paipiange/nvzhuang

This is a dark-themed static catalogue of 1,407 AI-generated clothing images (2:3, no people in frame), organized into eight seasonal and style categories, plus an optional AI virtual try-on workbench that composites an uploaded portrait with a selected garment. It suits designers, e-commerce creators and hobbyists looking for clothing references or a self-hostable try-on demo, not people seeking real product photos.

Browsing without any setup

The front end is a plain static site: no build step, no API key, no backend. Open the online preview or serve index.html locally to get category filtering, live search in Chinese or English, infinite scrolling, a full-screen viewer with arrow-key navigation, and URL parameters for sharing a language or search term. Search matches localized keyword metadata, so Chinese terms work even though the underlying data.js field names and keywords are Japanese/English. The repository itself does not include the full-resolution PNGs (roughly 1 GB+); the download button points to the original site's assets, so plan on web-resolution images only.

What the optional try-on adds

The AI try-on requires the separate GoFrame backend in backend/. You choose a garment, upload a portrait, and the backend sends both images to an image model to transfer the clothing while keeping the face. Documented adapters cover OpenAI Images-compatible endpoints (GPT Image 2 and other OpenAI-compatible domestic providers), Doubao Seedream, Gemini image models, and a local ComfyUI workflow (Qwen-Image-Edit-2511 with a 4-step Lightning LoRA, falling back to MiniMax H3 if not ready). API key, base URL and model name are all configurable, so switching providers does not need code changes.

Requirements and limits worth knowing

• Cloud providers consume your own third-party quota; the README treats Gemini as the default cloud option and notes Doubao works domestically without a proxy. Local ComfyUI needs a suitable GPU.
• Backend deployment is documented for Docker Compose, single-container Docker, and direct go run; a Windows batch script is provided. For public deployments the README lists Bearer-token auth, CORS allow-listing, default rate limits of 10 requests/minute and 100/day, concurrency caps, and strict image validation (JPEG/PNG/WebP only, 8 MiB per image, 16 MiB combined, 256–8192 px sides, 20 MP total).
• Configurations saved in the browser's settings center stay local to that browser; personal base URLs are blocked from resolving to localhost, private and reserved addresses to avoid the public backend being used as an internal request proxy. The README advises clearing local configuration on shared devices. Successful generations are stored in the browser's IndexedDB and are not uploaded or synced.
• All images are AI-generated conceptual designs, not real brand product photos.

The project is MIT-licensed, and the README states code and assets may be used commercially, modified and redistributed. See the README for model configuration, the backend deployment guide for security details, and the online preview to try browsing.