StablePackGen

StablePackGen is an AI-powered audio sample pack generator that creates complete, ready-to-use sample libraries using Stability AI's Stable Audio Open model.

Features

• Complete Sample Packs: Generate entire collections of audio samples organized by category
• Precise Duration Control: Automatically trim samples to exact lengths for perfect timing in your productions
• Category Organization: Samples are sorted into logical categories (kicks, snares, hats, percussion, synths, pads, FX)
• High-Quality Audio: Leverages Stable Audio Open for professional-quality sound design
• LLM-Guided Generation: Uses Claude to help craft detailed prompts for each sample

Requirements

• Python 3.8+
• PyTorch
• CUDA-capable GPU (recommended)

Installation

1. Clone this repository:

git clone https://github.com/yourusername/StablePackGen.git
cd StablePackGen

2. Install dependencies:

pip install -r requirements.txt

3. Set up your Anthropic API key:

export ANTHROPIC_API_KEY='your-api-key-here'

Usage

To generate a complete sample pack:

python stablepackgen.py

By default, this will create a techno sample pack with the following categories:

• Kicks
• Snares
• Hi-hats
• Percussion
• Synths
• Pads
• FX

All samples will be saved in their respective category folders in the current directory.

Customization

Modifying Sample Types

To customize the types of samples generated, edit the generate_sample_pack_plan() function in the script. You can:

• Add new sample categories
• Modify existing prompts
• Change duration settings
• Customize output file paths

Changing Genres

The system is currently set up for techno, but you can easily modify the prompts for other genres like:

• House
• Ambient
• Trap
• Lo-fi
• Drum & Bass

Generation Settings

You can adjust the generation settings in the agent_step() method:

• cfg_scale: Controls how closely the generation follows the prompt (higher = more faithful)
• steps: Number of diffusion steps (higher = better quality but slower)

How It Works

StablePackGen uses a combination of:

1. LangGraph Workflow: Orchestrates the sample generation process
2. Claude AI: Helps craft detailed prompts for each sample type
3. Stable Audio Open: Generates high-quality audio based on text prompts
4. Audio Processing: Processes and trims samples to exact durations

Requirements.txt

torch
torchaudio
einops
langgraph
langchain-core
langchain-anthropic
anthropic
stable-audio-tools
librosa
soundfile
numpy
typing-extensions

License

MIT

Acknowledgments

• Stability AI for the Stable Audio Open model
• Anthropic for the Claude AI model
• LangChain for the LangGraph workflow framework