Research map / Video / Video generation
D2Cache: Second-Order Delta Caching for Higher Video Diffusion Acceleration
Enhuai Liu, Yunke Wang, Changming Sun, Chang Xu
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 |
|---|---|---|---|
| scale second-order deltas by timestep-embedding-derived factor s (ablation) | D2Cache · Latte superfast | VBench: 74.31% -> 76.03% | Improved |
| add second-order delta correction on top of TeaCache schedule (ablation) | TeaCache-superfast · Latte | VBench: 75.61% -> 76.03% | Improved |
| add second-order delta correction on top of TeaCache schedule | TeaCache-superfast · LTX-video | VBench: 70.92% -> 73.42% | Improved |
| add second-order delta correction on top of TeaCache schedule | TeaCache-superfast · Open-Sora 1.2 | VBench: 77.07% -> 77.47% | Improved |
| add second-order delta correction on top of TeaCache schedule | TeaCache-superfast · Wan2.1 | VBench: 79.70% -> 80.12% | Improved |
| add second-order delta correction in the slow (low acceleration) setting | TeaCache-slow · Latte | VBench: 77.12% -> 77.11% | No clear effect |
| add second-order delta correction in the slow setting | TeaCache-slow · Open-Sora 1.2 | VBench: 78.96% -> 78.95% | No clear effect |
| add second-order delta correction | TeaCache · all four models | latency: under 0.3s overhead | No clear effect |
| add second-order delta correction on complex prompts | TeaCache-superfast · Wan2.1 complex prompts | Laplacian variance: 621.21 -> 746.39 | Improved |
| add second-order delta correction on complex prompts | TeaCache-superfast · Wan2.1 complex prompts | Tenengrad: 332.58 -> 476.62 | Improved |
Closest papers by meaning
- Timestep Embedding Tells: It's Time to Cache for Video Diffusion Model (CVPR 2025 · similarity 0.92)
- MagCache: Fast Video Generation with Magnitude-Aware Cache (NeurIPS 2025 · similarity 0.91)
- FasterCache: Training-Free Video Diffusion Model Acceleration with High Quality (ICLR 2025 · similarity 0.91)
- Cache Me if You Can: Accelerating Diffusion Models through Block Caching (CVPR 2024 · similarity 0.90)
- CacheQuant: Comprehensively Accelerated Diffusion Models (CVPR 2025 · similarity 0.89)
- DeepCache: Accelerating Diffusion Models for Free (CVPR 2024 · 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.
