Skip to content

About

A browsable library of pose and costume references, with a rigged 3D mannequin for every pose.

Topics

Resources

Stars

12 stars

Watchers

0 watching

Forks

Repository files navigation

AI Library

AI Library logo

A browsable library of pose and costume references, with a rigged 3D mannequin for every pose.

  • 288 poses, each drawn from five angles: front, above, below, side and diagonal.
  • 1,407 costumes across eight categories, searchable by colour, garment, fabric and accessory.
  • A rigged 3D model for every pose, shown side by side with its drawing, ready to adjust and download as .glb or .bvh.

Built with Next.js 16, React 19, Tailwind CSS 4, shadcn/ui and three.js.

The app uses the matching mark as its browser icon and generates a 1200 × 630 Open Graph image for link previews. The logo artwork is in public/logo.svg, and the social card is served at /opengraph-image and included in the Open Graph and Twitter metadata.

The pose library with a pose open in the detail panel

A pose on the rigged 3D mannequin, side by side with its drawing

The costume library filtered to winter, with a costume open

Sample references

Every pose is drawn from five angles. P-GL-001, S-curve stand, hip pushed to the side:

Front Above Below Side Diagonal
P-GL-001 from the front P-GL-001 from above P-GL-001 from below P-GL-001 from the side P-GL-001 from the diagonal

One costume from each season:

Spring · C-SP-0001 Summer · C-SU-0001 Autumn · C-AU-0001 Winter · C-WI-0004
Spring costume C-SP-0001 Summer costume C-SU-0001 Autumn costume C-AU-0001 Winter costume C-WI-0004

The images

All images are in public/media and are served as static files:

Folder Contents
public/media/poses/thumbs 1,439 pose thumbnails (JPEG)
public/media/poses/full 1,439 full-size pose drawings, 900 × 1200 (WebP)
public/media/costumes/thumbs 1,407 costume thumbnails (WebP)
public/media/costumes/img 1,407 costume images (WebP)

Every image can also be downloaded from its detail panel in the app.

To serve images from your own storage or CDN, upload the public/media folder and set NEXT_PUBLIC_MEDIA_BASE_URL to its base URL.

Getting the rigged model

  1. Open any pose and choose 3D rig & export.
  2. Compare the mannequin with the drawing using Side by side and the angle tabs.
  3. Optionally adjust joints with Pose and the body shape with Body, then Save pose to library. Mirror pose flips it left to right and Vary pose nudges every joint a little for a fresh variation.
  4. Choose Download .glb for a rigged, posed mesh, or Download .bvh for the skeleton alone.
  5. Choose Use in studio to send the mannequin, exactly as framed, to the studio as the pose reference in place of the drawing, together with its depth map.

Save as new pose draws the current pose in the library's style from all five angles (front, above, below, side and diagonal) at 900 × 1200, with a depth map for each, and keeps it under My poses in this browser. Saved poses can be browsed, downloaded, reopened in the 3D rig and used in the studio like any other pose, so editing, mirroring and varying a pose gives you new ones.

Depth shows the posed mannequin as a depth map (lighter is nearer the camera), and Depth .png downloads it. A depth map tells an image model what the outline cannot: which arm or leg is in front, and how the body turns and leans. In the studio it is sent beside the render, and the prompt explains how to read it.

The model uses a 17-bone skeleton (pelvis, spine, chest, neck, head, upper arms, forearms, hands, thighs, shins and feet), which imports into Blender and any tool that reads glTF or BVH.

The fitted rig for a pose is also available as JSON:

GET /api/fits/:code        e.g. /api/fits/P-GL-001

It returns the bone rotations, the camera angle of each drawing and the measured body proportions.

Pose from your image puts the pose from any photo or drawing onto the mannequin. The image is read in the browser and never uploaded.

See it in action: turn a photo into a pose and inspect the fitted mannequin, or compare the source image with the drawing and explore the mannequin views.

Turn a photo into a pose

Video files: Part 1 · Part 2

Compare the source image with the drawing and explore the mannequin

Video files: Part 1 · Part 2 · Part 3

Studio

The studio puts your own model into any pose and costume with an image-generation model.

Watch the studio workflow: add a model photo, pose and costume, describe the scene, then generate an image.

Video files: Part 1 · Part 2 · Part 3 · Part 4 · Part 5 · Part 6

The studio: a pose and a costume sent as references, and the generated photograph

  1. Add up to three photos of the person to draw, and optionally describe them. Optionally add an environment photo, whose background, light and camera style the result follows.
  2. Choose up to six poses and, optionally, one costume. Any pose or costume can also be sent here with Add to studio from its detail panel.
  3. Under Image settings, add one or more services. Puter needs no key: you sign in with your own Puter account, which pays for what you generate, and the image is made in your browser. Cloudflare Workers AI takes your account ID and an API token with Workers AI permission, and is billed to that Cloudflare account with a free daily allowance; it uses the FLUX.2 models, which take up to four reference images, so the studio keeps the most useful four and shrinks them to fit 512 px. The other presets take an API key: Google Gemini, OpenAI, OpenRouter, xAI, Together, SiliconFlow, Zhipu, Seedream and BytePlus, and any other OpenAI-compatible service can be added by its base URL and model. Each service lists its known models, and Custom model takes any other model ID. Keys are kept in your browser and sent only with each generate request.
  4. Tick the services to generate with, describe the scene and send. Each pose becomes its own image, or all of them share one sheet, and every selected service makes its own version.

The default prompt asks for a photorealistic fashion editorial. It names the role of each reference image in order (model photos, environment, outfit, pose), keeps the outfit and the pose exactly, and adds what you type as the scene and direction.

A default negative prompt (malformed hands and limbs, extra fingers, watermarks, text and so on) is sent with every image; change it under the options menu with Negative prompt. Services that take a separate negative prompt get it as one, and the others read it as an "Avoid" line.

Every chat and generated image is kept in this browser, so a reload brings them back. Images lists everything generated, newest first: open any image full size, download or delete it, or download or delete them all. API keys are never stored with them.

Chats are listed beside the studio (under Chats on smaller screens): each one keeps its own turns and images, several can generate at once, and the page asks before a reload or close while anything is still being made.

Google Drive. After the first image, the studio asks whether to save new images to Google Drive automatically. Saving uses Google's drive.file permission, so the site can only see the files it creates, in an AI Library folder; turn it on or off in Image settings, or save single images from the gallery and viewer. To enable it on your own deployment, create an OAuth client ID (type Web application) in Google Cloud, add the site's origins as authorized JavaScript origins, enable the Google Drive API, and set NEXT_PUBLIC_GOOGLE_CLIENT_ID before building. Without it, Drive saving stays hidden.

Without a key, Copy prompt gives you the full prompt to use in any other image tool, with the model photos, the costume image and the pose drawing as references in that order.

To use a key held on the server instead, set IMAGE_API_KEY in the environment, and optionally IMAGE_PROTOCOL (gemini, openai, seedream, chat or cloudflare), IMAGE_BASE_URL and IMAGE_MODEL; Cloudflare also needs IMAGE_ACCOUNT_ID.

API

Endpoint Returns
GET /api/poses?q=&group=&limit= poses with their five images, a ready-to-use prompt and a link to their rig
GET /api/costumes?q=&category=&colour=&limit= costumes with their keywords, a ready-to-use prompt and images
GET /api/fits/:code the fitted rig for one pose
POST /api/generate one generated image, used by the studio

The search endpoints leave out the adult categories unless include_adult=true is passed.

Features

  • Filter poses by group and class, and costumes by category, colour and keyword.
  • A written description of every pose and costume, ready to copy as a prompt, plus a default negative prompt. Duo poses say how the woman and man are placed. Full-body poses also describe where each hand and leg is, read from the drawing; the camera line follows the chosen angle.
  • Search, sort and shuffle; every view is a shareable URL.
  • Favourites, recently viewed, saved rigs and boards for collecting one character's references.
  • My poses: new poses made in the 3D rig, drawn from five angles with depth maps.
  • Compare up to four items, full-screen zoom, and image adjustments (mirror, invert, contrast, tint).
  • Export and import your favourites, boards and saved rigs as JSON.
  • A switch to hide the adult categories (Glamour and Duo poses; Lingerie, Swimwear and Glamour costumes).
  • Dark and light themes; works on phones, tablets and desktops.

Project structure

data/                      pose and costume catalogue, and the joint data the rigs are fitted from
docs/screenshots/          the screenshots in this README
public/media/              pose and costume images
public/assets/             the base mannequin body (glTF)
src/app/(library)/         poses, costumes, collections, boards, pose-from-image and studio screens
src/app/api/               rig fitting, search and image generation
src/components/library/    shell, sidebar, grid, inspector and dialogs
src/components/ui/         shadcn/ui primitives
src/hooks/client/          library state, URL state and the 3D viewer launcher
src/lib/                   catalogue, saved library state, rig fitting and the 3D viewer

Licence

The source code is released under the MIT License. The base mannequin body in public/assets is CC0; see public/assets/base-female-LICENSE.txt.

About

A browsable library of pose and costume references, with a rigged 3D mannequin for every pose.

Topics

Resources

Stars

12 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages