Open-source SUNO is here! Free offline AI music generator

Key Concepts

• Heart Moola: A new open-source AI music generator.
• Sunno & Udo: Previously leading AI music generators, now facing licensing restrictions due to lawsuits from major record labels (Warner Music & Universal Music Group).
• Open-Source AI Music: The ability to generate music locally and offline without licensing limitations.
• Benchmarks: Comparative metrics used to evaluate the performance of AI models (Lyric Error Rate, Song Quality, Style Consistency, Audio Quality, Musicality, Structure Coherence).
• Hugging Face: A platform for sharing and downloading AI models and datasets.
• Virtual Environment (conda): An isolated environment for Python projects to manage dependencies.
• Triton: A library required for running the Heart Moola model.
• CFG Scale: A parameter controlling how closely the AI follows the provided prompt and lyrics.
• Apache 2 License: A permissive open-source license allowing commercial use.

The Rise of Open-Source AI Music: Heart Moola as a Sunno Alternative

The video focuses on Heart Moola, a newly released open-source AI music generator positioned as a viable, and potentially superior, alternative to existing options like Sunno and Udo. The core argument is that Heart Moola offers a return to the freedom and flexibility of early AI music generation, circumventing the increasing restrictions imposed by commercial platforms due to copyright concerns.

The Problem with Commercial AI Music Generators

Initially, Sunno and Udo allowed users to freely generate, download, and even commercially utilize AI-generated music. However, both companies faced lawsuits from Warner Music and Universal Music Group, alleging unauthorized training on copyrighted material. As a result:

• Sunno: Partnered with Warner Music Group, leading to licensing requirements and limitations on music generation.
• Udo: Disabled downloads entirely, restricting users to listening to music solely on the platform.

This shift towards partnerships with major labels introduces usage limitations and licensing requirements, diminishing the original appeal of these platforms. As the narrator states, “Sununo and Audio isn't really what it used to be.”

Heart Moola: A Promising Open-Source Solution

Heart Moola emerges as a solution to this problem, offering an open-source model that allows users to generate music locally and offline without these restrictions. Benchmarks presented in the video demonstrate Heart Moola’s performance is comparable to Sunno version 5:

• Lyric Error Rate: Heart Moola performs better than Sunno v5.
• Overall Song Quality, Style Consistency, Audio Quality, Musicality, Structure Coherence: Heart Moola scores close to Sunno v5.

This positions Heart Moola as a “pseudo level music generator which you can use offline,” a significant advantage over its commercial counterparts.

Demo Examples & Capabilities

The video showcases several demos generated using Heart Moola, highlighting its versatility:

• Taylor Swift Style (Acoustic Folk Country): Successfully replicated the style and incorporated provided lyrics with impressive accuracy. The narrator notes the output isn’t “production quality” but is significantly better than other open-source models.
• Modern Rock/Heavy Metal: Demonstrated the ability to generate energetic and powerful rock music, though still exhibiting some “muddy” audio quality.
• K-Pop (Dance Electronic): Produced a song with a strong K-Pop melody and even incorporated Vocaloid-style autotune.
• Spanish (Latin Chill Urban Cafe): Successfully generated a chill song with correct Spanish pronunciation.
• J-Pop (Energetic Electronic): Created a song with a Yosobi-like vibe.
• Hindi (Bollywood Upbeat Dance Indie Pop): Results were less successful, suggesting potential limitations in the model’s training data for Bollywood music.
• Instrumentals: Heart Moola struggles with generating high-quality instrumentals without vocals. The narrator advises it’s best suited for full songs with singing.

Installation & Setup: A Step-by-Step Guide

The video provides a detailed, step-by-step guide to installing and running Heart Moola locally:

1. GitHub Repository: Download the necessary files from the official GitHub repository (link in description).
2. Git Installation: Instructions are provided for installing Git if it’s not already present on the user’s system.
3. Python & Conda (Miniconda): The video recommends using Miniconda (a minimalist version of Anaconda) to create a virtual environment with Python 3.10. This isolates the project’s dependencies.
4. Virtual Environment Activation: Activate the created virtual environment using the command conda activate heartmoola.
5. Dependency Installation: Install required packages using pip install -r requirements.txt.
6. Model Download: Download the model weights from Hugging Face using specific commands (with adjustments for Windows users to avoid quotation issues). This involves downloading several large files (up to 16GB).
7. Triton Installation: Install Triton (version 2.1) if not already present, using a pre-built wheel for Windows to avoid lengthy compilation times.
8. Music Generation: Run the music generation script (run_music_generation.py) with optional parameters to customize the output (e.g., max_audio_length_ms, top_k, temperature, cfg_scale).

Licensing & Future Implications

Heart Moola is released under the Apache 2 license, granting users significant freedom, including commercial use. The narrator expresses optimism about this open-source approach, stating, “God bless this team for making this music generator and kind of saving AI music from being, you know, consumed by restrictions and copyright issues from these big label companies.”

Conclusion

Heart Moola represents a significant development in the AI music landscape, offering a powerful, open-source alternative to increasingly restricted commercial platforms. While not yet perfect, its performance is remarkably close to industry leaders like Sunno, and its freedom from licensing limitations makes it a compelling option for creators and enthusiasts alike. The detailed installation guide provided in the video empowers users to leverage this technology locally and explore the potential of AI-driven music generation.