MosaiChunk Gives Autoregressive Video Models Persistent Visual Memory
Researchers from Cornell, UC Berkeley, Harvard and Impossible introduce MosaiChunk, which uses selective KV memory collage to give autoregressive video generation models persistent visual memory across long horizons.
2026-10-13 ~ 2026-10-13 · 2 related posts
- MosaiChunk adds persistent visual memory to autoregressive video models — xiuyu_l · 2026-10-13
- MosaiChunk: compositing spatio-temporal KV memory for autoregressive video generation — camenduru · 2026-10-13