MusicGen Large
MusicGen Large is Meta's highest-capacity model in the MusicGen family, released as part of the AudioCraft framework. It generates high-quality mono and stereo music conditioned on text descriptions and optional audio continuations, producing tracks with rich instrumentation, better temporal coherence, and stronger adherence to genre prompts compared to the smaller variants.
The large variant requires more compute than medium or small but delivers the most musically coherent output, making it the preferred choice when generation quality is the primary concern. It is open-weights and integrates directly with the AudioCraft library, supporting both research and production creative workflows.
Key Features
High-quality mono and stereo music generation from text prompts
Audio continuation mode for extending or completing musical passages
Rich multi-instrument composition with strong genre adherence
Open-weights model available via AudioCraft and Hugging Face
Configurable generation duration for short clips to extended tracks
Melody conditioning to guide harmonic structure from a reference audio
Ideal Use Cases
Generating full-quality music beds for professional video production
Extending or completing partial musical compositions
Research into neural music generation and audio-language alignment
Producing royalty-free music demos for licensing exploration
Creating diverse training data for downstream audio models
Example Prompts for MusicGen Large
Technical Specifications
| Provider | MusicGen |
| Category | Audio |
| Modality | Text -> Audio |
| License | CC BY-NC 4.0 (non-commercial) |
Frequently Asked Questions
Try MusicGen Large now
Start using MusicGen Large instantly — 100 free credits, no credit card required. Access 750+ AI models through one platform.