// ✦ Diffusion, Flow Matching & Generative Models
Classifier-Free Guidance (CFG)
Extrapolates between conditional and unconditional predictions to follow prompts.
无分类器引导 CFG 2022 Intermediate core
IN PLAIN WORDS
Randomly drop the condition in training so one model does both; at sampling, push along the difference. Higher guidance follows prompts more at the cost of diversity.
Used in: Nearly all text-to-image/video (guidance_scale)
01Original paper
02Where it sits
Learn the left column first; the right column builds on this one.
Classifier-Free Guidance (CFG)
Then you can learn
- Nothing in the atlas builds on it yet
Related:Latent Diffusion Models
Left: prerequisites · right: builds on this · dashed: related. Full map →
03Learning resources
Official docs, explainers, courses and reference code. All links checked.
04Where the course teaches it
On learning paths:Generative Models & Frontier Research
For AI / data
- Verification: verified, checked 2026-10-21.
- JSON: /api/learn/algorithms.json (id:
cfg)