Core

Minimal pipeline for Diffusion Guidance experiments.

---

source

MinimalDiffusion

 MinimalDiffusion (model_name, device, dtype, generator=None)

Loads a Stable Diffusion pipeline.

The goal is to have more control of the image generation loop. This class loads the following individual pieces: - Tokenizer - Text encoder - VAE - U-Net - Sampler

The self.generate function uses these pieces to run a Diffusion image generation loop.

This class can be subclasses and any of its methods overriden to gain even more control over the Diffusion pipeline.