This open-source website provides a simple interface for modifying images using text-based instructions. You can upload an image, provide a text prompt describing how to change that image, and generate new images based on the prompt.
The image generation is powered by InstructPix2Pix, an open-source machine learning model that combines the knowledge from GPT-3 and Stable Diffusion to generate a large dataset of image editing examples. This model was created at the University of California, Berkeley by Tim Brooks, Aleksander Holynski, and Alexei A. Efros.
The model is hosted on Replicate, which exposes a cloud API for running predictions. This website is built with Next.js and hosted onVercel, and uses Replicate's API to run the InstructPix2Pix model. The source code is publicly available on GitHub. Pull requests welcome!