When I read about Stable Diffusion model, they usually talk about adjusting convolution layers or U-Net weights. I believe they both should be related together and the U-Net is the part that accepts the encoded image+text embedding from the VAE encoder and uses convolutional layers to extract features from the image, then adds the noise to this features, then denoises them and sends the output as a latent vector/matrix to to the VAE decoder.

But I am not sure if my understanding is completely correct?



You must log in to answer this question.