Pick your favorite cute animal and watch the denoising diffusion model transform random noise into a stunning image
Gradually adds Gaussian noise to an image until it becomes pure noise. This is the training phase of the model.
A neural network learns to reverse the noise process, gradually removing noise step by step to generate a new image.
CFG scale controls how strongly the model follows the prompt. Higher values produce more prompt-adherent but less diverse images.
An encoder-decoder architecture with skip connections that predicts noise at each timestep for the denoising process.