jintae-00/ReDesign ? reverse-engineered prompt

Build me a tool that takes a flat design image and turns it back into something editable, like a layered design file.

I want to drop in a screenshot or exported mockup, and have it figure out the text, shapes, images, groups, and layer order, then save the result as a clean JSON structure I can inspect or use later. It should try to recover real text styling too, not just plain OCR, and it should separate overlapping parts as best it can so the final result is editable instead of just one flat picture.

Please make it work for a single image first, and keep it organized so it could also run on a batch of design images later. If you need current docs or model setup details, look them up online. Add a simple way to configure API keys and checkpoints, and include a clear example of how to run it on an image and get the reconstructed output.