Research map / Image generation / Fast diffusion sampling
DiCache: Let Diffusion Model Determine Its Own Cache
Jiazi Bu, Pengyang Ling, Yujie Zhou, Yibin Wang, Yuhang Zang, Dahua Lin, Jiaqi Wang
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 |
|---|---|---|---|
| add Dynamic Cache Trajectory Alignment (DCTA) (ablation) | DiCache · HunyuanVideo | PSNR: 31.98 -> 32.79 | Improved |
| add DCTA (SSIM/LPIPS) (ablation) | DiCache · HunyuanVideo | SSIM: 0.9314 -> 0.9396 | Improved |
| increase probe depth from m=1 to m=5 (ablation) | DiCache online probe · HunyuanVideo | LPIPS and speedup: LPIPS 0.1492 -> 0.1367; speedup 2.34x -> 2.10x | Mixed |
| lower reuse threshold from 0.10 to 0.05 (ablation) | DiCache · HunyuanVideo | LPIPS and speedup: LPIPS 0.1492 -> 0.1047; speedup 2.34x -> 1.76x | Mixed |
| raise reuse threshold from 0.10 to 0.20 (ablation) | DiCache · HunyuanVideo | LPIPS and speedup: LPIPS 0.1492 -> 0.1886; speedup 2.34x -> 2.90x | Mixed |
| use shallow-layer probe (online) instead of input differences as caching error indicator | DiCache · Wan/Hunyuan/Flux feature analysis | Spearman correlation with output differences: around 0.8 at m in [1,3] | Improved |
| combine DiCache with Sparse VideoGen | HunyuanVideo · HunyuanVideo 720x1280, 129 frames | speedup: 1.67x -> 3.08x | Improved |
| combine DiCache with Sparse VideoGen | HunyuanVideo · HunyuanVideo 720x1280, 129 frames | LPIPS: 0.1760 -> 0.2051 | Hurt |
| use offline-prior or empirical-law caching (TeaCache-fast) instead of online probe | TeaCache · HunyuanVideo | LPIPS: 0.1492 -> 0.2898 | Hurt |
Later papers that compared with this method
3 later accepted papers report DiCache as a baseline in their results tables; 3 of them report a better number on most of the results they share with it.
- DIFFSPARSE: ACCELERATING DIFFUSION TRANSFORMERS WITH LEARNED TOKEN SPARSITY (ICLR 2026 · better on 1 of 1 shared result)
- Evolutionary Caching to Accelerate Your Off-the-Shelf Diffusion Model (ICLR 2026 · better on 1 of 1 shared result)
- Accelerating Diffusion Models via Equal-Risk Caching (ECCV 2026 · better on 1 of 1 shared result)
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
- Accelerating Diffusion Models via Equal-Risk Caching (ECCV 2026 · similarity 0.91)
- ERTACache: Error Rectification and Timesteps Adjustment for Efficient Diffusion (ICLR 2026 · similarity 0.91)
- Cache Me if You Can: Accelerating Diffusion Models through Block Caching (CVPR 2024 · similarity 0.90)
- Learning-to-Cache: Accelerating Diffusion Transformer via Layer Caching (NeurIPS 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.
