Imogen Heap's ai.Mogen: what an artist-controlled AI voice looks like, and what it costs fans

Most artists meet AI as a defendant meets a lawsuit: their voice turns up in a model they never agreed to, and they spend money trying to get it out. Imogen Heap went the other way. She built a model of her own voice, licensed her style to one AI company on terms she helped set, and is reported to take 70% of each sale. It is the clearest working example we know of an artist-controlled AI deal. It is also worth reading closely, because "artist-owned" turns out to mean something narrower than it sounds.

What has Heap built?

Three pieces, which are easy to blur together.

ai.Mogen, the voice. On the page for her 2025 song "I AM ___", Heap describes ai.Mogen as a spoken and singing voice model that began as a chatbot more than six years ago, built to help her cope with incoming messages. For the song's "Aftercare" section she sang the part herself, then transferred it into a model she trained on her own voice, and says she spent hundreds of hours making it sound right. She writes that, to her knowledge, she is the first musician to intentionally train and use a private model of her singing voice for commercial use. That is her claim; we haven't been able to test it. The same page credits the voice work to three models: Jen, Voiceswap and Stability AI.

StyleFilters, the licensing deal. Jen, an AI music platform, unveiled StyleFilter at Web Summit on 14 November 2024 with Heap as its first artist. The demo version was trained on two of her releases, "What Have You Done To Me?" and "Last Night of an Empire", and Jen said the technique can learn a style from a single song. Heap called it "song as a service", Decrypt reported at the time. Jen's StyleFilters page now lists five Heap filters, adding "Headlock", "Goodnight and Go" and "Just for Now". Each is built from "a fully-licensed track, directly from rights holders". Users pay from $4.99 for 90 minutes of low-strength generation or $7.99 for high strength. They own what they make, but may not use it to train other AI models.

Auracles, the permission layer. Heap leads Auracles, which the Jen press release described as a non-profit focused on data provenance that lets artists control and permission their information. The same release said the AI version of her voice was "authorized for use through Auracles". On the "I AM ___" page she adds that Auracles is now linked to the Do Not Train registry, so musicians can opt out of model training through their Auracle ID. In a SoundCloud interview recorded at Web Summit 2025, she described it as the way she grants platforms permission to train on her work, including her own voice for ai.Mogen.

What does the money look like?

Rolling Stone reported in April 2025 that Jen pays artists 70% of the revenue each time someone uses their StyleFilter. Jen's own page doesn't give a percentage. It says only that each use "generates revenue for the artists whose work inspires yours". So the 70% rests on the press report, not on published terms.

If 70% applies to the full list price, which neither side has said, Heap gets about $3.49 from a $4.99 purchase and about $5.59 from a $7.99 one. That is our arithmetic, before any app-store or payment fees. In the same Rolling Stone piece Heap said she had made $12,000 from 400 stem downloads at $30 each, far more than streaming paid her. On those numbers, one stem sale is worth roughly eight or nine low-strength StyleFilter sales.

She also explained why she said yes: she is pro-AI, she told Rolling Stone, only with people whose ethics she shares, and if artists stay out, the field is left to people doing it purely for profit.

What we think

Heap's model gets three things right that most AI music deals get wrong. Consent comes first, not after a takedown. The training data is named, down to the song titles. And the artist's share is high and was reported publicly. Compare BandLab's purchase of Aiode, where session players get a share of the tokens spent on their models but nobody has published how big it is. A 70% figure an artist can check is worth more than a vague "revenue share" in a press release.

But "artist-owned" needs to be read carefully. The "I AM ___" credits list three companies' voice models behind ai.Mogen. Heap owns the permission and the voice. She doesn't own the infrastructure that renders it. If a vendor changes its terms, pivots or shuts down, the artist's control is only as strong as the contract. HYBE's AI voice company Supertone shows the risk: it was dissolved with no public word on what happens to voice models built from artists. Any artist copying Heap should get three things in writing: deletion rights over the model, a ban on using it to improve the vendor's base model, and the right to take the trained model with them if the deal ends.

The second limit is scale. When we checked, Jen's StyleFilters page showed only Heap's five filters, although the 2024 launch was billed as the first in a series of artist and producer collaborations. Heap brings a decade of rights-tech work, her own permissions platform and a recent viral hit. A mid-level artist has none of that leverage, and a 70% share of very few sales is still very little. The model works. What's missing is a cheap, standard way for ordinary artists to reach the same terms.

That is why the permission layer matters more than the voice. A registry where an artist records who may train on what, and that platforms are expected to check, is the piece that could scale. But it only works if the big generators actually check it before they train.

The practical lesson for working artists is smaller and more immediate. Your voice, your stems and your session files are the training data. Keep clean, dated originals, whether in a hardware rig or a browser DAW such as Veena, so you can prove what is yours and license it on your own terms when someone asks.

Sources

• I AM ___, Imogen Heap
• Jen showcases StyleFilter technology with Imogen Heap at Web Summit, press release, 14 Nov 2024
• StyleFilters, Jen
• Imogen Heap Just Went Viral With 'Headlock', Rolling Stone (via Rolling Stone India), 25 Apr 2025
• Imogen Heap interview, Sound Advice, SoundCloud, 11 Dec 2025
• Musician Imogen Heap Is Offering AI-Powered 'Songs as a Service', Decrypt, 19 Nov 2024