sapanablog/egyptian-relief-colorization — reverse-engineered prompt

Build me a Python project that can take a grayscale photo of an ancient Egyptian relief and turn it into a plausible color reconstruction. I want the full flow to match the thesis style idea in the README, first detect likely pigment regions, then turn those regions into color hints, then use a transformer based colorization model to produce the final image.

Please make it feel complete and runnable, not just model code. It should work on the included example relief images and also let me point it at my own image with one simple command. Save the intermediate mask and hint image as well as the final colorized output so I can see each stage. If there are pretrained weights or setup steps needed, wire that up and document it clearly.

Keep the code clean and practical, with a short README that explains how to install, run inference, and where outputs are saved. If anything is missing in the repo, look up current docs online and fill in the gaps.