wooyeolbaek/attention-map-diffusers
A tool for extracting and visualizing cross-attention maps from text-to-image diffusion models using HuggingFace Diffusers.
Not currently ranked — collecting fresh signals.
star history
This repository provides utilities for extracting and visualizing cross-attention maps from diffusion models. It hooks into the HuggingFace Diffusers library to capture attention weights during the diffusion process, enabling analysis of how text prompts influence image generation. The tool is compatible with Stable Diffusion 2/3, SDXL, and Flux models, supporting both batch operations and layer-wise/timestep-based attention map saving.
Frequently asked
- What is wooyeolbaek/attention-map-diffusers?
- A tool for extracting and visualizing cross-attention maps from text-to-image diffusion models using HuggingFace Diffusers.
- Is attention-map-diffusers open source?
- Yes — wooyeolbaek/attention-map-diffusers is open source, released under the MIT license.
- What language is attention-map-diffusers written in?
- wooyeolbaek/attention-map-diffusers is primarily written in Python.
- How popular is attention-map-diffusers?
- wooyeolbaek/attention-map-diffusers has 409 stars on GitHub.
- Where can I find attention-map-diffusers?
- wooyeolbaek/attention-map-diffusers is on GitHub at https://github.com/wooyeolbaek/attention-map-diffusers.