Metraipolis › archive › dc_bc6d07c189bb

UnStep: Training-Free Acceleration of Causal Video Diffusion with Fewer Steps Than Distillation — arXiv:2609.32518

Source: arXiv (en)
Original: https://arxiv.org/abs/2609.32518
License (as archived): CC0-1.0
Archived: 2026-10-07T23:50:58.205Z
SHA-256 of the archived text: 9e08c3fe18291ae54ee9274775822b2a3d299e6a8b7dbd1045da2cbf03eb83ad
Stable URL of this page: https://metraipolis.com/documento/dc_bc6d07c189bb

How to cite

UnStep: Training-Free Acceleration of Causal Video Diffusion with Fewer Steps Than Distillation — arXiv:2609.32518 — arXiv (en), https://arxiv.org/abs/2609.32518 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_bc6d07c189bb (Metraipolis archive dc_bc6d07c189bb, sha256 9e08c3fe18291ae54ee9274775822b2a3d299e6a8b7dbd1045da2cbf03eb83ad)

BibTeX

@misc{dcbc6d07c189bb,
  title = {UnStep: Training-Free Acceleration of Causal Video Diffusion with Fewer Steps Than Distillation — arXiv:2609.32518},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.32518},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_bc6d07c189bb; sha256 9e08c3fe18291ae54ee9274775822b2a3d299e6a8b7dbd1045da2cbf03eb83ad},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.32518. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Distilling bidirectional multi-step video diffusion transformers into few-step causal models has become a common approach for streaming video generation. While these few-step students are significantly faster than the teachers they are distilled from, they remain slow for real-time generation. In this work we present UnStep, a training-free wrapper that accelerates few-step causal video models at inference by running them with fewer diffusion transformer (DiT) steps than during distillation and limiting the temporal window retained in the attention KV cache. We propose two inference-only mechanisms to recover quality lost by step reduction and attention windowing: renoising the generated latent frames to a near-clean level and reusing the existing clean-cache pass to refine them, and applying truncated SVD to the DiT attention value and output projections. We also accelerate inference with a quality-preserving runtime stack for the DiT and VAE decoder, including more efficient attention calls and KV indexing, fused Triton RoPE with cached coefficients, and VAE decoding with optimized memory layout, precision, and convolution kernels. By reducing computation and optimizing the runtime stack, UnStep sets a new throughput regime for causal video diffusion, by running substantially faster than current methods, reaching 50 FPS on a single H100 without quality loss, and 77 FPS on GB200, all without retraining.

Tema degli esploratori: statistica e scienza dei dati (retro-catalogo arXiv: annunci del 2026-09-29 in cs.CV).
Categorie arXiv: Computer Vision and Pattern Recognition (cs.CV).
Pagina del lavoro: https://arxiv.org/abs/2609.32518 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-10-07T23:29:27.059Z dalla pagina https://arxiv.org/catchup/cs.CV/2026-09-29?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).