大纲

This notebook is going to cover the basics of how to use Stable Diffusion to create and modify images using existing pipelines. We'll also take a brief look at the key components within the pipeline, while leaving further exploration of them to the deep dive notebook. Specifically, we will cover:

• Generating images from text using the StableDiffusionPipeline and experimenting with the available arguments
• Seeing some of the key pipeline components in action
  • The VAE that makes this a 'latent diffusion model'
  • The tokenizer and text encoder that process the text prompt
  • The UNet itself
  • The scheduler, and exploring different schedulers
• Replicating the sampling loop with the pipeline components
• Editing existing images with the Img2Img pipeline
• Using inpainting and Depth2Img pipelines

步骤

Setup

Generating Images from Text

Pipeline Components

Additonal Pipelines