An advanced AI-powered tool that generates Minecraft textures and models using deep learning. This project uses PyTorch to train and run generative models that can create custom Minecraft-compatible textures and block models from text prompts.
- Texture Generation: Create 16x16 Minecraft-compatible textures from text descriptions
- Model Generation: Generate JSON models for Minecraft blocks and items
- Training Pipeline: Custom training system for fine-tuning the generators
- Minecraft Compatibility: All outputs are automatically formatted to work with Minecraft's resource pack system
- Clone the repository:
git clone https://github.com/yourusername/minecraft-ai-generator.git cd minecraft-ai-generator
- Install dependencies:
`` pip install torch torchvision pillow numpy
``
python src/minecraft_ai_generator/main.py --generate
Then follow the prompts to enter your texture descriptions. For example:
- "A weathered oak plank texture"
- "Smooth polished granite"
- "Glowing ancient debris"
- Prepare your dataset in the following structure:
dataset/ ├── textures/ │ ├── wood_planks.png │ ├── stone_brick.png │ └── ...
- Start training:
python src/minecraft_ai_generator/main.py --train --dataset dataset --epochs 100
src/minecraft_ai_generator/main.py- Main entry point and CLI interfacetexture_generator.py- Texture generation modelmodel_generator.py- 3D model generationtraining.py- Training pipeline and dataset handling
- Uses a GAN-based architecture
- 16x16 RGB output
- Conditional generation based on text prompts
- Built-in Minecraft compatibility checks
- Transformer-based architecture
- Generates Minecraft-compatible JSON models
- Includes validation for Minecraft's model format
- Supports custom display settings
- Custom dataset loader for Minecraft textures
- Automatic labeling based on texture names
- Checkpoint system for model saving
- Progress tracking and logging
- Latent dimension: 100
- Multiple deconvolutional layers
- BatchNorm and LeakyReLU activations
- Final Tanh activation for [-1, 1] output range
- Batch size: 32
- Learning rate: 0.0002
- Adam optimizer with β1=0.5, β2=0.999
- Checkpoints saved every 10 epochs
- Python 3.8+
- PyTorch 1.8+
- PIL (Pillow)
- NumPy
- CUDA-capable GPU (optional, but recommended for training)
[Your chosen license]
Contributions are welcome! Please feel free to submit a Pull Request.
- Training requires significant GPU resources
- Generation time can vary based on hardware
- Limited to 16x16 textures currently