本ページの本文は英語で提供されています。タイトルと導入は日本語化されています。
これは何か
Diffusers is an open-source library from Hugging Face, released in 2022, offering a unified implementation of diffusion models. It splits a generation run into a model, a scheduler and a pipeline, so any part can be swapped without rewriting the flow. It targets the difficulty of aligning and reproducing results across the many implementations behind text-to-image, image-to-image and image editing.
なぜ覚えておく価値があるか
Its modular split into model, scheduler and pipeline made diffusion experiments and comparisons reproducible, rather than collections of result images.
主な仕様
- Components
- Model + scheduler + pipeline
- Pipelines
- Text-to-image, image-to-image, editing and more