Skip to content
Projects

atuizz/posewear has 115 GitHub stars and 45 forks. GitHot measured +2 stars in 24 hours and +27 in 7 days. Last checked 2026-11-15 15:55 UTC. Last push 2026-10-31.

How these counts are measured

Measured star growth

in 24h
+2
in 7 days
+27
in 15 days
Update delayed
in 1 month
Update delayed

Net star changes over overlapping periods, not daily totals. Missing measurements are left blank. Stars measure attention, not software quality.

About

A pose, outfit, and portrait reference library with prompts in Chinese and English, plus AI try-on in your browser.

Creative & Media

README.en.md

Original
Portrait + pose + outfit = result

The result is an AI-generated example using a portrait, pose and outfit from the galleries as references. Source images belong to their authors; see Acknowledgements.

PoseWear

PoseWear 衣姿

Pose, outfit and portrait reference galleries with bilingual prompts, plus a bring-your-own-key AI try-on studio that runs in your browser.

License: MIT Cloudflare Pages Live demo GitHub stars

Live demo · Features · Quick start · Deploy · FAQ · 中文

Home

Features

  • Three galleries: poses (multi-angle line sketches), outfits (by season and type), and portraits (face references, CC0). Every image has a Chinese description and an English prompt you can copy.
  • Search and filters: search in Chinese or English, filter by category, sort, or pick a random image. Each item has its own static page for sharing.
  • Try-on studio: upload a photo, pick an outfit and a pose, then generate. You can also generate without a photo. It supports several outputs per run, a before/after slider, and single or zip download.
  • Generation history: stored locally in IndexedDB. You can view, delete or clear items, and undo.
  • Clear errors: an invalid key, a rate limit, a timeout or a safety block each get their own message and a suggested next step.
  • Reduce false safety blocks: on by default. The studio rewrites its prompt into a neutral e-commerce try-on description without changing the outfit or pose. Turn it off to send the prompt as is.
  • Favorites, recent items and prompt combos, all stored locally; an 18+ toggle with age confirmation; PWA and offline support; dark mode; a Chinese/English UI.

Providers

Provider You need Default model Notes
OpenAI API key (gpt-image requires a verified org) gpt-image-1 images/edits with multiple references; strictest moderation
Gemini Google AI Studio key gemini-2.5-flash-image a.k.a. Nano Banana
Doubao Seedream Volcengine Ark key doubao-seedream-4-0-250828 routed through the site proxy by default for CORS
Custom relay Base URL, key and model any OpenAI-compatible; choose edits (multipart) or generations (JSON)

Screenshots

Studio Dark mode Mobile
Studio Dark mode Mobile

Quick start

You need Python 3.9+ with Pillow. Tests and deploys also need Node.js 18+.

git clone https://github.com/atuizz/posewear.git && cd posewear
pip install pillow
python3 scripts/fetch-assets.py   # downloads data and images from the original galleries (several GB)
python3 build.py                  # outputs dist/
python3 -m http.server 8787 -d dist

Set POSEWEAR_RAW to use a different asset folder, and POSEWEAR_SITE to set your own site URL for the sitemap and share links.

Deploy

The repository ships without assets, so build locally first, then upload dist/ to Cloudflare Pages:

npm install && npx wrangler login
POSEWEAR_SITE=https://<project>.pages.dev python3 build.py
npx wrangler pages deploy dist --project-name <project>

functions/api/proxy.js is deployed as a Pages Function that only forwards API requests. Adding a rate-limit rule for /api/proxy is recommended.

Tests

The Playwright tests mock every provider endpoint, so no real key is needed. They cover success, a bad key, rate limits, 5xx errors, timeouts, cancel, safety blocks and multiple outputs.

npm install && npx playwright install chromium
python3 -m http.server 8787 -d dist &
node tests/v12.js && node tests/v13.js

FAQ

Is my API key safe? It stays in your browser's localStorage. Requests go straight to the provider. They only go through /api/proxy if you turn on proxy forwarding or use a custom relay; the proxy forwards requests as they are and never stores keys or images.

Why was my image blocked? Providers moderate both the input photos and the output. Real photos with revealing outfits are the most likely to be blocked, and OpenAI is the strictest. The rewrite option reduces false positives on normal outfits. It does not bypass moderation of prohibited content.

Can I use the images? No images are included in this repo. All assets belong to their original authors, so check each source site's terms. The portrait gallery is CC0.

Contributing

Issues and PRs are welcome. Run python3 build.py and both test files before you open a PR, and add screenshots for UI changes. Please don't commit any assets or keys.

Acknowledgements

AI Pose Gallery · AI Costume Gallery by @giaaibusi · AI Pose Depth Gallery by ギア / @giaaibusi (hotlinked, unofficial Chinese index; images are not in this repo) · AI Portrait Gallery by @agi_aibusi (CC0) · wzc520pyfm/gen-ai-pose-image · paipiange/nvzhuang

License and asset boundaries

What In this repo? License
Code, UI copy, build scripts Yes MIT
Pose and outfit images and data No — downloaded by fetch-assets.py Owned by @giaaibusi (AI Pose Gallery / AI Costume Gallery); no open license is stated
Portrait images No — same as above CC0 1.0 by @agi_aibusi (AI Portrait Gallery)
Depth-pose images No — hotlinked with <img> from AI Pose Depth Gallery Owned by ギア / @giaaibusi. This repo only has the text metadata and Chinese names
Depth-pose text metadata Yes, depth/packs/ Japanese text belongs to the original author; Chinese names are ours
README screenshots Yes The pictures shown belong to their authors as above
  • MIT covers the code only, not any image or data.
  • To repost, use commercially or redistribute pose/outfit images, ask the author @giaaibusi first and follow the source site's terms.
  • Portraits are CC0: free to use, modify and sell without credit. They are AI-generated fictional people — don't use them to impersonate real people. The author asks that they not be bulk-reposted as-is.
  • If you deploy your own copy, keep the source and author credits on the pages.