Research map / Video / Video generation
Accelerating Diffusion-based Video Editing via Heterogeneous Caching: Beyond Full Computing at Sampled Denoising Timestep
Tianyi Liu, Ye Lu, Linfeng Zhang, Chen Cai, Jianjun Gao, Yi Wang, Kim-Hui Yap, Lap-Pui Chau
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 HetCache instead of TeaCache-fast | Wan2.1-VACE · VACE-Benchmark inpainting (Wan2.1-VACE) | speedup: 2.53× -> 2.67× | Improved |
| use HetCache instead of TeaCache-fast | Wan2.1-VACE · VACE-Benchmark inpainting (Wan2.1-VACE) | VFID: 54.86 -> 54.51 | Improved |
| use HetCache instead of TeaCache-fast | Wan2.1-VACE · VPBench text-guided editing (Wan2.1-VACE) | speedup: 1.79× -> 1.91× | Improved |
| add context clustering and correlation guidance to uniform context sampling (ablation) | HetCache · VACE-Benchmark inpainting (ablation) | VFID: 54.54 -> 54.51 | No clear effect |
| add context clustering and correlation guidance to uniform context sampling (ablation) | HetCache · VACE-Benchmark inpainting (ablation) | speedup: 3.13× -> 2.67× | Hurt |
| add K-Means context clustering on top of correlation-only selection (ablation) | HetCache · VACE-Benchmark inpainting (ablation) | VFID: 55.36 -> 54.51 | Improved |
| use HetCache instead of TeaCache at higher resolution | Wan-VACE-1.3B · Higher-resolution 25x720P inpainting | speedup: 2.40× -> 2.91× | Improved |
| use HetCache instead of TeaCache on longer video | Wan-VACE-1.3B · Longer 57x480P inpainting | speedup: 2.44× -> 3.06× | Improved |
| increase context ratio (ablation) | HetCache · VACE-Benchmark inpainting (ablation) | quality | Improved |
| vary K-Means cluster count K (ablation) | HetCache · VACE-Benchmark inpainting (ablation) | quality | No clear effect |
Closest papers by meaning
- FADE: Frequency-Aware Diffusion Model Factorization for Video Editing (CVPR 2025 · similarity 0.90)
- FasterCache: Training-Free Video Diffusion Model Acceleration with High Quality (ICLR 2025 · similarity 0.90)
- Timestep Embedding Tells: It's Time to Cache for Video Diffusion Model (CVPR 2025 · similarity 0.90)
- AsymRnR: Video Diffusion Transformers Acceleration with Asymmetric Reduction and Restoration (ICML 2025 · similarity 0.90)
- Ouroboros-Diffusion: Exploring Consistent Content Generation in Tuning-free Long Video Diffusion (AAAI 2025 · similarity 0.90)
- MagCache: Fast Video Generation with Magnitude-Aware Cache (NeurIPS 2025 · similarity 0.89)
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.
