diffusers is a python library for running state-of-the art diffusion models like StableDiffusion, which has the same capabilities as other text-to-image AI tools like Midjourney and DALL·E2 .

What distinguishes StableDiffusion from the others is that the model and development are open source. You can obtain the model weights yourself from CompVis page on HuggingFace. HuggingFace also hosts a basic web frontend and the same model powers DreamStudio.

Importantly for research purposes, the diffusers library allows programmatic access which makes it possible to explore the model more systematically, iterating over input parameters and prompt constructions.

If you know a smattering of Python, diffusers can be set up and run in just a dozen or so lines of code. I’ve made a basic Jupyter notebook that runs on Google Colab for getting started.