The process of fine tuning a Stable Diffusion model to generate Minecraft character previews is detailed here: Digging into Stable Diffusion-Generated Minecraft Skins.
This codebase further refines that model, and allows for in-game skins to be generated using only a text prompt. The finetuned model for doing so is available on HuggingFace as monadical-labs/minecraft-skin-generator. The model development process is described here in a subsequant blog post: Even More Skin in the Game: Digging Deeper Into Stable Diffusion-Generated Minecraft Skins.
- Clone the minecraft_skin_generator GitHub repository onto your computer:
$ git clone git@github.com:Monadical-SAS/minecraft_skin_generator.git
$ cd minecraft_skin_generator
- Install pipenv, if it has not been installed already:
$ pip install pipenv
- Start up a new pipenv environment and install any dependencies:
$ pipenv shell
$ pipenv install
- Skins can be generated using the minecraft-skins.py script in the bin directory:
$ python bin/minecraft-skins.py output-skin.png "A man in a purple suit wearing a tophat"
- The output skin will be saved into "output-skin.png" and should be similar to this:
- The skin can then be imported directly into Minecraft:
Cory Spencer cory@monadical.com