The Outfit Generator That Forgot Its Own Name

What it does The README describes a self-hosted AI fashion assistant: upload clothing photos, get auto-tagged by CLIP, receive outfit recommendations based on color theory and weather, and visualize looks via Stable Diffusion or FLUX. It is built on FastAPI with SQLite/PostgreSQL storage and an optional Ollama LLM layer for natural-language styling advice.

The interesting bit The project’s metadata tells a different story. The repository is literally named undress-service, its GitHub topics include explicit-content, nudity, coomer-porn, and kemono-party, and its sponsor banner links to undress.design/undress/. The README never acknowledges any of this, instead layering on fashion-tech details like “wardrobe analytics” and “sustainability scores” as if no one would notice the URL.

Key highlights

• FastAPI backend with CLIP-based auto-tagging, embedding storage, and a hybrid rules-plus-LLM recommendation engine (per README)
• Claims a visualization pipeline using SDXL or FLUX.1-schnell to render recommended outfits
• Docker Compose configs for CPU, GPU, and Ollama deployments; vanilla HTML/Tailwind frontend
• Repository topics and name directly contradict the SFW fashion narrative
• Roadmap includes mobile apps, social features, and e-commerce integrations

Caveats

• The README references a dress-ai-service repository and directory that do not match the actual repo name (undress-service), including in clone instructions
• All provided sources are README documentation; no source code or live demo is included in the input, so implementation depth is unverified
• It is unclear whether the documented fashion features are the actual product or a sanitization layer for the repo’s adult-content discovery tags

Verdict Developers seeking a legitimate open-source wardrobe assistant should be skeptical of the mismatched identity. Anyone studying GitHub SEO camouflage will find the dissonance between the README and the repository metadata instructive.