Research map / Image generation / Fast diffusion sampling
Consistency Models
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Distill an EDM diffusion model into a consistency model instead of using progressive distillation | EDM · CIFAR-10, 1-step | FID: 8.34 -> 3.55 | Helped |
| Distill an EDM diffusion model into a consistency model instead of using progressive distillation | EDM · ImageNet 64×64, 1-step | FID: 15.39 -> 6.20 | Helped |
| Distill an EDM diffusion model into a consistency model instead of using progressive distillation | EDM · LSUN Bedroom 256×256, 1-step | FID: 16.92 -> 7.80 | Helped |
| Distill an EDM diffusion model into a consistency model instead of using progressive distillation | EDM · LSUN Cat 256×256, 1-step | FID: 29.6 -> 11.0 | Helped |
| Use 2-step rather than 1-step consistency sampling | CD · CIFAR-10 | FID: 3.55 -> 2.93 | Helped |
| Use 2-step rather than 1-step consistency sampling | CT · ImageNet 64×64 | FID: 13.0 -> 11.1 | Helped |
| Train consistency distillation with LPIPS instead of squared L2 or L1 (ablation) | CD · CIFAR-10 | FID | Helped |
| Use Heun's second-order solver instead of Euler's first-order solver (ablation) | CD · CIFAR-10 | FID | Helped |
| Use N=18 discretization steps in consistency distillation (ablation) | CD · CIFAR-10 | FID | Helped |
| Increase N during training rather than keeping it fixed (ablation) | CT · CIFAR-10 | convergence speed | Helped |
| Adapt the EMA decay μ during training rather than keeping it fixed (ablation) | CT · CIFAR-10 | convergence speed | Helped |
| Use LPIPS rather than squared L2 for progressive distillation (ablation) | PD · CIFAR-10, few-step generation | FID | Helped |
Closest papers by meaning
- Improved Techniques for Training Consistency Models (ICLR 2024 · similarity 0.93)
- Theory of Consistency Diffusion Models: Distribution Estimation Meets Fast Sampling (ICML 2024 · similarity 0.93)
- Simplifying, Stabilizing and Scaling Continuous-time Consistency Models (ICLR 2025 · similarity 0.91)
- Truncated Consistency Models (ICLR 2025 · similarity 0.91)
- ACT-Diffusion: Efficient Adversarial Consistency Training for One-step Diffusion Models (CVPR 2024 · similarity 0.91)
- Sampling is as easy as keeping the consistency: convergence guarantee for Consistency Models (ICML 2024 · similarity 0.91)
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.