Research map / Video / Video generation
BWCache: Accelerating Video Diffusion Transformers through Block-Wise Caching
Hanshuai Cui, Zhiqing Tang, Zhifei Xu, Zhi Yao, Wenyi Zeng, Weijia Jia
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use BWCache instead of TeaCache | TeaCache · Open-Sora | speedup: 1.50x -> 1.61x | Improved |
| use BWCache instead of TeaCache | TeaCache · Open-Sora-Plan | speedup: 4.41x -> 2.24x | Hurt |
| use BWCache instead of TeaCache | TeaCache · Open-Sora | LPIPS: 0.1496 -> 0.0879 | Improved |
| apply BWCache to model | Open-Sora · VBench | VBench: 80.33% -> 80.03% | No clear effect |
| apply BWCache to model | Latte · VBench | VBench: 78.95% -> 78.28% | Hurt |
| apply BWCache instead of Skip-DiT | Skip-DiT · Latte | VBench: 75.76% -> 78.28% | Improved |
| raise indicator threshold from 0.15 to 0.25 (ablation) | BWCache · Open-Sora | VBench: 80.03% -> 79.03% | Hurt |
| lengthen reuse interval from 5% to 25% (ablation) | BWCache · Open-Sora | VBench: 80.28% -> 79.31% | Hurt |
| lengthen reuse interval from 5% to 25% (ablation) | BWCache · Open-Sora | latency (s): 34.70 -> 25.76 | Improved |
| use caching instead of reducing steps to 19 (ablation) | Open-Sora · Open-Sora | LPIPS: 0.3139 -> 0.0879 | Improved |
| scale to 8 GPUs with DSP | BWCache · Open-Sora (204 frames) | speedup: 2.4x -> 17.2x | Improved |
| use BWCache instead of TeaCache with 8 GPUs | TeaCache · Open-Sora-Plan (221 frames) | speedup: 28.1x -> 13.3x | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- FasterCache — FasterCache: Training-Free Video Diffusion Model Acceleration with High Quality (ICLR 2025) · better on 3 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Model Reveals What to Cache: Profiling-Based Feature Reuse for Video Diffusion Models (ICCV 2025 · similarity 0.93)
- FasterCache: Training-Free Video Diffusion Model Acceleration with High Quality (ICLR 2025 · similarity 0.92)
- Accelerating Diffusion Transformers with Token-wise Feature Caching (ICLR 2025 · similarity 0.92)
- AsymRnR: Video Diffusion Transformers Acceleration with Asymmetric Reduction and Restoration (ICML 2025 · similarity 0.91)
- Cache Me if You Can: Accelerating Diffusion Models through Block Caching (CVPR 2024 · similarity 0.91)
- HarmoniCa: Harmonizing Training and Inference for Better Feature Caching in Diffusion Transformer Acceleration (ICML 2025 · similarity 0.90)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
