AI Music Generation

AI music generation refers to the use of machine learning to create music, ranging from short instrumental loops to complete songs with synthesised vocals and lyrics. Modern systems typically accept a natural-language prompt describing a genre, mood, instrumentation, or theme, and produce finished audio in seconds. The field sits within generative AI alongside text, image, and video generation, and it has moved quickly from research demonstrations to consumer products used by millions.

Early research systems such as Google's MusicLM and Meta's MusicGen showed that models could generate coherent music from text descriptions. They were followed by consumer platforms, most prominently Suno and Udio, which generate full songs including vocals from a short prompt, and Stability AI's Stable Audio for instrumental and sound-design work. Voice and audio companies such as ElevenLabs have extended adjacent capabilities in synthetic speech and sound.

How the technology works

AI music systems generally rely on two families of model. Transformer-based models treat audio as a sequence of discrete tokens, often produced by a neural audio codec that compresses sound into compact representations, and generate music token by token much as a language model generates text. Diffusion models, by contrast, start from random noise and iteratively refine it into a coherent audio waveform or spectrogram, conditioned on the text prompt. Many production systems combine these approaches and add separate components for lyrics, melody, and arrangement. The models are trained on large collections of recorded music, learning statistical patterns of rhythm, harmony, timbre, and song structure.

Copyright and legal disputes

The central controversy in AI music generation concerns the data used to train these models. In June 2024 the Recording Industry Association of America filed landmark copyright lawsuits against Suno and Udio, alleging that the companies had used vast quantities of copyrighted sound recordings for training without authorisation. The dispute has produced a mix of settlements and continuing litigation. Universal Music Group reached a licensing settlement with Udio in October 2025, and Warner Music Group settled with Suno in November 2025, with both arrangements moving toward licensed catalogues, opt-in mechanisms for artists, and per-generation royalties. Other claims remained active into 2026, with a summary-judgment hearing in the Suno case expected in mid-2026. In May 2025 the United States Copyright Office issued guidance suggesting that training on copyrighted works to produce content that competes with them may fall outside fair use, particularly where access was unauthorised.

Uses and concerns

AI music tools are used for background music in videos and games, rapid prototyping by songwriters, advertising, and personal creative projects. Alongside the copyright questions, the technology raises concerns about the economic impact on working musicians, the disclosure of AI-generated content, and the use of artists' vocal likenesses without consent, an issue highlighted by early viral imitations of well-known performers.

| Tool | Developer | Focus | |------|-----------|-------| | Suno | Suno | Full songs with vocals | | Udio | Udio | Full songs with vocals | | MusicGen | Meta | Instrumental, open model | | Stable Audio | Stability AI | Instrumental, sound design |

References

1. Agostinelli, A., et al. (2023). MusicLM: Generating Music From Text. Google Research.
2. RIAA. (2024). Record Companies Bring Landmark Cases for Responsible AI Against Suno and Udio. riaa.com.
3. United States Copyright Office. (2025). Copyright and Artificial Intelligence, Part 3: Generative AI Training. copyright.gov.
4. Billboard. (2026). AI Music Timeline: From Fake Drake to Suno and Udio Label Settlements. billboard.com.