ID-Forcing Enables Minute-Long Video Generation Without Fine-Tuning

ID-Forcing, a new test-time framework highlighted on HuggingFace Papers, fixes the drift (color, texture and motion decay) that arises when autoregressive video diffusion models generate beyond their training length. By aligning KV caching with conditioning instead of relying on flawed KV conditioning, it produces minute-long videos without fine-tuning.

2026-10-06 ~ 2026-10-07 · 2 related posts

• ID-Forcing extends short-horizon video diffusion models to minute-scale generation — SeoulNatlUniv · 2026-10-06
• ID-Forcing Keeps Long Video Generation In-Distribution, Enabling Minute-Scale Videos Without Fine-Tuning — _akhaliq · 2026-10-07