Texture-Craft

Texture-Craft

Makes transparent 16x16 item textures from text prompts with a diffusion model, through a Gradio Space, a local REST server or an MCP tool. For game makers and agent builders.

Made by ArshPathan 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: This project turns a text prompt into a small transparent item texture, 16x16 pixels, using a diffusion model. You can run it as a Gradio Space with MCP tools, as a local REST server, or as a local MCP server over stdio. Each result comes back as a PNG encoded in base64.

Who it suits: It suits indie game makers, modders and tool builders who need many small item icons fast. It also suits developers who want an agent to create textures through an MCP tool.

Industry | Who uses it | How they use it | Example
Games | Indie developers | Make item icons for a game | Ask for a ruby sword with a brown handle
Modding | Mod authors | Fill a pack with new item art | Generate several variants of a potion bottle
Software | Agent builders | Give an agent a texture tool over MCP | Call the generate_texture tool from an MCP client
Design | Pixel artists | Get rough tiles to refine by hand | Export a tile and clean it up

Anyone who needs large, detailed artwork can skip it, since the output is a 16x16 tile.

Where to use it: It works in several places.

• A web browser, on the page of the Space at Hugging Face
• An MCP client, through the Space or the local stdio server
• Windows, macOS or Linux, as a local copy with Python 3.12
• Any program that can call the local REST server

Use case | Who | What they get back
Item textures | Game makers | A transparent 16x16 PNG for each prompt
Variants | Artists | Several versions of one prompt, each with its own seed
Agent tooling | Developers | A texture produced from inside an agent workflow
Asset pipelines | Studios | Tiles as base64 PNG or a data URI

How to set it up: The online page needs nothing installed. A local copy needs the following.

• Python 3.12
• Windows, macOS or Linux
• The packages listed in requirements.txt
• uv, for the quick start in the README
• No Hugging Face token, since the public model needs none

1. Install Python 3.12.
2. Install the Hugging Face hub package and download a copy of the Space.
3. Move into the new folder and install Gradio 6.31.0 and the packages in requirements.txt.
4. Start the Space app with python app.py.
5. For the local REST server, sync the project and run server.py.
6. For a local MCP server, point your client at local_mcp_server.py.

pip install -U huggingface_hub
hf download ArshPathan/TexuteCraft --repo-type space --local-dir TexuteCraft
cd TexuteCraft
pip install gradio==6.31.0
pip install -r requirements.txt
python app.py

uv sync
python server.py

{
  "mcpServers": {
    "texture-craft": {
      "command": "C:\\path\\to\\Texture-Craft\\.venv\\Scripts\\python.exe",
      "args": ["C:\\path\\to\\Texture-Craft\\local_mcp_server.py"]
    }
  }
}

How to use it: Write a prompt and ask for one or more variants. In the Space or an MCP client, call the generate_texture tool. On a local copy, this command starts the Space page.

python app.py

Tool | What it does
generate_texture | The MCP tool that turns a prompt into a texture
server.py | Runs the local REST server
local_mcp_server.py | Runs the local stdio MCP server built on FastMCP
app.py | Runs the Gradio Space with its built-in MCP server
Health endpoint | Reports whether the runtime is on GPU or CPU

Example: You ask for a ruby broadsword with glowing runes, two variants, seed 42, 30 steps and a guidance scale of 9.0. You get back two transparent tiles, each as a base64 PNG and a data URI, with its seed and latency.

For newcomers and experts: If you are new to this, start with a short, concrete prompt such as a blue potion bottle with a cork. Ask for one variant, then try again with a different seed to see how the result shifts.

If you know the field, generation runs a Stable Diffusion 1.5 based model, yuuki14202028/minecraft-item-16px, and keeps a transparent PNG by removing magenta and edge-connected flat backgrounds. Transparency is inferred from the image edges as well as the model's magenta chroma key. The Space exposes only texture generation and accepts no filesystem paths from callers.

User prompt
TextureEngine.generate_single
Stable Diffusion inference
16x16 transparent tile
Base64 PNG and data URI returned

Good to know: Backgrounds that are flat or repeating work best. Complex backgrounds, or an item that touches the image corners, may need manual cleanup.

• License: MIT for the app code, while the model weights fall under the CreativeML OpenRAIL-M terms of Stable Diffusion 1.5, which restrict some uses
• Cost to run: none named, and no Hugging Face token is needed for the public model
• Data it reaches: your prompt, and the model it names
• Not for: creating resource packs or reading files from paths you supply

On Hugging Face, 13 November 2026

• Likes: 1
• Built with: gradio
• License: mit

Links

• Space: https://huggingface.co/spaces/ArshPathan/TexuteCraft
• Files: https://huggingface.co/spaces/ArshPathan/TexuteCraft/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.