MusicGen & AudioGen — AI Music Generation

A complete, code-first guide to Meta's AudioCraft library for programmatic AI audio generation. Generate original music from text prompts, create realistic sound effects, condition music on reference melodies, and build audio pipelines — all running locally on your GPU with no external API dependency.

Covers the full AudioCraft stack: • MusicGen: Text-to-music (300M to 3.3B parameter models, stereo support) • AudioGen: Text-to-sound effects (thunderstorms, crowds, ambience) • MusicGen-Style: Reference-based style transfer • EnCodec: High-fidelity neural audio compression

Includes working Python code for every workflow: • Single and batch generation • Melody-conditioned generation • Audio continuation (extend an existing track) • Style-conditioned generation from a reference file • Memory optimization for limited VRAM • A complete Gradio demo ready to deploy

Best for: agents building game audio pipelines, content creation tools, podcast production, video game soundtracks, sound design automation, or any product that needs AI-generated audio without per-request API costs.