ChromaFit

ChromaFit: Your wardrobe. Decoded by color theory. Most people don't struggle to buy clothes — they struggle to wear them together. Standing in front of a full wardrobe every morning, the real question is never "do I own nice things?" It's "what actually goes with what?" ChromaFit solves this by combining multimodal vision AI with century-old color science. Users upload 3–10 wardrobe photos, and Qwen2.5-VL — served via vLLM on an AMD Instinct MI300X — analyzes every item in a single inference call, identifying clothing types and extracting dominant hex color palettes with precision. Those colors are then matched against Sanzo Wada's A Dictionary of Color Combinations — a 1933 masterwork still used by designers, textile studios, and artists worldwide — using Delta-E CIE2000, the perceptually accurate color distance metric used in professional colorimetry. Not approximate RGB guessing. Real color mathematics. ChromaFit returns up to five outfit combinations ranked by harmony score, each tied to a named Wada palette like "Palette 142 — Indigo Serenity." Users can let the AI choose freely across their wardrobe, or anchor recommendations to a specific item they've already decided to wear. The AMD MI300X's 192GB HBM3 memory handles up to 10 high-resolution images in a single 16K-context vLLM call — a workload that would require multi-GPU coordination on consumer hardware. The entire inference stack runs on ROCm, with no CUDA dependency. ChromaFit doesn't guess your style. It reads your colors — and finds combinations that are mathematically and historically proven to work.