GangTailorUpgrade/undress-service

👗 Dress AI Service – Self‑hosted AI fashion assistant

What it is – An open‑source web service that lets you digitise your personal wardrobe, get AI‑driven outfit recommendations and visualise the looks with generative‑image models. Everything runs on your own machine (Docker or plain Python), so no photos leave your computer.

Core capabilities (as described in the README)

Capability	How it works
Wardrobe digitisation	Upload photos of clothing items. The backend runs CLIP inference to auto‑tag each item (category, colour, pattern, fabric, season) and stores the embedding for similarity search.
Outfit generation	A hybrid engine combines rule‑based fashion logic (colour theory, occasion, weather) with CLIP‑based similarity and, optionally, an LLM (Ollama) that reasons about style.
Virtual try‑on	Generates photorealistic renderings of the recommended outfit using Stable Diffusion XL, FLUX‑1‑schnell or other local diffusion models.
Weather awareness	Pulls real‑time data from OpenWeather (API key required) to suggest layers appropriate for current conditions.
Analytics & helpers	Wardrobe statistics (most‑worn colours, gaps), colour‑palette extraction, duplicate detection, favourite collections, export to PNG/PDF, shareable links.

Tech stack (from the README)

Backend – FastAPI (Python 3.11) with async endpoints, SQLite by default (PostgreSQL optional).
AI/ML – CLIP for image tagging, Stable Diffusion XL / FLUX‑1‑schnell for image generation, optional Ollama LLM (e.g., llama 3.2) for natural‑language reasoning.
Frontend – Vanilla HTML + Tailwind CSS, served by the same FastAPI process; drag‑and‑drop upload and live preview.
Containerisation – Docker + Docker‑Compose (GPU‑aware compose files provided).
Other services – Weather API integration, optional PostgreSQL, optional Ollama server.

Quick start (Docker – recommended)

# clone the repo (note the repo name is "undress-service" but the app lives in the "dress‑ai‑service" folder)
git clone https://github.com/GangTailorUpgrade/undress-service.git
cd dress-ai-service
cp .env.example .env               # edit if you want a weather key or custom DB URL
docker-compose up --build          # adds a FastAPI server on http://localhost:8080

The UI is reachable at http://localhost:8080.

Local Python alternative – create a venv, install requirements.txt, run scripts/download_models.py to fetch CLIP and diffusion checkpoints, then start with uvicorn app.main:app --reload.

API surface (summary)

POST /api/v1/wardrobe/upload – add a clothing item.
GET /api/v1/wardrobe/items – list stored items.
POST /api/v1/outfits/generate – request a set of outfits for a given occasion/weather.
POST /api/v1/outfits/{id}/visualize – produce a diffusion image of a specific outfit.
Additional endpoints for CRUD, favouriting, analytics and health checks. Swagger UI is available at /docs.

Who might find this useful?

Fashion hobbyists who want a private digital closet and style suggestions.
Boutique owners looking for a self‑hosted catalogue + AI styling tool without third‑party data leakage.
Developers / researchers interested in a ready‑made pipeline that combines CLIP tagging, rule‑based recommendation and diffusion‑based visualisation.
Privacy‑conscious users who want AI‑powered styling without cloud uploads.

Extensibility points mentioned

Model swap – you can point the .env variables to different diffusion checkpoints (realistic, anime, sketch).
Plugin system – custom recommendation engines can be added.
Ollama integration – optional local LLM for richer outfit descriptions.
Future roadmap – mobile app, social sharing, e‑commerce import, 3‑D avatars, sustainability scoring, AI shopping assistant.

License

MIT – free for personal, academic or commercial use.

Bottom line: Dress AI Service is a genuine, self‑hosted AI application that combines image classification, generative diffusion, and optional LLM reasoning to turn a photo collection of clothes into a smart, privacy‑preserving fashion assistant.