Self Gradient Forcing: Native Long Video Extrapolation

Research official 2 src. ~1 min

Self Gradient Forcing (SGF) is a two-pass training strategy for autoregressive video diffusion that restores missing supervision in the context-gradient path, enabling extrapolation to multi-minute videos from only 5-second training windows with improved identity preservation and temporal stability.

Why it matters

Ranked #2 trending on HuggingFace Daily Papers (>=100 upvotes); addresses a fundamental limitation in long-video generation.

Importance: 4/5

#2 trending HF Daily Paper with >=100 upvotes.

Sources