Research map / Image generation / Fast diffusion sampling
FACM: Flow-Anchored Consistency Models
Yansong Peng, Kai Zhu, Yu Liu, Pingyu Wu, Hebei Li, Xiaoyan Sun, Feng Wu
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 |
|---|---|---|---|
| add Flow-Anchoring (expanded time interval) to MeanFlow with 75% FM (ablation) | MeanFlow · ImageNet 256x256 FID@10 epochs | FID (NFE=1): 15.54 -> 4.31 | Improved |
| add interpolation (alpha(t)=1) (ablation) | FACM · ImageNet 256x256 FID@10 epochs | FID (NFE=1): 4.31 -> 3.42 | Improved |
| add residual clamping (ablation) | FACM · ImageNet 256x256 FID@10 epochs | FID (NFE=1): 3.42 -> 2.86 | Improved |
| add beta weighting (beta(t)=1) (ablation) | FACM · ImageNet 256x256 FID@10 epochs | FID (NFE=1): 2.86 -> 2.51 | Improved |
| add an FM anchor (0% to 75% FM) in MeanFlow (ablation) | MeanFlow · ImageNet 256x256 FID@10 epochs | collapse: 372.3-391.5 -> 43.03 | Improved |
| expanded time interval instead of auxiliary condition (ablation) | FACM · ImageNet 256x256 distilled | FID (NFE=1): 1.97 -> 1.81 | Improved |
| FACM instead of MeanFlow, distillation | LightningDiT · ImageNet 256x256 | FID (NFE=1): 2.75 -> 1.81 | Improved |
| FACM instead of MeanFlow, from scratch | LightningDiT · ImageNet 256x256 from scratch | FID (NFE=1): 2.65 -> 2.27 | Improved |
| set FM loss weight to zero or near zero (ablation) | FACM · ImageNet 256x256 | FID (NFE=1): 0.0-0.1 Collapse; 0.1-10.0 3.17-3.22 | Hurt |
| raise FM loss weight to 10.0-64.0 (ablation) | FACM · ImageNet 256x256 | FID (NFE=1): 3.17-3.22 -> 3.32-4.97 | Hurt |
| add cosine similarity term to FM loss (ablation) | FACM · ImageNet 256x256 | FID (NFE=1): 0.1-0.2 FID | Improved |
Closest papers by meaning
- Simplifying, Stabilizing and Scaling Continuous-time Consistency Models (ICLR 2025 · similarity 0.87)
- Truncated Consistency Models (ICLR 2025 · similarity 0.87)
- Improving Consistency Models with Generator-Augmented Flows (ICML 2025 · similarity 0.86)
- How to build a consistency model: Learning flow maps via self-distillation (NeurIPS 2025 · similarity 0.86)
- Blockwise Flow Matching: Improving Flow Matching Models For Efficient High-Quality Generation (NeurIPS 2025 · similarity 0.86)
- Improved Techniques for Training Consistency Models (ICLR 2024 · similarity 0.86)
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.
