Research map / Image generation / Fast diffusion sampling
CFG-Ctrl: Control-Based Classifier-Free Diffusion Guidance
Hanyang Wang, Yiyang Liu, Jiawei Chi, Fangfu Liu, Ran Xue, Yueqi Duan
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 standard CFG with SMC-CFG | SD3.5 · MS-COCO 5k | FID: 21.421 -> 20.044 | Improved |
| replace standard CFG with SMC-CFG | Flux-dev · MS-COCO 5k | FID: 27.323 -> 26.398 | Improved |
| replace standard CFG with SMC-CFG | Qwen-Image · MS-COCO 5k | FID: 35.431 -> 33.371 | Improved |
| apply any guidance (CFG) to Qwen-Image | Qwen-Image · MS-COCO 5k | FID: 24.894 -> 35.431 | Hurt |
| replace CFG-Zero* with SMC-CFG | SD3.5 · MS-COCO 5k | MPS: 7.0430 -> 7.5719 | Improved |
| replace CFG with SMC-CFG | SD3.5 · MS-COCO 5k | Aesthetic score: 5.5875 -> 5.5790 | No clear effect |
| replace CFG-Zero* with SMC-CFG | Flux-dev · MS-COCO 5k | CLIP score: 0.3742 -> 0.3743 | No clear effect |
| increase switching gain k from 0.1 to 0.4 (ablation) | SMC-CFG (lambda=5) · MS-COCO 5k | ImageReward: 1.0248 -> 1.0504 | Improved |
| increase k from 0.4 to 1.0 (ablation) | SMC-CFG (lambda=5) · MS-COCO 5k | ImageReward: 1.0504 -> 1.0453 | Hurt |
| increase k from 0.1 to 1.0 (ablation) | SMC-CFG (lambda=5) · MS-COCO 5k | CLIP score: 0.3709 -> 0.3741 | Improved |
| increase k from 0.1 to 0.7 (ablation) | SMC-CFG (lambda=5) · MS-COCO 5k | FID: 25.951 -> 26.416 | Hurt |
| vary lambda from 3 to 5 (mid value) (ablation) | SMC-CFG (k=0.1) · MS-COCO 5k | FID: 26.193 -> 25.951 (lambda 6: 26.143) | Mixed |
Closest papers by meaning
- Adaptive Classifier-Free Guidance via Dynamic Low-Confidence Masking (NeurIPS 2025 · similarity 0.90)
- Eliminating Oversaturation and Artifacts of High Guidance Scales in Diffusion Models (ICLR 2025 · similarity 0.90)
- Rectified CFG++ for Flow Based Models (NeurIPS 2025 · similarity 0.90)
- CFG++: Manifold-constrained Classifier Free Guidance for Diffusion Models (ICLR 2025 · similarity 0.89)
- Rethinking the Spatial Inconsistency in Classifier-Free Diffusion Guidance (CVPR 2024 · similarity 0.89)
- Diffusion Models without Classifier-free Guidance (NeurIPS 2026 · 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.
