Research map / Image generation / Fast diffusion sampling
C^2FG: Control Classifier-Free Guidance via Score Discrepancy Analysis
Jiayang Gao, Tianyi Zheng, Jiayang Zou, Fengxiang Yang, Shice Liu, Luyao Fan, Zheyu Zhang, Hao Zhang, Jinwei Chen, Peng-Tao Jiang, Bo Li, Jia Wang
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 |
|---|---|---|---|
| replace fixed CFG weight with exponentially decaying C2FG schedule | DiT-XL/2 · ImageNet 256x256 ODE | FID: 2.29 -> 2.07 | Improved |
| add C2FG to SiT-XL/2 (REPA) | SiT-XL/2 (REPA) · ImageNet 256x256 SDE | FID: 1.80 -> 1.51 | Improved |
| add C2FG to DiT-XL/2 at 512x512 | DiT-XL/2 · ImageNet 512x512 | FID: 6.81 -> 6.54 | Improved |
| add C2FG to DiT-XL/2 at 512x512 on recall | DiT-XL/2 · ImageNet 512x512 | Recall: 0.62 -> 0.60 | Hurt |
| add C2FG on top of interval guidance | SiT-XL/2 (REPA, Interval) · ImageNet 256x256 SDE | FID: 1.42 -> 1.41 | No clear effect |
| add C2FG on top of interval guidance with ODE sampler | SiT-XL/2 (REPA, Interval) · ImageNet 256x256 ODE | FID: 1.56 -> 1.54 | No clear effect |
| add C2FG to U-ViT | U-ViT · MS-COCO | FID: 5.37 -> 5.28 | Improved |
| add C2FG to Stable Diffusion 1.5 | SD1.5 · MS-COCO | CLIP score: 31.8 -> 31.9 | No clear effect |
| add C2FG to EDM2-S with autoguidance | EDM2-S-autog · ImageNet-64 | FID: 1.04 -> 1.03 | No clear effect |
| add C2FG at 20 sampling steps with ODE sampler | SiT-XL/2 (REPA) · ImageNet 256x256 | FID: 3.29 -> 3.10 | Improved |
| add C2FG at 20 sampling steps with SDE sampler | SiT-XL/2 (REPA) · ImageNet 256x256 | FID: 4.38 -> 4.30 | No clear effect |
Closest papers by meaning
- No Training, No Problem: Rethinking Classifier-Free Guidance for Diffusion Models (ICLR 2025 · similarity 0.93)
- Rectified Diffusion Guidance for Conditional Generation (CVPR 2025 · similarity 0.92)
- Feedback Guidance of Diffusion Models (NeurIPS 2025 · similarity 0.92)
- Diffusion Models without Classifier-free Guidance (NeurIPS 2026 · similarity 0.92)
- Eliminating Oversaturation and Artifacts of High Guidance Scales in Diffusion Models (ICLR 2025 · similarity 0.92)
- TFG: Unified Training-Free Guidance for Diffusion Models (NeurIPS 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.
