Study investigates the impact of visual memory in long-video generation
A study published on Hugging Face Papers investigates whether visual memory in video generation always improves the generated story.
Long-video generators increasingly retrieve past visual observations to prevent identity drift. However, because an image can be faithful to the past but incorrect for the present, questions remain as to whether visual memory always improves the generated story.
The research investigates this through VMem-Bench, a two-track diagnostic study. Track-A measures which references are selected from 92 real films (16,548 segments) before generation and without generator confounds. Track-B compares four retrieval policies with no-memory and gold-informed last-seen controls on 30 causally generated stories (2,824 segments) using a shared generator. The paper received 4 upvotes on Hugging Face Daily Papers.
This item is a summary. The full, checkable version is at the original source:
Read the original at Hugging Face Papers ↗