物馆在洛杉矶开幕", "date": "2026-06-20", "category": "赛展", "source": "The Art Newspaper / Artnet News / NPR / Google Blog", "src": "https://news.artnet.com/art-world/refik-anadol-dataland-review-2-2781630", "sources": null, "summary": "全球首座 AI 艺术博物馆 Dataland 在洛杉矶开幕，AI 生成艺术从实验室与网络走向机构化展陈。", "reason": null, "tags": ["加更","Refik Anadol","Dataland","AI艺术博物馆","沉浸式装置","Google","洛杉矶"], "topics": ["Google","艺术与展览"], "firsthand": true, "editor_pick": false},{"id": "2026-06-20-comfyui-v0251", "url": "https://www.transwonder.top/insights/#2026-06-20-comfyui-v0251", "title": "ComfyUI v0.25.1 发布，Kling V3-Turbo 正式接入", "date": "2026-06-20", "category": "产品", "source": "CSDN / AI认知工坊", "src": "https://blog.csdn.net/weixin_48502062/article/details/162153176", "sources": null, "summary": "ComfyUI v0.25.1 正式发布，核心更新为 Kling V3-Turbo 节点接入，并附入门指南，进一步扩展视频生成工作流能力。", "reason": null, "tags": ["ComfyUI","Kling","视频生成","工作流"], "topics": ["快手","ComfyUI","视频生成"], "firsthand": true, "editor_pick": false},{"id": "2026-06-20-baoyu-design", "url": "https://www.transwonder.top/insights/#2026-06-20-baoyu-design", "title": "宝玉迭代 baoyu-design Skill：支持 AI 配图并导出 PPTX", "date": "2026-06-20", "category": "产品", "source": "宝玉（@dotey）", "src": "https://x.com/dotey", "sources": null, "summary": "宝玉迭代 baoyu-design：支持 AI 配图并导出 PPTX。", "reason": null, "tags": ["Skill","AI配图","PPT"], "topics": ["设计工具","智能体与工作流"], "firsthand": false, "editor_pick": false},{"id": "2026-06-19-humanize-ppt", "url": "https://www.transwonder.top/insights/#2026-06-19-humanize-ppt", "title": "Humanize PPT v0.9：为「演讲」而生的开源 PPT Skill", "date": "2026-06-19", "category": "产品", "source": "卡尔的AI沃茨 / GitHub", "src": "https://github.com/LearnPrompt/humanize-ppt", "sources": null, "summary": "Humanize PPT v0.9：为『演讲』而生的开源 PPT Skill。", "reason": null, "tags": ["开源","PPT","Skill"], "topics": ["设计工具","智能体与工作流"], "firsthand": true, "editor_pick": false},{"id": "2026-06-19-comfyui-v3-nodes", "url": "https://www.transwonder.top/insights/#2026-06-19-comfyui-v3-nodes", "title": "ComfyUI 核心节点迁移 V3 架构，修复 FP8 scaled LoRA 问题", "date": "2026-06-19", "category": "产品", "source": "ComfyUI 更新日志", "src": "https://docs.comfy.org/zh/changelog", "sources": null, "summary": "ComfyUI 核心节点迁移 V3 架构，修复 FP8 scaled LoRA 问题。", "reason": null, "tags": ["ComfyUI","LoRA","工作流"], "topics": ["ComfyUI","智能体与工作流"], "firsthand": true, "editor_pick": false},{"id": "2026-06-18-siff-ai-tech-unit", "url": "https://www.transwonder.top/insights/#2026-06-18-siff-ai-tech-unit", "title": "上海国际电影节设立技术创制单元与 AI Backlot 计划：AI 进入主流电影节创作机制", "date": "2026-06-18", "category": "行业", "source": "Variety / Hollywood Reporter / China.org.cn", "src": "https://variety.com/2026/film/news/shanghai-film-fest-tech-unit-ai-industry-push-1236780477/", "sources": null, "summary": "第 28 届上海国际电影节（6/12–22）设立「技术创制单元」并推出 AI Backlot 计划，全球收到 500+ 申请，4 支跨国团队在电影节开放工作室实时制作 AI 短片，标志 AI 正式进入主流电影节的创作机制。", "reason": null, "tags": ["上海国际电影节","AI电影","AI Backlot","行业机制","AIGC"], "topics": ["视频生成","艺术与展览","行业与融资"], "firsthand": false, "editor_pick": false},{"id": "2026-06-18-sasha-stiles-art-basel-ai-poetry", "url": "https://www.transwonder.top/insights/#2026-06-18-sasha-stiles-art-basel-ai-poetry", "title": "AI 诗人 Sasha Stiles 在 Art Basel 2026 发布 NFC 版画新作《WORDS BEYOND WORDS》", "date": "2026-06-18", "category": "案例", "source": "Art Basel", "src": "https://www.artbasel.com/stories/sasha-stiles-how-technology-renews-relationship-with-storytelling", "sources": null, "summary": "AI 诗人 Sasha Stiles 与定制 AI 分身 Technelegy 共创的限量版画《WORDS BEYOND WORDS》在 Art Basel Basel 2026 发布，作品内嵌 NFC 芯片，提供可验证出处与独家数字内容。", "reason": null, "tags": ["AI诗歌","Art Basel","生成艺术","NFC艺术"], "topics": ["艺术与展览"], "firsthand": true, "editor_pick": false},{"id": "2026-06-18-runway-gen45", "url": "https://www.transwonder.top/insights/#2026-06-18-runway-gen45", "title": "Runway Gen-4.5 + Studio 内置编辑器上线：视频生成与后期整合进单一工具", "date": "2026-06-18", "category": "模型", "source": "Runway 官方", "src": "https://runwayml.com/changelog", "sources": null, "summary": "Runway 上线 Studio 内置编辑器，Gen-4.5 居文生视频基准榜首。", "reason": null, "tags": ["Runway","Gen-4.5","视频生成","Studio","AI影像","后期制作"], "topics": ["Runway","视频生成","图像生成"], "firsthand": true, "editor_pick": false},{"id": "2026-06-18-hf-oft-lora", "url": "https://www.transwonder.top/insights/#2026-06-18-hf-oft-lora", "title": "HuggingFace：别只用 LoRA——公平基准下 OFT 等技术更优", "date": "2026-06-18", "category": "论文", "source": "HuggingFace Blog", "src": "https://huggingface.co/blog", "sources": null, "summary": "HuggingFace：公平基准下 OFT 等技术在部分任务上优于 LoRA。", "reason": null, "tags": ["PEFT","LoRA","微调","模型训练"], "topics": ["研究论文"], "firsthand": true, "editor_pick": false},{"id": "2026-06-18-graduation-show-2026-ai", "url": "https://www.transwonder.top/insights/#2026-06-18-graduation-show-2026-ai", "title": "2026 届毕设展 AI 含量攀升：央美、清华、中传、国美深度调研", "date": "2026-06-18", "category": "案例", "source": "每日经济新闻", "src": "https://www.mrjjxw.com/articles/2026-06-18/4431297.html", "sources": null, "summary": "每日经济新闻对央美、清华、中传、国美四校 2026 届毕设展的深度调研显示，AI 辅助创作比例明显攀升，呈现方式与评价标准随之分化。", "reason": null, "tags": ["毕业设计","AI艺术","高校","艺术设计教育"], "topics": ["艺术与展览","设计教育"], "firsthand": true, "editor_pick": false},{"id": "2026-06-18-google-vids", "url": "https://www.transwonder.top/insights/#2026-06-18-google-vids", "title": "Google Vids 三连更：Veo 3.1 对所有账号免费、Lyria 3 AI 音乐生成、AI 虚拟人扩容至 53 种", "date": "2026-06-18", "category": "产品", "source": "Google Workspace Blog", "src": "https://workspaceupdates.googleblog.com/2026/06/enhanced-ai-avatars-vids.html", "sources": null, "summary": "Google Vids 升级：Veo 3.1 免费、Lyria 3 AI 配乐、AI 虚拟人扩至 53 种。", "reason": null, "tags": ["Google","Veo 3.1","Lyria 3","AI虚拟人","视频生成","AI音乐","Google Vids"], "topics": ["Google","视频生成","音频与音乐"], "firsthand": true, "editor_pick": false},{"id": "2026-06-18-freestyle-style-transfer", "url": "https://www.transwonder.top/insights/#2026-06-18-freestyle-style-transfer", "title": "FreeStyle：复旦+阶跃星辰等联合提出社区 LoRA 驱动的大规模任意风格迁移方法", "date": "2026-06-18", "category": "论文", "source": "复旦大学 / 阶跃星辰 / 西湖大学 / 港大", "src": "https://arxiv.org/abs/2606.20506", "sources": null, "summary": "复旦大学、阶跃星辰、西湖大学与港大联合提出 FreeStyle（arXiv:2606.20506），通过系统挖掘社区 LoRA 构建大规模内容×风格双参考数据集，实现任意风格迁移并完整保留内容结构，支持水墨、刺绣、油画、吉卜力等多样风格。", "reason": null, "tags": ["风格迁移","LoRA","图像生成","开源"], "topics": ["图像生成","研究论文"], "firsthand": true, "editor_pick": false},{"id": "2026-06-18-figma-ai-design-agent-aleph2", "url": "https://www.transwonder.top/insights/#2026-06-18-figma-ai-design-agent-aleph2", "title": "Figma AI 设计 Agent 接入联网搜索，Runway Aleph 2.0 嵌入 Weave 画布", "date": "2026-06-18", "category": "产品", "source": "Figma / Runway", "src": "https://www.figma.com/blog/direct-every-frame-with-runway-aleph-2/", "sources": null, "summary": "Figma AI 设计 Agent 新增联网搜索能力，Runway Aleph 2.0 同步嵌入 Figma Weave 画布，设计师可在 Figma 内直接用文字提示编辑视频帧。", "reason": null, "tags": ["Figma","Runway","设计工作流","AI视频编辑"], "topics": ["Figma","Runway","视频生成"], "firsthand": true, "editor_pick": false},{"id": "2026-06-18-adobe-firefly-project-memory", "url": "https://www.transwonder.top/insights/#2026-06-18-adobe-firefly-project-memory", "title": "Adobe Firefly「项目记忆」私测启动：跨会话保存角色设定、产品角度与品牌资产", "date": "2026-06-18", "category": "产品", "source": "Adobe 官方 / WinBuzzer", "src": "https://winbuzzer.com/2026/06/19/adobe-firefly-adds-project-memory-in-private-beta-xcxwbn/", "sources": null, "summary": "Firefly 私测『项目记忆』，跨会话保存角色、产品角度与品牌资产，保持连续创作一致性。", "reason": null, "tags": ["Adobe","Firefly","项目记忆","Elements","Projects","持续创作"], "topics": ["Adobe"], "firsthand": true, "editor_pick": false},{"id": "2026-06-18-adobe-firefly-creative-skills", "url": "https://www.transwonder.top/insights/#2026-06-18-adobe-firefly-creative-skills", "title": "Adobe Firefly Creative Skills 上线：对话式设计与 AI 助手进驻全系 CC", "date": "2026-06-18", "category": "产品", "source": "搜狐科技", "src": "https://www.sohu.com/a/1038594646_121956424", "sources": null, "summary": "Adobe Firefly Creative Skills 发布，对话式 AI 设计能力与 AI 助手全面入驻 Creative Cloud 全系应用，独立于同期「项目记忆」更新。", "reason": null, "tags": ["Adobe","Firefly","Creative Cloud","AI助手"], "topics": ["Adobe","智能体与工作流"], "firsthand": false, "editor_pick": false},{"id": "2026-06-18-adobe-creative-agents", "url": "https://www.transwonder.top/insights/#2026-06-18-adobe-creative-agents", "title": "Adobe 把「创意智能体」铺进 Photoshop、Premiere、Illustrator、InDesign（公测）", "date": "2026-06-18", "category": "产品", "source": "Adobe 官方 / Engadget", "src": "https://news.adobe.com/news/2026/06/adobe-unveils-major-expansion", "sources": null, "summary": "Adobe 把创意智能体铺进 PS/Pr/Ai/Id 公测，AI 从生成器转向流程助手。", "reason": null, "tags": ["Adobe","创意工具","Agent","设计工作流"], "topics": ["Adobe","智能体与工作流"], "firsthand": true, "editor_pick": false},{"id": "2026-06-17-tsinghua-art-grad-show-2026", "url": "https://www.transwonder.top/insights/#2026-06-17-tsinghua-art-grad-show-2026", "title": "清华美院 2026 届本科毕业展「艺科融合」：AI 辅助创作在多院系作品中成为常态", "date": "2026-06-17", "category": "案例", "source": "新浪财经 / 清华美院", "src": "https://finance.sina.com.cn/roll/2026-06-17/doc-inictnur5949943.shtml", "sources": null, "summary": "清华美术学院 2026 届本科毕业展「艺科融合」6 月 12—25 日在清华艺术博物馆展出，240 余名毕业生近千件作品中，AI 辅助创作已作为普通工具出现在多个专业院系。", "reason": null, "tags": ["清华美院","毕业展","AI辅助创作","艺术教育"], "topics": ["艺术与展览","设计教育"], "firsthand": false, "editor_pick": false},{"id": "2026-06-17-meitu-qixiang-v6", "url": "https://www.transwonder.top/insights/#2026-06-17-meitu-qixiang-v6", "title": "美图奇想大模型 V6 发布，同步推出 Picchi / Artflo / MVLAND 等四款 AI 新品", "date": "2026-06-17", "category": "模型", "source": "美图 / 新浪科技", "src": "https://finance.sina.com.cn/tech/roll/2026-06-17/doc-inicttap5914460.shtml", "sources": null, "summary": "美图奇想大模型 V6 正式发布，同步推出 Picchi、Artflo、MVLAND、MeituHub 四款 AI 新品及站酷 AI 工具箱，覆盖图像生成至多媒体创作全流程。", "reason": null, "tags": ["美图","奇想大模型","AI设计工具","图像生成"], "topics": ["图像生成","大模型","设计工具"], "firsthand": false, "editor_pick": false},{"id": "2026-06-16-disney-adobe-firefly-foundry", "url": "https://www.transwonder.top/insights/#2026-06-16-disney-adobe-firefly-foundry", "title": "Disney 幻想工程 × Adobe Firefly Foundry：训练 IP 自定义模型，草图直出品牌级概念图", "date": "2026-06-16", "category": "案例", "source": "Adobe 官方", "src": "https://news.adobe.com/news/2026/06/adobe-and-disney-imagineering-collaborate", "sources": null, "summary": "Disney 幻想工程与 Adobe Firefly Foundry 合作，基于内部 IP 资产训练自定义生成模型，可从草图生成米奇、冰雪奇缘、汽车总动员等 IP 的品牌级概念图，生成结果已直接导入商业制作流程，为 Firefly Foundry 企业自定义模型的标志性落地案例。", "reason": null, "tags": ["企业案例","Adobe","Disney","IP生成"], "topics": ["Adobe"], "firsthand": true, "editor_pick": false},{"id": "2026-06-16-digital-smart-art-theory", "url": "https://www.transwonder.top/insights/#2026-06-16-digital-smart-art-theory", "title": "《数智艺术论》央美首发：56 万字，提出「50%+30%+20%」人才培养框架", "date": "2026-06-16", "category": "论文", "source": "央美美术馆 / 四川美术出版社", "src": "https://www.cafamuseum.org/exhibit/newsdetail/4105", "sources": null, "summary": "《数智艺术论》在中央美术学院首发，56 万字，五大板块 13 专题，提出「50% 专业+30% 跨学科+20% AI 基础」的高校人才培养框架，填补国内数智艺术系统性理论空白。", "reason": null, "tags": ["数智艺术","学术专著","人才培养","艺术教育"], "topics": ["设计教育","研究论文"], "firsthand": true, "editor_pick": false},{"id": "2026-06-16-boogu-image-01", "url": "https://www.transwonder.top/insights/#2026-06-16-boogu-image-01", "title": "Boogu-Image-0.1 开源：10B 参数统一图像生成与编辑模型，Apache-2.0", "date": "2026-06-16", "category": "模型", "source": "GitHub / 机器之心 / HuggingFace", "src": "https://github.com/boogu-project/Boogu-Image", "sources": null, "summary": "Boogu-Image-0.1 以 Apache-2.0 协议开源，10B 参数统一图像生成与编辑，中英双语密集文本渲染为开源第一，ComfyUI 整合包同步上线。", "reason": null, "tags": ["开源模型","图像生成","文字渲染","ComfyUI"], "topics": ["ComfyUI","图像生成"], "firsthand": true, "editor_pick": false},{"id": "2026-06-16-ai-higher-ed-art-reform", "url": "https://www.transwonder.top/insights/#2026-06-16-ai-higher-ed-art-reform", "title": "《艺术教育》：AI 赋能高校艺术设计学科重构与革新", "date": "2026-06-16", "category": "观点", "source": "艺术教育杂志", "src": "https://www.yjys.cn/", "sources": null, "summary": "《艺术教育》刊文梳理 AI 技术介入高校艺术设计学科的路径，探讨课程体系、教学方法与评价标准的重构方向。", "reason": null, "tags": ["艺术设计教育","AI","学科重构","高校"], "topics": ["设计教育"], "firsthand": true, "editor_pick": false},{"id": "2026-06-15-minimax-hub", "url": "https://www.transwonder.top/insights/#2026-06-15-minimax-hub", "title": "MiniMax Hub 发布：图·视频·配音·音乐·剪辑一站式多模态 AI 创作平台", "date": "2026-06-15", "category": "产品", "source": "Variety / Sixth Tone / 多源", "src": "https://variety.com/2026/film/festivals/minimax-hub-ai-video-generator-shanghai-film-festival-1236781901/", "sources": null, "summary": "MiniMax 于上海国际电影节期间发布 MiniMax Hub，整合图像、视频、配音、音乐与剪辑的一站式多模态创作平台，由 Agent 拆解任务并保留人工决策节点，4 支跨国团队用于电影节 AI 短片制作。", "reason": null, "tags": ["MiniMax","多模态","AI视频","创作平台","上海国际电影节"], "topics": ["视频生成","大模型","音频与音乐"], "firsthand": false, "editor_pick": false},{"id": "2026-06-15-freeorbit4d", "url": "https://www.transwonder.top/insights/#2026-06-15-freeorbit4d", "title": "FreeOrbit4D：单目视频→任意相机轨迹重拍，无需训练，单卡可跑（CVPR 2026）", "date": "2026-06-15", "category": "论文", "source": "arXiv / CVPR 2026 / 机器之心", "src": "https://arxiv.org/abs/2601.18993", "sources": null, "summary": "UIUC 与 Eyeline Labs 提出 FreeOrbit4D：输入单目视频，无需额外训练即可重拍出任意相机轨迹视角；于 CVPR 2026 Workshop 展示，代码开源，单卡 A40 可运行。", "reason": null, "tags": ["4D重建","相机控制","视频生成","CVPR"], "topics": ["视频生成","研究论文"], "firsthand": true, "editor_pick": false},{"id": "2026-06-15-dreamx-world-1", "url": "https://www.transwonder.top/insights/#2026-06-15-dreamx-world-1", "title": "高德 AI 团队开源 DreamX-World 1.0 通用交互式世界模型", "date": "2026-06-15", "category": "论文", "source": "arXiv / 机器之心 / chooseAI", "src": "https://arxiv.org/abs/2606.16993", "sources": null, "summary": "高德 AI 团队在 arXiv 发布 DreamX-World 1.0，可生成 1 分钟长视频，支持 6 自由度相机控制与「世界记忆」机制，采用 MIT+Apache-2.0 双协议开源。", "reason": null, "tags": ["世界模型","开源","视频生成","arXiv"], "topics": ["视频生成","3D 与世界模型","研究论文"], "firsthand": true, "editor_pick": false},{"id": "2026-06-10-midjourney-v81", "url": "https://www.transwonder.top/insights/#2026-06-10-midjourney-v81", "title": "Midjourney V8.1 成为默认版本：生成速度 4–5 倍提升，原生 2K 高清输出", "date": "2026-06-10", "category": "模型", "source": "Midjourney 官方", "src": "https://docs.midjourney.com/hc/en-us/articles/32199405667853-Version", "sources": null, "summary": "Midjourney V8.1 成默认版本，速度提升约 4–5 倍，支持原生 2K 输出。", "reason": null, "tags": ["Midjourney","V8.1","图像生成","2K","速度提升"], "topics": ["Midjourney","图像生成"], "firsthand": true, "editor_pick": false},{"id": "2026-06-10-midjourney-niji-v7", "url": "https://www.transwonder.top/insights/#2026-06-10-midjourney-niji-v7", "title": "Midjourney Niji V7 正式上线：动漫风格连贯性与 sref 风格参考全面升级", "date": "2026-06-10", "category": "模型", "source": "Midjourney 官方", "src": "https://www.midjourney.com/updates", "sources": null, "summary": "Niji V7 发布，动漫/插画风格连贯性与 sref 风格参考全面升级。", "reason": null, "tags": ["Midjourney","Niji V7","动漫","插画","风格参考"], "topics": ["Midjourney"], "firsthand": true, "editor_pick": false},{"id": "2026-06-08-midjourney-v7-review", "url": "https://www.transwonder.top/insights/#2026-06-08-midjourney-v7-review", "title": "Midjourney V7 深度评测：人物一致性 88%，图内文字可读", "date": "2026-06-08", "category": "教程", "source": "AIHotTool / 多源", "src": "https://www.aihottool.com/midjourney-v7-2026-review/", "sources": null, "summary": "多方深度评测显示 Midjourney V7 人物一致性达 88%，图内文字实现可读渲染，较 V6 在角色连贯性和排版精度上有显著提升。", "reason": null, "tags": ["Midjourney","V7","人物一致性","文字渲染"], "topics": ["Midjourney"], "firsthand": true, "editor_pick": false},{"id": "2026-06-05-runway-seedance2-api", "url": "https://www.transwonder.top/insights/#2026-06-05-runway-seedance2-api", "title": "Runway API 接入 Seedance 2.0 Fast，支持关键帧控制与参考图生成", "date": "2026-06-05", "category": "产品", "source": "Runway", "src": "https://releasebot.io/updates/runwayai", "sources": null, "summary": "Runway API 正式接入 Seedance 2.0 Fast，支持关键帧控制、参考图及参考视频生成，输出 4–15 秒视频，提供 480p/720p 两档分辨率。", "reason": null, "tags": ["Runway","Seedance","视频生成","API"], "topics": ["字节跳动","Runway","视频生成"], "firsthand": false, "editor_pick": false},{"id": "2026-06-03-ideogram-4-open-weights", "url": "https://www.transwonder.top/insights/#2026-06-03-ideogram-4-open-weights", "title": "Ideogram 4.0 发布首个开放权重文生图模型，ComfyUI 首日支持", "date": "2026-06-03", "category": "模型", "source": "Ideogram 官方", "src": "https://ideogram.ai/news/ideogram-4.0/", "sources": null, "summary": "Ideogram 发布 4.0 开放权重文生图模型（9.3B DiT），支持 JSON 精控色彩与文字布局，ComfyUI 首日支持，可本地免费部署。", "reason": null, "tags": ["Ideogram","开源模型","ComfyUI","文生图"], "topics": ["ComfyUI","图像生成"], "firsthand": true, "editor_pick": false},{"id": "2026-06-02-scorsese-flux", "url": "https://www.transwonder.top/insights/#2026-06-02-scorsese-flux", "title": "斯科塞斯用 FLUX 为好莱坞新片分镜，出任 Black Forest Labs 顾问", "date": "2026-06-02", "category": "案例", "source": "Variety / Hollywood Reporter", "src": "https://variety.com/2026/film/news/martin-scorsese-supports-ai-company-storyboard-movies-1236765037/", "sources": null, "summary": "斯科塞斯用 FLUX 为新片绘制分镜，并出任 Black Forest Labs 顾问。", "reason": null, "tags": ["Martin Scorsese","FLUX","Black Forest Labs","分镜","好莱坞","AI辅助创作"], "topics": ["Black Forest Labs"], "firsthand": false, "editor_pick": true},{"id": "2026-06-01-iciad-aigc", "url": "https://www.transwonder.top/insights/#2026-06-01-iciad-aigc", "title": "2026 ICIAD 国际文化创新艺术设计奖设「AIGC 与计算设计」类别", "date": "2026-06-01", "category": "赛展", "source": "设计竞赛网", "src": "https://www.shejijingsai.com/2026/03/1517443.html", "sources": null, "summary": "2026 ICIAD 设『AIGC 与计算设计』类别（原始时间：2026-06）。", "reason": null, "tags": ["赛事","AIGC","生成艺术"], "topics": ["艺术与展览"], "firsthand": true, "editor_pick": false},{"id": "2026-06-01-haochuangyi-baidu-aigc", "url": "https://www.transwonder.top/insights/#2026-06-01-haochuangyi-baidu-aigc", "title": "2026 中国好创意大赛 · 百度 AIGC 未来创作专项赛", "date": "2026-06-01", "category": "赛展", "source": "设计竞赛网", "src": "https://www.shejijingsai.com/2026/02/1501689.html", "sources": null, "summary": "2026 中国好创意大赛设百度 AIGC 未来创作专项赛（原始时间：2026-06）。", "reason": null, "tags": ["赛事","高校","AIGC"], "topics": ["设计教育","艺术与展览"], "firsthand": true, "editor_pick": false},{"id": "2026-06-01-comfyui-v0230", "url": "https://www.transwonder.top/insights/#2026-06-01-comfyui-v0230", "title": "ComfyUI v0.23.0 发布：新增 NVIDIA PixelDiT 与 Microsoft Lens 模型支持", "date": "2026-06-01", "category": "产品", "source": "ComfyUI 官方 / NVIDIA Blog", "src": "https://docs.comfy.org/changelog", "sources": null, "summary": "ComfyUI v0.23.0 新增 NVIDIA PixelDiT 与 Microsoft Lens 模型支持。", "reason": null, "tags": ["ComfyUI","FLUX.2","NVIDIA","PixelDiT","本地部署","工作流","开源"], "topics": ["ComfyUI","智能体与工作流"], "firsthand": true, "editor_pick": false},{"id": "2026-05-21-siggraph-art-gallery", "url": "https://www.transwonder.top/insights/#2026-05-21-siggraph-art-gallery", "title": "SIGGRAPH 2026 Art Gallery「In-Betweens」入选作品公布，7 月洛杉矶登场", "date": "2026-05-21", "category": "赛展", "source": "SIGGRAPH 2026 官网", "src": "https://s2026.siggraph.org/program/art-gallery/", "sources": null, "summary": "SIGGRAPH 2026 Art Gallery『In-Betweens』入选公布，7 月洛杉矶登场。", "reason": null, "tags": ["SIGGRAPH","Art Gallery","交互装置","生成艺术","科技艺术","洛杉矶","学生竞赛"], "topics": ["研究论文","艺术与展览"], "firsthand": true, "editor_pick": false}]}