Cross-Attention Visualization Utils of Stable-Diffusion Models by Generative or Discriminative Mode

This repository is a trivial reimplementation of cross-attention map visualization utils by generative and discriminative mode, referring prompt-to-prompt codebase.

Installation

conda create -n diff.attn python=3.8
conda activate diff.attn
pip install -r requirements.txt

Usage

python attn_utils --help

or

jupyter lab # then open demo.ipynb from the left folder