Most diffusion-based video models are trained on short clips. When they’re used to relight longer videos in chunks, the lighting can visibly jump between them.
Meet HorizonRelight, new work from our research team and @USC. It propagates target-domain context from one sliding window to the next, helping each section pick up where the last one left off.
This means more consistent lighting across long videos, with fewer boundary artifacts and unwanted appearance changes.
#ECCV2026 paper + demo:
context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
Why it matters
Cross-window context windowThe maximum amount of text a model can consider at once — its working memory for the current conversation or task.Full definition → propagation cuts boundary artifacts in long-video relighting, a practical fix for anyone building video-generation pipelines that need visual consistency over time.
Key quotes
“Most diffusion-based video models are trained on short clips.”
“When they’re used to relight longer videos in chunks, the lighting can visibly jump between them.”
