arXiv paper presents Wave Forcing for speed-of-light video generation
Original title: Wave Forcing: Towards Speed-of-Light Streaming Video Generation
Wave Forcing co-designs generation training and execution for continuous video output achieving 1F1C with 147.6 DiT FPS on eight H200 GPUs.
The work introduces Wave Forcing which uses chunk-wise causal attention and staggered denoising to enable continuous long-video generation without quality degradation or camera drift. A three-stage training procedure introduces mixed-noise denoising motion supervision and causal adaptation. Wave Runtime handles early K/V delivery and communication overlap. Experiments on Wan2.1-T2V-1.3B show 5-step Wave Forcing maintains quality comparable to baselines while delivering higher throughput than Ulysses and Ring context-parallel methods.
- 1F1C execution after pipeline fill
- Three-stage training with mixed-noise and motion supervision
- Wave Parallelism for early K/V delivery
- 147.6 DiT FPS and 134.8 end-to-end FPS on eight H200 GPUs
- 7.07x single-GPU throughput
From the original article
1 min read at huggingface.co
Real-time long-video generation requires continuous output with consistent appearance and natural motion. Existing chunk-wise autoregressive methods face challenges in both generation efficiency and long-video quality: serialized denoising and history-cache refresh require two forward passes even with one-step denoising, and longer rollouts can degrade quality and cause camera drift.
By Yijie Jin, Yuxuan Liu, Jiajun Xu · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.