Research map / Image generation / Fast diffusion sampling
Momentum Guidance: Plug-and-Play Guidance for Flow Models
Runlong Liao, Jian Yu, Baiyu Su, Chi Zhang, Lizhang Chen, Qiang 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 Momentum Guidance (no CFG) | DiT-XL rectified flow · ImageNet-256 (NFE 16) | FID-50K: 7.76 -> 4.46 | Improved |
| add Momentum Guidance on top of CFG 1.2 | DiT-XL rectified flow · ImageNet-256 (NFE 16) | FID-50K: 3.26 -> 2.00 | Improved |
| add MG on top of CFG 1.6 with 64 NFEs | DiT-XL rectified flow · ImageNet-256 | FID-50K: 3.34 -> 3.17 | No clear effect |
| add MG to CFG interval (w=1.4, 16 NFE) | DiT-XL rectified flow · ImageNet-256 | FID-50K: 2.352 -> 1.553 | Improved |
| add MG to CFG interval (w=1.6, 64 NFE) | DiT-XL rectified flow · ImageNet-256 | FID-50K: 1.612 -> 1.618 | No clear effect |
| add MG to CFG at strong scales | DiT-XL rectified flow · ImageNet-256 | Precision: 0.90 -> 0.87 (CFG 2.0, NFE 16); recall 0.45 -> 0.51 | Mixed |
| add MG to RAE | RAE · ImageNet-256 (w=1.0) | FID-50K: 1.535 -> 1.376 | Improved |
| add MG to SD3 at CFG 1 | Stable Diffusion 3 · HPS prompts | HPSv2.1: 22.87 -> 27.37 | Improved |
| add MG to SD3 at CFG 7 | Stable Diffusion 3 · HPS prompts | HPSv2.1: 30.41 -> 30.56 | Improved |
| add MG to FLUX.1-dev at CFG 3.5-4 | FLUX.1-dev · HPS prompts | ImageReward: 1.117 -> 1.115 (CFG 3.5) | Hurt |
| add MG to HunyuanVideo | HunyuanVideo · VBench | total score: 0.816 -> 0.819 | Improved |
| add MG to HunyuanVideo | HunyuanVideo · VBench scene dimension | Scene score: 0.387 -> 0.356 | Hurt |
Closest papers by meaning
- FlowTurbo: Accelerating Flow-Based Image Generation Models via Multi-Stage Refinement (TPAMI 2026 · similarity 0.91)
- FlowTurbo: Towards Real-time Flow-Based Image Generation with Velocity Refiner (NeurIPS 2024 · similarity 0.90)
- On the Guidance of Flow Matching (ICML 2025 · similarity 0.88)
- Accelerating Rectified Flow Models via Trajectory-Aware Caching (NeurIPS 2026 · similarity 0.88)
- Eliminating Oversaturation and Artifacts of High Guidance Scales in Diffusion Models (ICLR 2025 · similarity 0.88)
- Devil is in the Details: Density Guidance for Detail-Aware Generation with Flow Models (ICML 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.
