Research map / Image generation / Fast diffusion sampling
ScalingCache: Extreme Acceleration of DiTs through Difference Scaling and Dynamic Interval Caching
Lihui Gu, Jingbin He, Lianghao Su, Kang He, Wenxiao Wang, Yuliang Liu
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 differential scaling coefficient alpha (ablation) | ScalingCache · Wan2.1 1.3B | PSNR: 22.50 -> 26.61 | Improved |
| add differential scaling coefficient alpha (ablation) | ScalingCache · Flux 1.dev | PSNR: 31.04 -> 32.28 | Improved |
| add dynamic caching interval (ablation) | ScalingCache · Flux 1.dev | PSNR: 29.83 -> 32.28 | Improved |
| add dynamic caching interval (ablation) | ScalingCache · Wan2.1 1.3B | PSNR: 25.95 -> 26.61 | Improved |
| add dynamic caching interval (ablation) | ScalingCache · Flux 1.dev | speedup: 2.6x -> 3.0x | Mixed |
| use per-sample custom alpha instead of mean alpha (ablation) | ScalingCache · Wan2.1 1.3B | PSNR: 26.92 -> 27.76 | Improved |
| use mean alpha instead of no alpha (ablation) | ScalingCache · Wan2.1 1.3B | PSNR: 26.37 -> 26.92 | Improved |
| lower Sf from 12 to 6 (ablation) | ScalingCache · HunyuanVideo | VBench: 81.13 -> 79.67 | Hurt |
| lower Sf from 6 to 4 (ablation) | ScalingCache · HunyuanVideo | speedup: 4.5x -> 5.8x | Mixed |
| ScalingCache vs EasyCache | Wan2.1 1.3B · T2V | LPIPS: 0.095 -> 0.071 | Improved |
| ScalingCache vs TaylorSeer | FLUX 1.dev · T2I | LPIPS: 0.230 -> 0.131 | Improved |
| ScalingCache vs TaylorSeer | Wan2.2 5B i2v · PSNR | PSNR: 29.51 -> 33.29 | Improved |
Closest papers by meaning
- MagCache: Fast Video Generation with Magnitude-Aware Cache (NeurIPS 2025 · similarity 0.90)
- FasterCache: Training-Free Video Diffusion Model Acceleration with High Quality (ICLR 2025 · similarity 0.89)
- DeepCache: Accelerating Diffusion Models for Free (CVPR 2024 · similarity 0.89)
- CacheQuant: Comprehensively Accelerated Diffusion Models (CVPR 2025 · similarity 0.89)
- Towards Precise Scaling Laws for Video Diffusion Transformers (CVPR 2025 · similarity 0.89)
- Accelerating Diffusion Transformer via Increment-Calibrated Caching with Channel-Aware Singular Value Decomposition (CVPR 2025 · similarity 0.88)
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.
