Research map / Image generation / Fast diffusion sampling
SoFlow: Solution Flow Models for One-Step Generative Modeling
Tianze Luo, Haotian Yuan, Zhuang 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 Flow Matching loss data (75% ratio) (ablation) | SoFlow · ImageNet 256 DiT-B/4 1-NFE | FID-50K: 66.36 -> 58.57 | Improved |
| use loss weighting p=1.0 instead of p=0 (ablation) | SoFlow · ImageNet 256 DiT-B/4 1-NFE | FID-50K: 68.25 -> 58.57 | Improved |
| use exponential instead of cosine l-to-t schedule (ablation) | SoFlow · ImageNet 256 DiT-B/4 1-NFE | FID-50K: 60.17 -> 58.57 | No clear effect |
| use constant instead of exponential schedule (ablation) | SoFlow · ImageNet 256 DiT-B/4 1-NFE | FID-50K: 58.57 vs 59.52 | No clear effect |
| use loss weighting p=1.5 instead of 1.0 (ablation) | SoFlow · ImageNet 256 DiT-B/4 1-NFE | FID-50K: 58.57 -> 63.72 | Hurt |
| use linear schedule with Euler parameterization instead of trigonometric/trigonometric (ablation) | SoFlow · ImageNet 256 DiT-B/4 1-NFE with CFG | FID-50K: 19.01 -> 11.59 | Improved |
| increase CFG strength from 1.5 to 3.0 (ablation) | SoFlow · ImageNet 256 DiT-B/4 1-NFE with CFG | FID-50K: 34.47 -> 11.59 | Improved |
| lower velocity mix ratio m from 1.0 to 0.25 (ablation) | SoFlow · ImageNet 256 DiT-B/4 1-NFE with CFG | FID-50K: 17.22 -> 11.59 | Improved |
| replace MeanFlow loss with solution consistency loss (efficient attention) | SoFlow · ImageNet 256 training cost | GPU memory: 51.45 GB -> 35.44 GB | Improved |
| replace MeanFlow loss with solution consistency loss (efficient attention) | SoFlow · ImageNet 256 training cost | training speed: 2.39 -> 2.94 iters/sec | Improved |
| use SoFlow instead of MeanFlow | XL/2 · ImageNet 256 1-NFE | FID-50K: 3.43 -> 2.96 | Improved |
| use SoFlow instead of MeanFlow | XL/2 · ImageNet 256 2-NFE | FID-50K: 2.93 -> 2.66 | Improved |
Later papers that compared with this method
1 later accepted paper reports SoFlow-XL as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Temporal Equilibrium MeanFlow: Bridging the Scale Gap for One-Step Generation (CVPR 2026 · better on 2 of 2 shared results)
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
- Mean Flows for One-step Generative Modeling (NeurIPS 2025 · similarity 0.91)
- Self-Corrected Flow Distillation for Consistent One-Step and Few-Step Image Generation (AAAI 2025 · similarity 0.89)
- Bespoke Solvers for Generative Flow Models (ICLR 2024 · similarity 0.89)
- D-Flow: Differentiating through Flows for Controlled Generation (ICML 2024 · similarity 0.88)
- Shortcutting Pre-trained Flow Matching Diffusion Models is Almost Free Lunch (NeurIPS 2025 · similarity 0.88)
- Diff2Flow: Training Flow Matching Models via Diffusion Model Alignment (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.
