Unverified50% confidenceSolutionExact time
Mosaic Chunk introduces a selective memory mechanism that solves the problem of video models forgetting objects that temporarily leave the frame, and the code and model have been open-sourced.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/20/2026
First Seen
Valid until: 1/18/2027
Sources
Related Entities
Related Claims
Unverified被动观看视频激活的是陈述性记忆系统，而编程能力依赖程序性记忆，需要通过反复实践才能内化73% similarUnverifiedHigh-end video models typically introduce temporal attention mechanisms or character identity embedding modules, allocating more computational resources to subject features during inference to suppress inter-frame drift.70% similarUnverifiedVideo Depth Anything builds on Depth Anything V2 with spatiotemporal attention heads, a temporal consistency loss, and strategies for long videos, making temporal behavior part of the model and inference design.69% similarUnverifiedThe scene graph serves as a generative constraint that guides the sampling of diverse reset states, thereby covering object poses and grasp modes that do not appear in the video.68% similarUnverified即梦视频生成插件采用了类似Sora的视频大模型架构，在图像扩散模型基础上增加了时间维度的一致性约束67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1078340API
curl https://kongchang.com/api/v1/knowledge/claims/1078340MCP
get_claim(id=1078340)