Research map / Image generation / Fast diffusion sampling
SeaCache: Spectral-Evolution-Aware Cache for Accelerating Diffusion Models
Jiwoo Chung, Sangeek Hyun, MinKyu Lee, Byeongju Han, Geonho Cha, Dongyoon Wee, Youngjun Hong, Jae-Pil Heo
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 |
|---|---|---|---|
| filter features with SEA filter before computing cache distance (vs TeaCache raw distance, d=0.3) | TeaCache · FLUX.1-dev | PSNR: 20.762 -> 26.285 | Improved |
| filter features with SEA filter (vs TeaCache, d=0.12) | TeaCache · HunyuanVideo | PSNR: 23.40 -> 32.39 | Improved |
| filter features with SEA filter (vs TeaCache, d=0.09 vs 0.2) | TeaCache · Wan2.1 1.3B | PSNR: 20.84 -> 26.60 | Improved |
| filter features with SEA filter at aggressive budget (vs TeaCache d=0.6 vs 0.6) | TeaCache · FLUX.1-dev | PSNR: 17.214 -> 21.332 | Improved |
| use complementary 1-SEA filter instead of SEA filter (ablation) | SeaCache · FLUX and HunyuanVideo | PSNR vs refresh ratio | Improved |
| add gain normalization to the SEA filter (ablation) | SeaCache · FLUX and HunyuanVideo | PSNR vs refresh ratio | Improved |
| use timestep-dependent SEA filter instead of static low-pass filter (LPF 30%) (ablation) | SeaCache · FLUX and HunyuanVideo | PSNR vs refresh ratio | Improved |
| apply SEA filter to DiCache block features | DiCache · FLUX | PSNR at same refresh ratio | Improved |
| use SeaCache instead of TeaCache/TaylorSeer on CycleReward | FLUX.1-dev 50% budget · DrawBench | average rank (lower better): 2.01 -> 1.91 | Improved |
| use SeaCache instead of MagCache at matched 50% refresh ratio | MagCache · Wan2.1 1.3B T2V | PSNR: 25.55 -> 29.55 | Improved |
| use SeaCache instead of MagCache at matched 52% refresh ratio | MagCache · FLUX.1-dev | PSNR: 29.96 -> 30.37 | Improved |
| use SeaCache instead of TeaCache in VBench average rank (30% budget) | TeaCache · Wan2.1 1.3B | average rank (lower better): 1.53 -> 2.13 | Hurt |
Closest papers by meaning
- DeepCache: Accelerating Diffusion Models for Free (CVPR 2024 · similarity 0.92)
- CacheQuant: Comprehensively Accelerated Diffusion Models (CVPR 2025 · similarity 0.91)
- Cache Me if You Can: Accelerating Diffusion Models through Block Caching (CVPR 2024 · similarity 0.90)
- Timestep Embedding Tells: It's Time to Cache for Video Diffusion Model (CVPR 2025 · similarity 0.90)
- Spectral Progressive Diffusion for Efficient Image and Video Generation (NeurIPS 2026 · similarity 0.90)
- FasterCache: Training-Free Video Diffusion Model Acceleration with High Quality (ICLR 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.
