Research map / Image generation / Fast diffusion sampling
FMBoost: Boosting Latent Diffusion with Flow Matching
Johannes S Fischer, Ming Gui, Pingchuan Ma, Nick Stracke, Stefan Andreas Baumann, Vincent Tao Hu, Björn Ommer
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 |
|---|---|---|---|
| use coupling flow matching instead of one-step L1/L2 regression for latent upsampling (ablation) | latent upsampler · FacesHQ | FID: 4.52 -> 1.36 | Improved |
| use coupling flow matching instead of a diffusion upsampler (ablation) | latent upsampler · FacesHQ | FID: 2.72 -> 1.36 | Improved |
| use data-dependent coupling (start from low-res latent) instead of naive flow matching from noise (ablation) | flow matching upsampler · FacesHQ | p-FID: 2.10 -> 1.62 | Improved |
| use data-dependent coupling instead of naive flow matching (ablation) | flow matching upsampler · FacesHQ | FID: 1.37 -> 1.36 | No clear effect |
| use CFM instead of diffusion upsampler at 4 steps (256 to 1024) | SD1.5 upsampling · Zero-shot LAION-5k | p-FID: 50.53 -> 24.02 | Improved |
| use CFM instead of diffusion upsampler at 40 steps (512 to 1024) | SD1.5 upsampling · Zero-shot LAION-5k | FID: 22.29 -> 21.67 | Improved |
| add Gaussian noise augmentation to the low-res latent (ablation) | CFM · FID and p-FID, 256 to 1024 | FID | Improved |
| upsample low-res latent in pixel space instead of nearest or bilinear in latent space (ablation) | CFM · convergence | convergence speed and quality | Improved |
| boost SD1.5 with CFM instead of using LCM-LoRA SDXL for 1024 images | high-resolution synthesis · Zero-shot LAION-5k 1024x1024 | FID: 28.98 -> 21.67 | Improved |
Closest papers by meaning
- Simpler Diffusion: 1.5 FID on ImageNet512 with Pixel-space Diffusion (CVPR 2025 · similarity 0.90)
- One Diffusion Step to Real-World Super-Resolution via Flow Trajectory Distillation (ICML 2025 · similarity 0.90)
- Cross Flow: One-Step Generation Across Latent and Pixel Spaces (NeurIPS 2026 · similarity 0.90)
- Improving Progressive Generation with Decomposable Flow Matching (NeurIPS 2025 · similarity 0.89)
- Diff2Flow: Training Flow Matching Models via Diffusion Model Alignment (CVPR 2025 · similarity 0.89)
- Shortcutting Pre-trained Flow Matching Diffusion Models is Almost Free Lunch (NeurIPS 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.
