Hey HN! I’ve been working with AI models like Stable Diffusion and DALLE-3 for a while now and realized that building real applications with them usually requires building a multi-step image processing workflow, either to chain several models or combine AI results with basic image processing functions like cropping and resizing.

For example, you might want to take an input image, enhance lighting, upscale it, crop to square, and then add a text overlay. You can use a hosted AI model to solve some pieces of the puzzle, but ultimately you need to write your own code to glue everything together. I got pretty frustrated doing this for every project and decided to build ML Blocks.

ML Blocks lets you create custom image processing workflows that can handle end-to-end image generation or analysis pipelines using simple node-based workflows. Just connect a bunch of blocks and you can instantly deploy your entire workflow as a single API endpoint.

This initial version has around 20 blocks to play with. I will continue to add blocks as requested and I can see a possible future where this becomes like a Zapier for Image Processing, where blocks are community-contributed with a standardized API interface. I’d love to get your thoughts on this!