The core is pure JavaScript with zero dependencies. It works in Lab color space (not RGB) using Wu quantization, CIE perceptual distance metrics, and a DNA fingerprinting system that matches each image to one of 20+ built-in archetypes (Golden Hour, Dark Portrait, Fine Art Scan, etc.).

There's a browser-based web UI that just needs Node.js — drop an image, get 4 different possible separations side by side, export as layered OpenRaster (for GIMP/Krita), or flat PNG, or layered PSD (for Photoshop). There's also a CLI and Photoshop plugin.

Apache 2.0: https://github.com/electrosaur-labs/reveal

Some of my screen prints: https://www.electrosaur.org

The real proof of moving away from Euclidean RGB distances isn't in the composite — it's in the separated plates. Look at the individual channels of a complex texture like weathered terracotta through a Lab-aware engine, and the structural integrity of the gradients is night and day compared to legacy indexing. Even 1976-era Lab is a massive leap from RGB heuristics.