Research map / Image generation / Fast diffusion sampling
Rectified flow straightening
Straighten flow trajectories with reflow to speed up generation.
70 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2019–2026) · 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.
7 papers in 2023–24 and 61 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic grew, from 1.5% to 4.6%.
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
- 2019 · FloWaveNet : A Generative Flow for Raw Audio (ICML 2019 · 28 citations)
- 2020 · Normalizing Flows With Multi-Scale Autoregressive Priors (CVPR 2020 · 11 citations)
- 2023 · Flow Straight and Fast: Learning to Generate and Transfer Data with Rectified Flow (ICLR 2023 · 84 citations)
- 2024 · Improving the Training of Rectified Flows (NeurIPS 2024 · 6 citations)
- 2025 · Mean Flows for One-step Generative Modeling (NeurIPS 2025 · 26 citations)
- 2026 · Understanding, Accelerating, and Improving MeanFlow Training (CVPR 2026 · 1 citation)
Earliest papers here
- FloWaveNet : A Generative Flow for Raw Audio (ICML 2019 · 28 citations)
- Normalizing Flows With Multi-Scale Autoregressive Priors (CVPR 2020 · 11 citations)
- Flow Straight and Fast: Learning to Generate and Transfer Data with Rectified Flow (ICLR 2023 · 84 citations)
What has been tried
23 papers here report 245 measured changes (a component added, removed or swapped, compared with a baseline): helped 76%, no clear effect 5%, hurt 12%, mixed 7%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Replace the Flow Matching target with MeanFlow training that samples r≠t (25%) | ImageNet 256×256, 1-NFE · FID | 328.91 -> 61.06 | Helped | Mean Flows for One-step Generative Modeling (NeurIPS 2025 · 26 citations) |
| Add EDM initialization and increase batch size | CIFAR-10 · FID | 12.21 to 7.14 | Helped | Improving the Training of Rectified Flows (NeurIPS 2024 · 6 citations) |
| Train RFM S2 for 9 million iterations instead of one million | YFCC-4k · GeoScore | 2889 -> 3210 | Helped | Around the World in 80 Timesteps: A Generative Approach to Global Visual Geolocation (CVPR 2025 · 5 citations) |
| increase editing steps from 2 to 8 | Prompt-guided image editing · editing performance | Helped | FireFlow: Fast Inversion of Rectified Flow for Image Semantic Editing (ICML 2025 · 2 citations) | |
| add both VAE training stages instead of using neither | MEFB · PSNR | 59.41 -> 61.81 | Helped | Efficient Rectified Flow for Image Fusion (NeurIPS 2025 · 2 citations) |
| Add flow rectification and filtered self-generated data to QFlow | PDB, length 300, 500 steps · Fraction | 0.910_{±\text{0.029}} -> 0.969_{±\text{0.027}} | Helped | ReQFlow: Rectified Quaternion Flow for Efficient and High-Quality Protein Backbone Generation (ICML 2025 · 1 citation) |
| Use perturbed instead of standard rectification | OpenWebText · generative perplexity | 44.17 -> 24.13 | Helped | ReDi: Rectified Discrete Flow (NeurIPS 2025 · 1 citation) |
| use AYF-LMD instead of AYF-EMD | Image datasets · sample quality | Hurt | Align Your Flow: Scaling Continuous-Time Flow Map Distillation (NeurIPS 2025 · 1 citation) | |
| Add one velocity-refiner block | ImageNet 256×256 · FID↓ | 4.42 -> 2.80 | Helped | FlowTurbo: Towards Real-time Flow-Based Image Generation with Velocity Refiner (NeurIPS 2024 · 1 citation) |
| add progressive reflow | COCO-2017 validation · FID | 23.46 -> 22.97 | Helped | ProReflow: Progressive Reflow with Decomposed Velocity (CVPR 2025 · 0 citations) |
| add reflow to straighten trajectories | GEOM-Drugs, fewer than 5 ODE steps · COV-R and COV-P | without reflow, 0% coverage at 1 step | Helped | Efficient Molecular Conformer Generation with SO(3)-Averaged Flow Matching and Reflow (ICML 2025 · 0 citations) |
| Reduce KL regularization strength α from 0.5 to 0.0 | Synthetic manifold exploration · Monte Carlo entropy | 7.00 -> 7.14 | Helped | Flow Density Control: Generative Optimization Beyond Entropy-Regularized Fine-Tuning (NeurIPS 2025 · 0 citations) |
What reviewers will expect
Datasets and baselines named in the 25 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 · 6 of 25 papers
- CIFAR-10 · 4 of 25 papers
- CelebA-HQ · 2 of 25 papers
- AFHQv2 · 2 of 25 papers
- FFHQ · 2 of 25 papers
Baselines
- LCM · 4 of 25 papers
- PeRFlow · 3 of 25 papers
- DPMSolver · 3 of 25 papers
- SDXL-Lightning · 3 of 25 papers
- SDXL-Turbo · 3 of 25 papers
- 2-ReFlow · 2 of 25 papers
- LCM-LoRA · 2 of 25 papers
- Flash Diffusion · 2 of 25 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 68 papers; from OpenAlex and the papers' affiliations).
- Massachusetts Institute of Technology · 3 papers
- Korea Advanced Institute of Science and Technology · 3 papers
- University of Texas, Austin · 2 papers
- Google · 2 papers
- NVIDIA · 2 papers
- Tsinghua University · 2 papers
- Carnegie Mellon University · 2 papers
- ByteDance · 2 papers
- Technology · 2 papers
Use it
Download BibTeX of 24 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
- Flow matching distillation (95 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).