yuval-alaluf/Attend-and-Excite
An attention-based technique for improving Stable Diffusion's ability to generate all subjects described in text prompts by guiding cross-attention during inference.
Not currently ranked — collecting fresh signals.
star history
Attend-and-Excite is a Generative Semantic Nursing method that intervenes in the diffusion model’s inference process to improve faithfulness of generated images. It analyzes Stable Diffusion’s cross-attention units to identify and strengthen activations for subject tokens that may be neglected during generation. The approach guides the model to attend to all subject tokens in the text prompt and excite their activations, helping ensure all described subjects appear in the final image.
Frequently asked
- What is yuval-alaluf/Attend-and-Excite?
- An attention-based technique for improving Stable Diffusion's ability to generate all subjects described in text prompts by guiding cross-attention during inference.
- Is Attend-and-Excite open source?
- Yes — yuval-alaluf/Attend-and-Excite is open source, released under the MIT license.
- What language is Attend-and-Excite written in?
- yuval-alaluf/Attend-and-Excite is primarily written in Jupyter Notebook.
- How popular is Attend-and-Excite?
- yuval-alaluf/Attend-and-Excite has 771 stars on GitHub.
- Where can I find Attend-and-Excite?
- yuval-alaluf/Attend-and-Excite is on GitHub at https://github.com/yuval-alaluf/Attend-and-Excite.