Research map / Image generation / Fast diffusion sampling
MIST: Moment-Aligned Invariant Stability Transform for Robust Flow Matching
Liang Peng, Deqing Li, Yujia Wu, Hao Meng, Kuan Cao, Yu Wu, Xiaoxiao Xu, Lin Qu
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 Invariant Alignment (drift correction + energy matching) to CFG (ablation) | CFG on SD3.5 medium · HPD v2 (G.S. 10) | Aesthetic score: 5.866 -> 6.032 | Improved |
| add Stability Thresholding to CFG (ablation) | CFG on SD3.5 medium · HPD v2 (G.S. 10) | CLIP score: 36.57 -> 37.22 | Improved |
| combine IA and ST (ablation) | CFG on SD3.5 medium · HPD v2 (G.S. 10) | PickScore: 22.44 -> 23.02 | Improved |
| add energy matching on top of drift correction (ablation) | IA · HPD v2 (G.S. 10) | PickScore: 22.58 -> 22.92 | Improved |
| add temporal decay on top of spatial suppression (ablation) | ST · HPD v2 (G.S. 10) | PickScore: 22.71 -> 22.72 | Improved |
| apply TD from the first step (Tclip=0 instead of 1) (ablation) | MIST · HPD v2 (G.S. 10) | Aesthetic / CLIP: 6.056 / 37.32 vs 6.053 / 37.33 | No clear effect |
| raise spatial suppression gamma from 1.5 to 3.0 (ablation) | MIST · HPD v2 (G.S. 10) | ImageReward: 1.2216 -> 1.1450 | Mixed |
| use MIST instead of CFG at high guidance (G.S. 15) | SD3.5 medium · HPD v2 | PickScore: 21.43 -> 22.98 | Improved |
| use MIST instead of CFG at high guidance (G.S. 15) | Flux-dev · HPD v2 | PickScore: 21.36 -> 22.98 | Improved |
| use MIST instead of CFG at low guidance (G.S. 5) | SD3 medium · HPD v2 | CLIP: 36.41 -> 36.34 | No clear effect |
| use MIST instead of CFG on text-to-video | Wan2.2 5B · VBench | total score: 81.64 -> 81.92 | Improved |
| increase sampling steps from 28 to 50 | MIST vs CFG on SD3.5 · HPD v2 | PickScore: 22.44 -> 22.73 (CFG); 23.02 with MIST at both | Improved |
Closest papers by meaning
- Stable Velocity: A Variance Perspective on Flow Matching (ICML 2026 · similarity 0.92)
- On the Guidance of Flow Matching (ICML 2025 · similarity 0.88)
- Value Gradient Guidance for Flow Matching Alignment (NeurIPS 2025 · similarity 0.87)
- Direct Product Flow Matching: Decoupling Radial and Angular Dynamics for Few-Shot Adaptation (NeurIPS 2026 · similarity 0.87)
- Rectified CFG++ for Flow Based Models (NeurIPS 2025 · similarity 0.87)
- Coreset-Induced Conditional Velocity Flow Matching (NeurIPS 2026 · similarity 0.86)
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.
