Classifier-Free Guidance
A conditional generation technique used in diffusion models that strengthens conditional control without requiring a separate classifier network. During training, conditioning information is randomly dropped; during inference, the model blends conditional and unconditional predictions to steer generation toward the desired condition.
Timeline (last 90 days)
Classifier-Free Guidance (CFG) is one of the most mainstream guidance techniques in diffusion models and flow matching models, discarding conditions with a certain probability during training and, during sampling, subtracting the unconditional prediction from the conditional one via a formula and amplifying the guidance strength.
Classifier-free guidance (CFG) originally came from diffusion generative models; the approach is to compute both conditional and unconditional predictions during sampling, then use a weight w to amplify the difference between them, with the formula ε = ε_uncond + w·(ε_cond − ε_uncond).
Guided diffusion applies gradients at each step of the reverse denoising process; common guidance approaches include classifier guidance and classifier-free guidance.
All Facts (3)
Classifier-Free Guidance (CFG) is one of the most mainstream guidance techniques in diffusion models and flow matching models, discarding conditions with a certain probability during training and, during sampling, subtracting the unconditional prediction from the conditional one via a formula and amplifying the guidance strength.
50%UnverifiedClassifier-free guidance (CFG) originally came from diffusion generative models; the approach is to compute both conditional and unconditional predictions during sampling, then use a weight w to amplify the difference between them, with the formula ε = ε_uncond + w·(ε_cond − ε_uncond).
50%UnverifiedGuided diffusion applies gradients at each step of the reverse denoising process; common guidance approaches include classifier guidance and classifier-free guidance.
50%