1407 AI dresses and a Go backend that puts them on you

What it does It’s two things stapled together. First, a static, no-key-needed gallery of 1407 AI-generated women’s clothing images (2:3, no people in frame), browsable in Chinese and English with bidirectional keyword search, eight seasonal/category filters, and a Taobao reverse-image-search shortcut to find real-world equivalents. Second, an optional Go (GoFrame v2) backend that does virtual try-on: you upload a portrait, pick a garment, and an image model merges them while keeping your face and build.

The interesting bit The try-on backend is pluggable across wildly different providers — OpenAI Images, Gemini, Doubao Seedream, Qwen, Zhipu, SiliconFlow, or a local ComfyUI workflow — all normalized behind aliases in one config file. The ComfyUI path runs fully offline with Qwen-Image-Edit-2511 plus a 4-step Lightning LoRA, and auto-falls back to MiniMax H3 if the model isn’t ready. Privacy handling is unusually thoughtful: browser-saved API keys are sent per-request only, never logged, and personal Base URLs are blocked from pointing at localhost or internal networks to stop your public backend becoming an SSRF proxy.

Key highlights

• 1407 garment-only reference images across 8 categories (spring through swimwear), MIT-licensed and explicitly cleared for commercial use
• Chinese↔English bidirectional search with automatic keyword localization
• Try-on history stored locally in IndexedDB; uploaded originals live only in page memory and vanish on refresh
• Adding a new OpenAI-compatible model is a config stanza, not a code change
• Gallery half is pure static frontend — works with zero deployment or keys

Caveats

• The gallery is entirely AI-generated imagery, so treat it as design reference, not a catalog of real garments
• Try-on quality depends entirely on whichever third-party model you wire up; the repo doesn’t benchmark them
• The whole thing is a Linux.do community meme (“3rd-level members get to see the crossdressing content”) — the README leans into the joke hard, which may or may not be your vibe

Verdict Worth a look if you build image-generation tooling and want a clean reference implementation of multi-provider try-on with sane key handling. If you just wanted a dress catalog, the static gallery alone is a decent freebie.