What it is
Udio is a website and app that makes music from a description. You type something like "a mellow acoustic folk song about a lighthouse keeper, female vocals", and about a minute later you have a short song, complete with singing, instruments and lyrics. You can write the lyrics yourself or let it write them. You can extend a song, remix it into another style, or change a single section. It works across a wide range of genres, from country and rock to electronic, choral music and jazz.
Udio was started by a small team of former researchers from Google DeepMind, Google's AI research lab: David Ding, who was its first chief executive, Conor Durkan, Charlie Nash, Yaroslav Ganin and Andrew Sanchez. The company behind it is officially called Uncharted Labs. It launched to the public on 10 April 2024, announcing 10 million dollars of investment led by the venture capital firm Andreessen Horowitz, with backing from the musicians will.i.am and Common, the producer Tay Keith, the music distributor UnitedMasters, and Instagram's co-founder Mike Krieger, among others. By late 2025, Andrew Sanchez was chief executive.
How does it work? Like other AI generators, Udio is built on deep learning: a system shown a very large number of recordings until it has learned, in a statistical way, how music tends to go. It learns what a country chorus usually sounds like, how drums differ between rock and hip-hop, and how a voice rises before a big final line. It does not paste together clips from existing songs. It produces new audio each time, which is why the same prompt gives you different songs on different tries. Which recordings it learned from turned out to be the most important question about Udio, and it is the reason most of this page is about courts and contracts rather than features.
The product changed quickly in its first eighteen months. Version 1.5, announced on 23 July 2024, improved sound quality, let people add a musical key to a prompt, let users upload their own audio to remix, and let them download "stems", the separate vocal, bass, drum and other tracks that producers use to rebuild a mix. Later features added style references and reusable voices.
Then, on 29 October 2025, Udio settled its lawsuit with Universal Music Group and switched off downloads the same day. At the time of writing, you can make and listen to songs on Udio, and share links to them, but you cannot download them, and the terms say you may use them only for personal, non-commercial purposes. Udio says it is building a new, licensed service, described further down. For a creator who needs music to publish, that changes everything about how Udio can be used.
Why it matters
It showed how good AI music had become. When Udio launched in April 2024, many listeners were startled by how human the singing sounded and how complete the songs felt. Along with its rival Suno, it made "type a sentence, get a song" an ordinary experience for millions of people, not a lab demo.
It became a test case for training on copyrighted music. The biggest open question in creative AI is whether a company may learn from other people's work without asking or paying. Udio's case moved further than most: from a lawsuit, to a settlement, to a partnership with the company that sued it. What happened next, including the downloads switched off overnight, is a preview of what licensed AI might look like for users.
It is a lesson about building on someone else's platform. Subscribers who had spent months making songs woke up to find they could no longer take them away. Whatever you think of the lawsuit, the practical lesson applies to every online tool: keep your own copies of anything you care about, and read the terms before you build a project on a service.
It raises the question of who owns an AI song. At the time of writing, Udio's terms say the company and its licensors own the output, not you. Even under older, more generous terms, the law gave limited protection. The US Copyright Office has said that material generated from prompts alone is not protected by copyright; the rulings are covered on the Midjourney page. Words or music you write yourself remain yours.
Real-world examples
"BBL Drizzy": an AI song in a rap feud (April and May 2024)
In spring 2024 the rappers Drake and Kendrick Lamar were trading diss tracks, songs written to insult each other. On 14 April 2024, only four days after Udio's launch, the comedian Willonius Hatcher, who performs as King Willonius, released "BBL Drizzy", a mock soul song poking fun at Drake. According to Billboard's account, Hatcher wrote the lyrics, taking his cue from a post by the rapper Rick Ross, and Udio generated the music and the singing voice.
On 5 May 2024 the producer Metro Boomin released an instrumental beat, also called "BBL Drizzy", built on a sample of Hatcher's track. Billboard reported that Metro did not know at first that the song he sampled had been made with AI, and called it the first notable example of AI sampling in mainstream hip-hop. Metro offered a prize for the best rap over his beat, and it spread quickly online. Udio later pointed to the song in its own court filing as an example of how people used its tool.
Credit is precise here. The joke and the words were Hatcher's. The sound was Udio's. The beat was Metro Boomin's. That mix of human and machine is exactly what copyright law is still sorting out.
The record labels sue Udio (June 2024 to 2026)
On 24 June 2024 the Recording Industry Association of America announced lawsuits on behalf of the three major label groups, Universal, Sony and Warner, against both Suno and Udio. The joint filing and the Suno case are covered on the Suno page. The case against Udio was filed in the federal court for the Southern District of New York and assigned to Judge Alvin K. Hellerstein. The labels said Udio had copied their recordings on a huge scale to train its AI without permission. To show it, they said they had prompted Udio with details of famous songs, such as the decade, genre and subject, and received results closely resembling recordings including "Dancing Queen" by ABBA and "All I Want for Christmas Is You" by Mariah Carey. They asked for damages of up to 150,000 dollars for each work infringed.
Udio answered on 1 August 2024. It argued that nobody owns a genre or a musical style, and it described its training as using existing recordings as data to analyse patterns in music, which it called a "quintessential fair use", the part of US law that permits some uses of copyrighted work without permission. In June 2025 an independent country musician, Tony Justice, filed a proposed class action in New York making similar claims on behalf of independent artists. In September 2025 the labels asked to add a new allegation: that Udio had "illegally scraped" recordings from YouTube by getting around YouTube's protections against downloading.
No court ever ruled on whether Udio's training was fair use. Universal and Warner settled instead, as the next example explains. Sony did not. In June 2026 the judge refused to let Sony add more than 30,000 further recordings to the original case, and on 20 July 2026 Sony filed a second lawsuit over those 30,117 recordings. At the time of writing, Sony's claims against Udio were still going.
The settlements and the download shutdown (October 2025 onwards)
On 29 October 2025 Universal Music Group and Udio announced that they had settled and signed licences for recordings and songwriting. Together they would build a new service, due in 2026, trained only on authorised and licensed music, where fans could create with the voices and songs of artists who choose to take part, and those artists would be paid. Neither side disclosed the money involved.
In a blog post the same day, Sanchez told users that downloads were unavailable from that day, calling it a significant sacrifice made "with a heavy heart". Paying subscribers were given extra monthly credits and a one-time bonus of 1,000 credits. The reaction was furious. Subscribers on Reddit and elsewhere said they had paid for a tool and lost their music without warning; some talked about suing. Late on 30 October Udio announced a one-off 48-hour window, which opened on 3 November, for everyone to download their existing songs. Billboard reported that songs downloaded in that window stayed under the old terms, which had let users own what they made. After it closed, songs stayed playable on Udio but could no longer be exported.
On 19 November 2025 Warner Music Group also settled with Udio and signed on to the new service. Udio went on to sign deals with Merlin, which licenses for many independent labels (January 2026), the publisher Kobalt (April 2026) and the distributor Believe. In May 2026 the trade press reported the new app's name, Starstruck: a mobile app for fans with modes to cover, remix and reimagine songs by artists who opt in, or to set your own lyrics to an artist's voice. Reports said it would be a "walled garden": what you make stays inside the app, you do not own it, and you cannot download it or put it on streaming services. As far as I could find, Starstruck had not launched publicly by late September 2026; check the current news.
Not everyone sees the deals as a win for musicians. In June 2026 the American Federation of Musicians, the US musicians' union, sued Universal and Warner, arguing that licensing its members' recordings to Suno and Udio was a new use that required paying the musicians who played on them. That case was also undecided at the time of writing.
An everyday scenario: the book trailer that couldn't be updated
Here is an illustrative scenario, not a real person. In summer 2025, Rosa, a self-published fantasy author, made a moody instrumental on Udio for her book trailer. She downloaded it, cut the trailer, and posted it. A year later she wants a longer version for her sequel. She logs in, finds the original, and extends it, but there is no download button, and the current terms say Udio's output is for personal, non-commercial use only. Her old trailer, made under the earlier terms, is still hers to use. The new version is not something she can publish. She makes the sequel's music with a tool whose current terms allow commercial use, and saves copies of every file she makes, everywhere.
Try it yourself
This exercise shows you what Udio can do today, and what it can't. At the time of writing, you can browse and listen to other people's songs on udio.com without an account, but making songs needs a free account. The free plan gave 100 credits a month and 10 a day, enough for three full-length songs of about two minutes a day. The paid plans, Standard at 10 dollars a month and Pro at 30 dollars a month, add more credits and editing tools, but no plan includes downloads. Prices, limits and terms change, so check them yourself. Allow about thirty minutes.
- Listen first. Go to udio.com and open the home page. Play three or four songs from the Spotlight section in genres you know well. Listen for what sounds convincing and what doesn't: the lyrics, the vocal tone, the ending.
- Read the rules before you make anything. Open the Terms of Service linked at the bottom of the site. Find the parts on downloading and ownership, and note the date at the top. That tells you what you can do with anything you make.
- Sign up for the free plan if you're happy with the terms. No card is needed for it at the time of writing.
- Write a descriptive prompt. In the create box, try something like this:
An upbeat 1960s-style soul song about a dog who steals socks, warm brass, female lead vocal, handclaps - Listen to both versions. Udio usually gives you more than one take. Compare them. Which follows your description more closely?
- Try your own lyrics. Choose the option to write custom lyrics and paste in four to eight lines of your own. Keep lines short and give each verse the same rhythm; it helps the melody fit.
- Change one thing at a time. Make the same song again with one change, such as "male vocal" or "slower tempo". This is the quickest way to learn what the tool responds to, the same iterative refinement that works for any AI tool.
- Notice what you can't do. Look for a download option: at the time of writing there isn't one. You can share a link, which lets friends listen on Udio.
- If you need music to publish, such as a podcast jingle, a video backing track or a book trailer, don't use Udio for it while its terms say personal and non-commercial use only. Compare the current terms of other options instead: Suno and ElevenLabs both offer music generation with commercial use on some paid plans at the time of writing, with their own conditions. Royalty-free music libraries, or a human composer, avoid the AI questions altogether.
Common mistakes and misunderstandings
"I made it, so I own it." At the time of writing, Udio's terms say that the company and its licensors own the output and that you may use it only for personal, non-commercial purposes. That is the opposite of how Udio worked before November 2025, which is why old guides and videos are misleading. Always read the current terms, not an article, including this one.
"I'll download it later." You can't, and that is the lesson of October 2025. Any online tool can change its terms or features overnight. If you make something you care about on any platform, download a copy while you're allowed to, and keep it somewhere you control.
Thinking the lawsuits proved Udio broke the law. They didn't. No court ruled on whether training Udio on the labels' recordings was fair use; Universal and Warner settled and Sony's case was still going at the time of writing. The settlements show that the companies chose to make deals, not what a judge would have decided.
Thinking the lawsuits are all over. The opposite mistake. Sony's two cases, the independent artists' class action and the musicians' union case were all undecided at the time of writing. The rules for AI music are still being settled.
Assuming an AI song can be copyrighted. In the United States, the Copyright Office has said that music generated from prompts alone is not protected, though lyrics or music you write yourself, and your own creative arrangement of the parts, can be. See the Midjourney page for the rulings, and ask a lawyer where you live about anything valuable.
Prompting with real artists' names. Asking for a song "by" a named singer, or pasting in someone else's lyrics, can break Udio's rules and, in some countries, the law. The licensed service promises artist voices only where artists have opted in. Until then, describe the sound you want ("raspy male vocal, 1970s Southern rock") instead of naming a person.
Believing the new Udio is the same tool with a licence. On the reports so far, Starstruck is built for fans playing with their favourite artists' songs inside an app, not for creators making original music to publish. If that's what you need, it may not be the tool for you even when it launches. Wait for the actual terms before deciding.
Mixing up Udio and Suno. They were founded around the same time, launched similar features and were sued on the same day, but their stories split in late 2025. Udio closed its doors to downloads; Suno's settlement with Warner kept downloads for paying users. Check which one an article is actually about.
Related elements
- Suno: Udio's closest rival, sued on the same day, which took a different path after settling with Warner.
- ElevenLabs: best known for AI voices, it also offers music generation.
- Midjourney: where the US Copyright Office's rulings on AI-made work are explained.
- Datasets: the collections of examples an AI learns from, at the heart of the labels' lawsuit.
- Pretraining: the stage where a model learns from huge amounts of data, and where the copyright questions begin.
- Deep Learning: the kind of AI that learned how music sounds by studying recordings.
- Iterative Refinement: changing one thing at a time until the song sounds right.
- Adobe Firefly: another creative tool whose pitch rests on training with licensed material.
- Guardrails: the filters that try to stop a generator copying real songs or voices.