In-Distribution Forcing for Long Video Generation at Test Time
ID-Forcing keeps KV caches in-distribution so short-horizon video models generate minute-scale clips without drift.
Autoregressive video diffusion models drift on long rollouts because cached key-value entries leave the training distribution, a failure the authors call the KV-provenance problem. In-Distribution Forcing is a test-time method whose self-caching stores each chunk without attending to prior KV entries so the rolling window stays in-distribution. The method remains competitive on standard video benchmarks and substantially outperforms prior KV-conditioning work on drift metrics and a user study, extending short-horizon models to minute-scale video.
- Cached KV entries drift out of distribution beyond the training horizon
- Self-caching stores each chunk without attending to prior KV entries
- Extends short-horizon autoregressive video models to minute-scale clips
- Outperforms prior KV conditioning on drift metrics and a user study
Full article184 words · extracted from huggingface.co · click to collapse
Modern autoregressive (AR) video diffusion models excel at short-horizon video generation, yet generating long videos remains challenging due to drifting, where colors and textures shift, and motion dynamics decay. Existing works primarily rely on KV conditioning, which selects or modifies cached key-value (KV) entries to mitigate drifting. However, we observe that KV conditioning alone is insufficient as it assumes cached KV entries remain in-distribution. This assumption fails beyond the training horizon: nothing constrains the construction of KV entries during rollout, giving rise to the KV-provenance problem where cached entries themselves become out-of-distribution (OOD). To address this, we propose In-Distribution Forcing (ID-Forcing), a test-time framework that aligns both KV caching and KV conditioning with training configurations. Its key mechanism, self-caching, prevents OOD KV entries at their source. Each chunk is cached without attending to prior KV entry, keeping the rolling window exactly in-distribution. Consequently, ID-Forcing seamlessly extends short-horizon models to minute-scale video generation. Extensive evaluations show that our method remains competitive on standard video generation benchmark while substantially outperforming prior work in mitigating drifting, as validated by both our drift metrics and a user study.
Text extracted automatically; images, tables and formatting may be missing. Original: https://huggingface.co/papers/2610.03120