Research map / Image generation / Fast diffusion sampling
Flow matching distillation
Distill and align flow matching models into few-step generators.
95 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2022–2026) · an approach in Fast diffusion sampling, keyword cluster “flow · matching · rectified”
State History Tried Checklist Preprints Who Export
Where it stands
Not enough benchmark data Fewer than four papers report the same benchmark, so there is no record to follow.
9 papers in 2023–24 and 85 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic grew, from 2.0% to 6.4%.
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2022 · Maximum Likelihood Training of Implicit Nonlinear Diffusion Model (NeurIPS 2022 · 1 citation)
- 2023 · The Surprising Effectiveness of Diffusion Models for Optical Flow and Monocular Depth Estimation (NeurIPS 2023 · 20 citations)
- 2024 · FlowDiffuser: Advancing Optical Flow Estimation with Diffusion Models (CVPR 2024 · 42 citations)
- 2025 · DepthFM: Fast Generative Monocular Depth Estimation with Flow Matching (AAAI 2025 · 37 citations)
- 2026 · One-step Latent-free Image Generation with Pixel Mean Flows (ICML 2026 · 1 citation)
Earliest papers here
- Maximum Likelihood Training of Implicit Nonlinear Diffusion Model (NeurIPS 2022 · 1 citation)
- The Surprising Effectiveness of Diffusion Models for Optical Flow and Monocular Depth Estimation (NeurIPS 2023 · 20 citations)
- FlowDiffuser: Advancing Optical Flow Estimation with Diffusion Models (CVPR 2024 · 42 citations)
What has been tried
41 papers here report 486 measured changes (a component added, removed or swapped, compared with a baseline): helped 71%, no clear effect 3%, hurt 22%, mixed 3%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Add Hidden State Denoising | Sintel train clean · EPE | 0.93 -> 0.86 | Helped | FlowDiffuser: Advancing Optical Flow Estimation with Diffusion Models (CVPR 2024 · 42 citations) |
| use image-to-depth transport rather than noise-to-depth transport | NYUv2 · δ1 | 92.4 -> 94.6 | Helped | DepthFM: Fast Generative Monocular Depth Estimation with Flow Matching (AAAI 2025 · 37 citations) |
| Add NER to NewImp | CC MAR 30% missing · MAE | 0.61 -> 0.35 | Helped | Rethinking the Diffusion Models for Missing Data Imputation: A Gradient Flow Perspective (NeurIPS 2024 · 17 citations) |
| add adversarial GAN loss | CIFAR-10 image generation · sample quality | Helped | Consistency Trajectory Models: Learning Probability Flow ODE Trajectory of Diffusion (ICLR 2024 · 16 citations) | |
| Use sin-cos matching instead of linear flow matching | COCO validation · FID-30K | 9.63 -> 8.95 | Helped | Flowing from Words to Pixels: A Noise-Free Framework for Cross-Modality Evolution (CVPR 2025 · 11 citations) |
| distill the sampling process into a one-step student | Promoter DNA sequence design · MSE | .0269 -> .0278 | Hurt | Dirichlet Flow Matching with Applications to DNA Sequence Design (ICML 2024 · 10 citations) |
| Limit trainable parameters with LoRA rather than full fine-tuning | NYUv2 · AbsRel | 5.7 -> 5.9 | Hurt | Diff2Flow: Training Flow Matching Models via Diffusion Model Alignment (CVPR 2025 · 7 citations) |
| Add GAN loss to consistency distillation | CelebA-HQ 256, 1 NFE · FID | 41.34 -> 11.55 | Helped | Self-Corrected Flow Distillation for Consistent One-Step and Few-Step Image Generation (AAAI 2025 · 2 citations) |
| use hierarchical optical-flow diffusion instead of latent-space denoising | SNU-FILM hard · FID | 22.283 -> 15.320 | Helped | Hierarchical Flow Diffusion for Efficient Frame Interpolation (CVPR 2025 · 2 citations) |
| Add test-time ensembling | FlyingChairs validation · EPE | 3.45 -> 3.08 | Helped | Scaling Properties of Diffusion Models For Perceptual Tasks (CVPR 2025 · 2 citations) |
| Add Semantic Feature Guidance | ImageNet 256×256 · FID | 81.5 -> 66.9 | Helped | Blockwise Flow Matching: Improving Flow Matching Models For Efficient High-Quality Generation (NeurIPS 2025 · 2 citations) |
| use t-EDM with ν=5 instead of ν=3 | HRRR VIL (Test) · KS | 0.114 -> 0.888 | Hurt | Heavy-Tailed Diffusion Models (ICLR 2025 · 2 citations) |
What reviewers will expect
Datasets and baselines named in the 46 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- ImageNet · 8 of 46 papers
- CIFAR-10 · 7 of 46 papers
- ImageNet-1K · 5 of 46 papers
- FFHQ · 5 of 46 papers
- NYUv2 · 4 of 46 papers
- KITTI · 3 of 46 papers
- ETH3D · 3 of 46 papers
- ScanNet · 3 of 46 papers
Baselines
- CTM · 4 of 46 papers
- EDM · 4 of 46 papers
- Marigold · 3 of 46 papers
- DPM Solver · 3 of 46 papers
- iCT · 3 of 46 papers
- ADM · 3 of 46 papers
- DiT-XL/2 · 3 of 46 papers
- SiT-XL/2 · 3 of 46 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 94 papers; from OpenAlex and the papers' affiliations).
- Meta · 4 papers
- Tsinghua University · 4 papers
- Stanford University · 3 papers
- Weizmann Institute of Science · 3 papers
- Apple · 3 papers
- University of Science and Technology of China · 3 papers
- CompVis @ LMU Munich · 2 papers
- Chinese Academy of Sciences · 2 papers
- National University of Singapore · 2 papers
- The Chinese University of Hong Kong · 2 papers
Use it
Download BibTeX of 22 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
- Rectified flow straightening (70 papers)
- Guidance and alignment for flow matching (34 papers)
Elsewhere in the topic
- Training-free diffusion acceleration (257 papers)
- Diffusion theory and design space (252 papers)
- Diffusion models as perception tools (198 papers)
- Masked diffusion language model decoding (147 papers)
- Guidance for conditional diffusion (123 papers)
- Diffusion theory and learned samplers (121 papers)
- Fast diffusion ODE solvers (120 papers)
- Distributional control of diffusion (109 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).