The state of AI music right now
Walk into any session in Nashville or any bedroom studio in Brooklyn and the same two names come up: Suno and Udio. Suno, founded in Cambridge, leads on vocals and overall song quality. Its latest model handles lyrics, harmonies, and full vocal performances better than anything else on the market. Udio counters with sharper instrumental fidelity and an inpainting editor that lets you regenerate a specific section without touching the rest of the track.
The gap between the two has narrowed, but the legal picture shifted faster than the technology. Universal settled with Udio and Warner settled with Suno in late 2025, which sounds like good news until you read the fine print. The American Federation of Musicians filed an amended lawsuit against both majors in July, arguing that recordings made by union members were licensed to these AI companies without compensation or credit. Meanwhile, a group of artists including Jason Isbell filed suit against Suno over identity copying. If you plan to release AI-assisted music, you are entering a space where the rules are still being written in courtrooms.
The scale of the shift is hard to ignore. Deezer recently reported that AI-generated music now makes up more than half of daily uploads on its platform, the first time machine-made tracks outnumbered human ones. That flood of content is exactly why platforms, distributors, and performing rights organizations are tightening their policies.
What the major tools actually deliver
| Tool | Price range | Best for | Strengths | Watch out for |
|---|
| Suno | Free tier; Pro $8-$10/mo; Premier $24-$30/mo | Full songs with vocals, fast workflow | Best-in-class vocals, stem and MIDI export on paid plans, large community | Free tier is non-commercial; active litigation |
| Udio | Free tier; Standard $10/mo; Pro $30/mo | Instrumental quality and surgical editing | Inpainting editor, 48 kHz stereo output, long session extensions | Downloads disabled during the Universal transition |
| ElevenLabs Music | Free tier; Starter $5; Creator $22; Pro $99 | Licensed, commercially clean tracks | Trained on licensed data, studio-grade output, broad commercial clearance | Self-serve plans exclude film, TV, and game use |
| Stable Audio 2.5 | API and enterprise licensing | Producers needing instrumental beds | Fully licensed training dataset, prompt-driven multi-part structure | Instrumental focus, no lead vocals |
| Google Lyria 3 | Bundled in Gemini and Vertex AI | Creators already in the Google ecosystem | SynthID watermark for provenance, image-to-music generation | Tracks capped at around three minutes |
Prices reflect current monthly billing in US dollars. Annual plans at most of these services bring the cost down further, and all of them offer a free tier that is genuinely useful for testing the waters.
Matching the tool to your situation
For YouTube, podcasts, and social content
If you need background music, intro stings, or a theme song for a channel, Suno's free tier is enough to experiment. The moment you want to monetize, upgrade to a paid plan. Commercial rights only attach to downloads made within a paid plan's monthly allocation, so trial downloads and remixes do not count. ElevenLabs Music is the safer bet if your video could end up in sponsored content, because its licensing covers broad commercial use through partnerships with Merlin and Kobalt. A creator in Austin who runs a weekly true-crime podcast told me she switched from stock music libraries to ElevenLabs for its instrumental beds and has not looked back.
For independent artists planning a real release
This is where the stakes rise. The US Copyright Office still refuses to register works produced purely by a machine with no meaningful human input. A raw prompt and nothing else will not get you a registration. But the picture changed for partially AI-assisted work. ASCAP, BMI, and SOCAN now accept registrations for songs that combine AI-generated elements with meaningful human authorship, and they treat royalty rates for those tracks the same as fully human works.
What counts as meaningful human authorship? The standard is qualitative. You need a paper trail: lyric drafts, notes on structure, prompts with timestamps, and evidence of editorial decisions between AI passes. One songwriter in Nashville described her process this way: she writes the lyric, feeds it to Suno with detailed arrangement notes, then pulls the stems into her DAW, re-records the vocals herself, and rewrites the bridge entirely. That track gets registered. A track generated from a single prompt does not.
For hobbyists and curious listeners
Riffusion, now folded into Producer.ai under Google, remains the best genuinely free option. It produces full songs with decent style control, and the public beta costs nothing. Udio's free tier is also worth exploring if you care about instrumental texture, though the download situation remains in flux. If you are making music purely for fun, none of the legal complexity applies to you yet, but keep receipts of what plan you used in case a track surprises you and becomes popular.
Keeping your AI tracks legally clean
Treat this as a five-step checklist before you upload anything anywhere.
Document your creative process as you work, not after. Keep the lyric drafts, the rejected variations, the notes on why you chose one take over another. This documentation is what defends a registration if it gets challenged.
Disclose AI use when you register and when you distribute. The Copyright Office requires you to exclude AI-generated content beyond de minimis amounts and describe your human contribution. Most distributors have similar disclosure policies now. Filing without disclosure risks cancellation of the registration.
Know which plan your download came from. Suno requires Pro or Premier for commercial use. Udio downloads are currently unavailable during the Universal transition, so commercial release on Udio is not possible right now. Check each platform's current terms before you commit to a release date.
Clear anything borrowed. If your AI track includes a sample, a recognizable voice, or an imitation of a real artist, that needs its own permission. The recent lawsuit from artists alleging identity copying shows this is not a theoretical risk.
Think about distribution before you think about royalties. Streaming fraud is a real enforcement target. A North Carolina man was sentenced to eighteen months in prison this year for uploading thousands of AI songs and using bots to generate $8 million in fake royalties. Legitimate AI-assisted music is fine; fabricated streams are not.
Where to start this week
Open a free account on Suno and generate ten songs. Pick the best one, then spend an afternoon rewriting the lyrics and re-recording any vocal parts yourself. Export the stems, note your edits, and register the finished track with your PRO. That workflow takes you from curiosity to a commercially usable, registrable song without much money changing hands.
The tools improve every few months, but the fundamentals stay the same: human input, honest disclosure, and a clear record of what you actually did. Start with those and the technology becomes a collaborator rather than a gamble.
研究の知恵