[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-f884bd95-631b-4904-8e8d-a8db751f9314":3},{"id":4,"title":5,"summary":6,"original_url":7,"source_id":8,"tags":9,"published_at":20,"created_at":21,"modified_at":22,"is_published":23,"publish_type":24,"image_url":13,"view_count":25},"f884bd95-631b-4904-8e8d-a8db751f9314","MV-Forcing：用「4D 几何桥」打通「长 × 多视角」,单一扩散模型端到端跑出 4D 视频","视频扩散两条赛道一直咬不上:时间自回归把单视角拉到分钟级(Sora、Veo 3.1);双向注意力做多视角一致(VideoMV、4Diffusion),却只能撑几秒静态。Cornell Tech 的 Fiebelman 等人在 **arXiv:2607.05376** 提出 MV-Forcing:以自回归 3D 重建作「4D 几何桥」传递视角间先验,让单一扩散模型同时吃下「长」与「多视角一致」。  机制分三层。 几何桥负责对齐——源视角 3D 重建后渲染下一视角的深度、法线、位姿先验,交给扩散做高频细节,3D 守一致、扩散守保真。联合去噪让两视角槽位都从噪声起互给先验,绕开 teacher 固定窗口,使生成真正无界。DMD + Spatio-Temporal Self-Forcing 把 few-step student 蒸馏出来,用视频级损失修补 exposure bias,延续 Xun Huang 团队 Self Forcing 思路,这次同时盖住时间与视角两个自回归轴。  为什么值得看。 世界模型与自动驾驶仿真要的正是「任意长度、任意视点、物理一致」的 4D 场景,此前只能堆算力或拼多通道管线。MV-Forcing 提出「几何归 3D,纹理归扩散」的轻量范式,若被更大模型验证,工业级 4D 视频生成的成本曲线有望再下一档。","https:\u002F\u002Farxiv.org\u002Fabs\u002F2607.05376","7437aeb9-930c-4866-a2e9-48003c1a792b",[10,14,17],{"id":11,"name":12,"slug":12,"description":13,"color":13},"5e628969-6d2a-437f-998a-104e4b16cfb1","ai-progress",null,{"id":15,"name":16,"slug":16,"description":13,"color":13},"7b67033c-19e6-4052-a626-e681bba64c7a","diffusion",{"id":18,"name":19,"slug":19,"description":13,"color":13},"ebe5dcd1-46b1-4298-b8c2-8e0e2f456e56","video-generation","2026-07-08T04:00:00Z","2026-07-08T04:08:17.182511Z","2026-07-08T04:08:17.182526Z",true,"agent",2]