Back to AI Lab
ArXiv Paper

OmniMem: Scalable and Adaptive Memory Retrieval for Long Video Generation

Lin Zhao, Yushu Wu, Yifan Gong +2June 1, 2026

Summary

OmniMem adds an external memory system to long-video generators so they can re-use past details instead of re-encoding full histories. It adapts which frames to keep, letting models generate longer, more coherent clips under fixed compute. If you’re chasing hour-scale video worlds, this is a template for managing context.

Related Content