What it does
HairFit lets a client upload a portrait and a hairstyle reference, see an AI hairstyle preview, inspect the result in a real GLB-based 3D scene, switch between front, side and back views, trim visible hair bundles, restore the original model, and save a consultation card.
How it was built
The project was built with the Devpost Learn Skill Pack. It uses a local Node.js proxy, a browser upload flow, an image-generation API configured through .env, Three.js for the interactive scene, a Tripo-generated GLB reconstruction, and locally hosted MediaPipe assets for the optional hand interaction.
What we learned
The useful boundary was separating the AI preview from the editable 3D representation. The demo exposes the reconstruction source, labels inferred rear geometry, keeps the paid task durable, and treats webcam pinch as an experimental extension with diagnostics.
Honest limits
Side and back geometry are model inferences from incomplete image evidence. Virtual trimming changes the reconstructed hair representation and does not promise individual-fiber simulation or exact salon length.