Stable Diffusion Masterclass - Theory, Code & Application

Classifier-Free Guidance - Improving Image Quality

When used with diffusion models, classifier-free guidance is a technique used for improving the quality of generated images and how well those images match the text prompt without the need for a separate classification model, deeming it to be called classifier-free guidance.

The alternative to classifier-free guidance is to use a separate model to classify the generated images into different categories and is called classifier guidance. This is similar to the discriminator network's role in a GAN.