Research map / Image generation / Fast diffusion sampling
UniPC: A Unified Predictor-Corrector Framework for Fast Sampling of Diffusion Models
Wenliang Zhao, Lujia Bai, Yongming Rao, Jie Zhou, Jiwen Lu
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 UniPC instead of DPM-Solver++ | diffusion sampler · ImageNet 256x256 guided, NFE=10 | FID: 9.56 -> 7.51 | Improved |
| Add UniC corrector to DDIM (ablation) | DDIM · CIFAR10 NFE=5 | FID: 55.04 -> 47.22 | Improved |
| Add UniC corrector to DPM-Solver++(2M) (ablation) | DPM-Solver++(2M) · CIFAR10 NFE=5 | FID: 33.86 -> 31.23 | Improved |
| Add UniC corrector to DPM-Solver++(3S) (ablation) | DPM-Solver++(3S) · CIFAR10 NFE=5 | FID: 51.49 -> 50.62 | Improved |
| Add UniC corrector to DPM-Solver++(3M) (ablation) | DPM-Solver++(3M) · CIFAR10 NFE=5 | FID: 29.22 -> 25.50 | Improved |
| Use B(h)=h instead of B(h)=e^h-1 (ablation) | UniPC · CIFAR10 and FFHQ, NFE 5-6 | FID: 26.20 -> 23.22 (CIFAR10 NFE5) | Improved |
| Use B(h)=e^h-1 instead of B(h)=h (ablation) | UniPC · CIFAR10, NFE=10 | FID: 3.97 -> 3.87 | Improved |
| Use B(h)=h instead of e^h-1 in guided sampling (ablation) | UniPC · guided sampling | FID | Hurt |
| Recompute eps for corrected sample (UniC-oracle, 2x NFE) (ablation) | UniC · LSUN Bedroom 5 steps | FID: 13.79 -> 6.06 | Improved |
| Customize order schedule 123432 instead of default 123321 (ablation) | UniPC · CIFAR10 NFE=6 | FID: 10.33 -> 9.03 | Improved |
| Increase order at every step (123456) (ablation) | UniPC · CIFAR10 NFE=6 | FID: 10.33 -> 22.98 | Hurt |
| Use UniPC instead of DPM-Solver++ | diffusion sampler · CIFAR10 NFE=5 | Inception Score: 7.27 -> 7.55 | Improved |
Later papers that compared with this method
13 later accepted papers report UniPC as a baseline in their results tables; 13 of them report a better number on most of the results they share with it.
- BézierFlow: Learning Bézier Stochastic Interpolant Schedulers for Few-Step Generation (ICLR 2026 · better on 6 of 6 shared results)
- A-FloPS: Accelerating Diffusion Models via Adaptive Flow Path Sampler (AAAI 2026 · better on 6 of 6 shared results)
- Are First-Order Diffusion Samplers Really Slower? A Fast Forward-Value Approach (ICML 2026 · better on 3 of 5 shared results)
- STORK: Faster Diffusion and Flow Matching Sampling by Resolving both Stiffness and Structure-Dependence (ICLR 2026 · better on 3 of 3 shared results)
- Analyzing and Improving Training-Free Fast Sampling of Text-to-Image Diffusion Models (ECCV 2026 · better on 3 of 3 shared results)
- Image Diffusion Preview with Consistency Solver (CVPR 2026 · better on 1 of 1 shared result)
- Unleashing High-Quality Image Generation in Diffusion Sampling Using Second-Order Levenberg-Marquardt-Langevin (ICCV 2025 · better on 10 of 11 shared results)
- S4S: Solving for a Fast Diffusion Model Solver (ICML 2025 · better on 2 of 2 shared results)
5 more not listed.
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
- Faster Diffusion: Rethinking the Role of the Encoder for Diffusion Model Inference (NeurIPS 2024 · similarity 0.91)
- PFDiff: Training-Free Acceleration of Diffusion Models Combining Past and Future Scores (ICLR 2025 · similarity 0.90)
- A Simple Early Exiting Framework for Accelerated Sampling in Diffusion Models (ICML 2024 · similarity 0.90)
- Accelerating Diffusion Sampling with Optimized Time Steps (CVPR 2024 · similarity 0.90)
- UDPM: Upsampling Diffusion Probabilistic Models (NeurIPS 2024 · similarity 0.90)
- TADA: Improved Diffusion Sampling with Training-free Augmented DynAmics (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.
