[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-video-diffusion-rope-physics-fix":3,"topics-all":38,"news-related-b3835eee-bafd-430b-93f2-1d209b572751":57},{"id":4,"title":5,"summary":6,"content":7,"original_url":8,"source_id":9,"tags":10,"translations":24,"news_slug":31,"published_at":32,"created_at":33,"modified_at":34,"is_published":35,"publish_type":36,"image_url":14,"view_count":37},"b3835eee-bafd-430b-93f2-1d209b572751","视频里的篮球为何悬空？病根是RoPE空间注意力衰减","视频扩散模型常把物理规律画崩，篮球弹到一半悬在半空。北邮团队对Wan2.1-T2V-1.3B做可解释性分析：运动规划集中在去噪前5步，3D RoPE的空间锚定偏置让错误轨迹过早锁定；只需在前5步缩小空间RoPE频率，无需重训即可缓解，VideoPhy 344个案例人评验证有效。","生成的视频里，篮球从半空落下、弹了几次，突然在半空中定住——同一个提示词换四个随机种子，失败模式还各不相同。这不是素材库翻车集锦，而是最先进视频扩散模型的日常输出。北邮智能科学技术中心的团队没有像多数工作那样往模型里塞物理先验或专用数据，而是打开模型内部，问了一个更根本的问题：运动轨迹到底在哪一步、被什么机制定下来？\n\n## 运动规划：前 5 步定生死\n\n团队以 Wan2.1-T2V-1.3B 为分析对象，发现文生视频的「运动规划」——把文本语义落到物体逐帧位置的过程——主要发生在 50 个去噪步的前 5 步，呈现「先定形状、再补细节」的规律。他们把跨注意力里的轨迹模式和因果干预实验结合起来，从众多注意力头里分出真正驱动运动的那一批：有的头只是「展示」运动模式，有的头才是「决定」运动的角色。\n\n## 病根：RoPE 的空间锚定偏置\n\n顺着这条线查自注意力，问题浮出水面：3D RoPE（旋转位置编码）带来空间锚定偏置——一个 query 倾向关注其他帧里空间上邻近的坐标，远处那个物理上合理的候选位置拿到的注意力严重不足。更麻烦的是「早承诺」效应：候选区域本来在互相竞争，一旦某几帧提前安定，空间锚定立刻给其他帧的邻近位置「拉票」，哪怕那违反物理。论文里最典型的失败种子里，第 10 帧稳定后，第 14 帧从较低的候选切换到较高的候选，篮球就此悬在半空。\n\n## 修法很轻：只动前 5 步的 RoPE 频率\n\n根因诊断出来，修复反而便宜。训练自由版：只在去噪前 5 步缩小高、宽两个方向的 RoPE 频率，削弱过度的空间衰减，之后恢复原频率，完全不动权重；训练版用 LoRA 适配自注意力与跨注意力，推理时前 5 步取 λ=0.70。在 VideoPhy 物理常识基准的 344 个案例上，由人工评判语义遵循与物理合理性，固体交互类场景提升最大。论文自称是首个针对文生视频扩散模型运动规划的可解释性研究，代码与项目页已随论文放出（https:\u002F\u002Farxiv.org\u002Fabs\u002F2609.23658，https:\u002F\u002Fsiriuslala.github.io\u002Fphysics）。\n\n## 所以呢\n\n几天前我们还在聊怎么给视频模型的物理能力打分，这周有人直接把病理切片做出来了。真正值得行业咀嚼的是那个反直觉结论：物理问题未必需要物理方案，调一下位置编码的频率就能止血。视频模型某些「不理解世界」的瞬间，可能只是注意力预算在空间上分配得太早、太死。下次再看到物体悬浮，先别急着骂数据——看一眼 RoPE。","https:\u002F\u002Farxiv.org\u002Fabs\u002F2609.23658","7437aeb9-930c-4866-a2e9-48003c1a792b",[11,15,18,21],{"id":12,"name":13,"slug":13,"description":14,"color":14},"7b67033c-19e6-4052-a626-e681bba64c7a","diffusion",null,{"id":16,"name":17,"slug":17,"description":14,"color":14},"01598627-1ea6-4b27-a5d8-874971571a71","llm",{"id":19,"name":20,"slug":20,"description":14,"color":14},"4f214978-cac1-4f39-aa4b-f92a0d0934b7","transformer",{"id":22,"name":23,"slug":23,"description":14,"color":14},"ebe5dcd1-46b1-4298-b8c2-8e0e2f456e56","video-generation",[25],{"id":26,"lang":27,"title":28,"summary":29,"content":30},"5d2a08b3-655a-4934-9bb9-ddd0bbfe81b3","en","Why Video Diffusion Breaks Physics: The RoPE Attention Flaw","BUPT traces video physics failures to 3D RoPE anchoring in early denoising; a training-free frequency fix passed 344 VideoPhy cases.","Generated video has a blooper reel nobody asked for: a basketball drops from mid-air, bounces a few times, then freezes in place — and with the same prompt across four random seeds, the failure modes differ each time. These aren't cherry-picked fails; they're the everyday output of state-of-the-art video diffusion models. Instead of stuffing physics priors or curated data into the model, a team from BUPT's Center of Intelligent Science and Technology opened the model up and asked a more basic question: at which step, and by which mechanism, is a motion trajectory actually decided?\n\n## Motion planning: the first 5 steps decide everything\n\nWorking with Wan2.1-T2V-1.3B, the team found that \"motion planning\" — the process that lands text semantics onto object positions frame by frame — happens largely within the first 5 of 50 denoising steps, following a \"first shape, then details\" pattern. Combining cross-attention trajectory patterns with causal intervention experiments, they separated the attention heads that genuinely drive motion from those that merely display its pattern.\n\n## The culprit: RoPE's spatial anchoring bias\n\nFollowing that thread into self-attention exposed the root cause: 3D RoPE (Rotary Position Embedding) induces a spatial anchoring bias — a query tends to attend to spatially nearby coordinates in other frames, so a physically plausible candidate farther away receives far too little attention. Worse is the \"early commitment\" effect: candidate regions initially compete, but once a few frames settle early, spatial anchoring rallies support for nearby positions in other frames even when they violate physics. In the paper's most emblematic failing seed, frame 14 switches from a lower candidate to a higher one after frame 10 stabilizes — and the basketball simply hangs in mid-air.\n\n## The fix is light: rescale RoPE frequency in the first 5 steps\n\nWith the root cause diagnosed, the fix turns out to be cheap. Training-free: scale down the height and width RoPE frequencies during only the first 5 denoising steps to weaken the excessive spatial decay, then restore the originals — no weight updates required. The training-based variant adapts self- and cross-attention with LoRA and uses λ = 0.70 for the first 5 steps at inference. On 344 VideoPhy cases judged by humans for semantic adherence and physical commonsense, the largest gains came on interactions involving solids. The paper bills itself as the first interpretability study of the motion-planning process in text-to-video diffusion models; code and project page are released alongside (https:\u002F\u002Farxiv.org\u002Fabs\u002F2609.23658, https:\u002F\u002Fsiriuslala.github.io\u002Fphysics).\n\n## So what\n\nDays ago the conversation was still about how to score video models' physics; now someone has produced the pathology slide. The counterintuitive takeaway worth chewing on: a physics problem didn't need a physics solution — adjusting the frequency of a position encoding stopped the bleeding. Some of what looks like \"not understanding the world\" in video models may just be attention budget allocated too early and too rigidly in space. Next time an object levitates, don't blame the data first — check the RoPE.","video-diffusion-rope-physics-fix","2026-09-22T15:07:01Z","2026-09-22T15:10:34.053427Z","2026-09-22T15:10:34.053437Z",true,"agent",548,[39,48],{"slug":40,"tag_slug":40,"title_zh":41,"title_en":42,"intro_zh":43,"intro_en":44,"id":45,"is_active":35,"created_at":46,"modified_at":47},"ai-for-science","AI for Science 2026：从 UniPert 到 GPT-Rosalind 的硬核进化","AI for Science 2026: from UniPert to GPT-Rosalind","生命科学、化学材料、物理世界模型——AI 正在从\"语言工具\"变成\"实验伙伴\"。本专题收录 AI 在三大科学方向的关键节点：UniPert 统一基因与化学扰动空间、GPT-Rosalind 端到端生命科学推理、达摩院 AI 智能体 28 小时找到 4 种超导新材料、Anthropic Claude Science 把工作台做成标准品。","From language tool to lab partner — AI is reshaping life sciences, chemistry\u002Fmaterials, and physical world models. This topic covers the key milestones: UniPert unifying genetic-chemical perturbation spaces, GPT-Rosalind's end-to-end life-sciences reasoning, DAMO's AI agent discovering 4 superconducting materials in 28 hours, and Anthropic's Claude Science workbench going mainstream.","988a4300-5fab-41c4-b5d8-63711a2dc757","2026-09-10T01:34:15.296649Z","2026-09-10T01:34:15.296663Z",{"slug":49,"tag_slug":49,"title_zh":50,"title_en":51,"intro_zh":52,"intro_en":53,"id":54,"is_active":35,"created_at":55,"modified_at":56},"h3-series","MiniMax H3 系列：从开源权重到 35 倍吞吐","MiniMax H3 Series: from open weights to 35x throughput","MiniMax H3 自 2026 年 8 月开源以来节奏密集：官方把生成、参考与编辑收回一个模型；ComfyUI 当天压进 RTX 3060；摩尔线程 3 小时完成国产 GPU 适配；fal 后训练版把吞吐拉到 35 倍；FastH3 蒸馏再砍推理成本。本专题持续追踪 H3 的发布—开源—蒸馏—部署全链路。","Since MiniMax open-sourced H3 in August 2026 the pace has been relentless: one unified omni-modal model, same-day ComfyUI support down to an RTX 3060, a 3-hour Day-0 port to Moore Threads GPUs, fal's post-trained H3 Max at 35x throughput, and FastH3 distillation cutting inference cost further. This topic tracks the full H3 chain — release, open weights, distillation, deployment.","83ef0daa-3c31-4cb3-86ed-e5ee58654d5f","2026-09-08T07:33:19.942193Z","2026-09-08T07:33:19.942209Z",{"items":58},[59,64,69,74,79,84],{"id":60,"title":61,"news_slug":62,"published_at":63},"bcdc10bc-2f08-4c39-8ffa-e7e34041c112","京东开源 JoyAI-Video-Edit:用 16B 多模态扩散 Transformer 把视频编辑推进「边播边改」实时流时代","jd-joyai-video-edit-real-time-streaming","2026-08-05T03:00:00+00:00",{"id":65,"title":66,"news_slug":67,"published_at":68},"79c21822-4e80-43f7-afea-baa14af4ba0a","Subliminal Clocks: 扩散语言模型里那块\"潜时钟\"被找到了","subliminal-clocks-dlm-latent","2026-07-06T06:00:00+00:00",{"id":70,"title":71,"news_slug":72,"published_at":73},"a07a31d6-3a63-4452-b898-02a6e575682e","Vera：Netflix 把视频编辑拆成编辑层 + 原视频","vera-netflix-caltech-mixture-transformers-edit","2026-06-24T04:00:00+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"0bc5be19-abbb-4898-8e1d-86cb127fbb8a","长上下文调优重塑视频生成：单镜头模型学会「讲连续的故事」","lct-long-context-tuning-video-scene-consistency","2026-05-03T10:05:00+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"75228f10-58dd-4e4e-870c-4031cee599c3","Sebastian Raschka 2026预测：Transformer统治依旧，但扩散模型正悄然崛起","sebastian-raschka-2026-transformer-vs-diffusion-moe","2026-04-23T08:03:00+00:00",{"id":85,"title":86,"news_slug":87,"published_at":88},"c4976de5-150e-4dc7-bf86-68d7d5f81be1","分文本条件增强：DiT-ST革新文本到图像扩散模型","dit-st-split-text-conditioning-tongji-69pct-geneval","2026-04-21T07:00:00+00:00"]