projtex · Projection Texturing API

projtex · Projection Texturing API

Textures 3D meshes from a text prompt and returns a GLB with one baked colour texture. Built for AI agents and developers, with a REST API and an MCP endpoint.

Made by townbox on Hugging Face. These notes are Mindket's, written from its Space page and README; the app's own files and license are on Hugging Face.

About

What it is: projtex is a service that paints 3D meshes. Give it a mesh and a text prompt, and it hands back the same mesh with a colour texture baked in, as a GLB file. Underneath, it renders images from virtual cameras, checks them against the geometry, projects them onto the mesh and bakes the blend into the UVs.

Who it suits: Developers and agent builders who need textured 3D assets without painting them by hand, and 3D artists who want a quick first pass.

Industry | Who uses it | How they use it | Example
Game development | Indie developers | Texture a prop from a written description | Describing a house model's siding, roof and door
Web 3D | Developers using three.js | Texture a GLB exported from their scene | Sending an exported GLB and getting a textured one back
AI agents | Agent builders | Let an agent call the service as a set of tools | An agent texturing a mesh through the texture_mesh tool

You can skip it if you do not work with 3D meshes or have no use for baked textures.

Where to use it:

• A web browser, for the Space's page and its built-in docs
• Any script or REST client that can call the service's API
• An MCP client, through the service's remote MCP endpoint
• Windows, macOS or Linux with Docker, as a copy on your own computer

Use case | Who | What they get back
Texture a mesh | Developers and artists | A GLB with one baked colour texture and a quality verdict
Texture terrain and ground | Level builders | A ground GLB with a top-down colour map, plus a zip of splat maps and detail textures
Add PBR materials | 3D artists | A GLB whose base colour has the lighting removed, plus a metallicRoughness texture

How to set it up:

• A browser, if you only use the Space online
• Docker and Python with pip, if you run a copy yourself
• A Hugging Face token, which the examples send in an X-HF-Token header

1. Open the Space's page on Hugging Face to use the service online. Nothing needs installing.
2. For a local copy, install the huggingface_hub package and download the Space's files.
3. Move into the new folder and build the Docker image.
4. Run the image with port 7860 open.

pip install -U huggingface_hub
hf download townbox/projection-texturing --repo-type space --local-dir projection-texturing
cd projection-texturing
docker build -t projection-texturing .
docker run -it -p 7860:7860 projection-texturing

How to use it: Each texturing run is an asynchronous job. You submit a job, then poll its status until the state reads done. The status carries progress, a quality summary, file links and a list of follow-up calls.

You can try it without a mesh by sending a built-in sample with a prompt in the job body.

{"sample": "lantern", "prompt": "black wrought iron, weathered oak post, warm glass"}

Tool | What it does
texture_mesh | Textures a mesh from a prompt
texture_terrain | Builds terrain and ground textures
get_job | Reports a job's status, with a preview image
regenerate_views | Redoes chosen camera views
disable_views | Leaves chosen views out of the result
fill_gaps | Fills uncovered spots in the texture
retouch | Repairs, replaces or upscales a local area
bake | Bakes the views into the mesh again
submit_view_image | Lets you supply your own images

Example: You upload a GLB of a house and ask for white clapboard siding, a grey shingle roof and a dark green door. You poll the job. When its state reads done, the answer lists the textured GLB, a contact sheet and a quality verdict.

For newcomers and experts: If you are new to 3D, start with the built-in sample and a short prompt. It needs no mesh of your own, so you can see what a finished job looks like.

For experts, three image models sit behind it, picked with a backend field: Qwen-Image-2.1-Turbo by default, the full Qwen-Image-2.1 and Bria Fibo 2.0 Edit Turbo. A mock backend gives free dry runs, and an external one takes your own images. Coverage is measured over what the cameras can see, and the underside is reported apart. The trade-off is quota, since each view spends about 30 seconds of the caller's GPU time.

Submit a mesh and a prompt as a job
Virtual cameras generate images
Each image is checked against the geometry
Images are projected and blended
The result is baked into the UVs

Good to know:

• License: mit
• Cost to run: image generation spends your own daily ZeroGPU quota, about 30 seconds a view, and dry runs with the mock backend are free
• Data it reaches: your uploaded mesh, a public mesh link or a built-in sample, and image models hosted in other Hugging Face Spaces
• Input formats: meshes go in as GLB, glTF, OBJ, PLY or STL files, or inside a zip

Title: Generation steps per model
Qwen-Image-2.1-Turbo: 8 steps
Qwen-Image-2.1: 40 steps

On Hugging Face, 13 November 2026

• Likes: 5
• Built with: docker
• License: mit

Links

• Space: https://huggingface.co/spaces/townbox/projection-texturing
• Files: https://huggingface.co/spaces/townbox/projection-texturing/tree/main

Try it

Open the Space in your browser, or copy it to your own Hugging Face account from its page. Nothing in this kit runs by itself.