MusicGen Models: Small, Medium, Large, Melody, Stereo Explained!

These models have revolutionized the way we generate high-quality music samples based on text descriptions or audio prompts. In this article, we will explore the various MusicGen models, their capabilities, and how they are changing the landscape of music generation.

Introduction to MusicGen

MusicGen is a great text-to-music model that has introduced a new dimension to the creative process of music composition. MusicGen uses the power of artificial intelligence and Transformer models to craft music directly from textual inputs.

It is a single-stage auto-regressive Transformer model that has been trained extensively to generate remarkable musical compositions.

Here, we’re discussing about the MusicGen AI Models.

MusicGen 6 Models

MusicGen offers a range of models, each with its own unique capabilities and applications.

MusicGen Small300MGeneral32kHz EnCodec tokenizer, 4 codebooks, parallel codebook generation
MusicGen Medium1.5BGeneral32kHz EnCodec tokenizer, 4 codebooks, enhanced complexity
MusicGen Large3.3BGeneral32kHz EnCodec tokenizer, 4 codebooks, highest quality
MusicGen Melody1.5BMelody-focused32kHz EnCodec tokenizer, 4 codebooks, melody emphasis
MusicGen Stereo Large3.3BStereophonic32kHz EnCodec tokenizer, 4 codebooks, stereophonic sound
MusicGen Stereo Melody3.3BStereophonic & Melody32kHz EnCodec tokenizer, 4 codebooks, melody + stereo

Let’s dive into the various MusicGen models:

1. MusicGen Small – 300M

The MusicGen Small model is the entry-level version, yet it packs a punch with its 300 million parameters. This model utilizes a 32kHz EnCodec tokenizer with 4 codebooks sampled at 50 Hz. What sets it apart is its ability to generate all 4 codebooks in a single pass, eliminating the need for a self-supervised semantic representation.

This efficiency enables it to predict codebooks in parallel, resulting in just 50 auto-regressive steps per second of audio. Perfect for those looking to experiment with music generation.

2. MusicGen Medium – 1.5B

The MusicGen Medium model boasts 1.5 billion parameters, providing greater complexity and richness in generated music. Similar to the Small model, it employs the 32kHz EnCodec tokenizer and 4 codebooks sampled at 50 Hz.

Its enhanced capacity allows for more intricate and captivating compositions.

3. MusicGen Large – 3.3B

For those who demand nothing but the best, the MusicGen Large model shines with its colossal 3.3 billion parameters. This model embodies the pinnacle of text-to-music generation, delivering the highest quality music samples.

Like its smaller counterparts, it uses the 32kHz EnCodec tokenizer and 4 codebooks sampled at 50 Hz. The sheer scale of this model ensures unparalleled musical complexity and sophistication.

4. MusicGen Melody – 1.5B

Audiocraft introduces MusicGen Melody, a specialized model within the MusicGen family. With 1.5 billion parameters, it offers a unique focus on generating melodious compositions.

This model is perfect for those who want to emphasize the melodic aspect of their music. Like other MusicGen models, it excels in generating music without the need for a self-supervised semantic representation.

5. MusicGen Stereo Large – 3.3B

The MusicGen Stereo Large model brings stereophonic capabilities into play. This model has been fine-tuned for stereo sound and features 3.3 billion parameters, ensuring an immersive and spatial audio experience.

It uses two separate audio channels, creating a multidimensional listening experience that adds depth and direction to your compositions.

6. MusicGen Stereo Melody Large – 3.3B

Combining the best of both worlds, the MusicGen Stereo Melody Large model is a powerhouse of creativity. It inherits the stereophonic capabilities of the Stereo Large model while maintaining a focus on melody.

This combination results in rich, immersive, and melodious music compositions.


MusicGen models have ushered in a new era of music generation, empowering musicians, composers, and AI enthusiasts to explore the limitless possibilities of text-to-music conversion. Whether you’re starting with the Small model or diving headfirst into the world of Stereo Melody, MusicGen offers a range of options to cater to your creative needs.

With the power of AI and Transformer models, the future of music composition has looked brighter. Explore, create, and let the music flow with MusicGen models.

Leave a Comment