Agent Atlas

The Builders

Imogen Heap

Wrote, sang and produced I AM ___, and runs parts of her own voice through a model she trained, ai.Mogen.

Sing it yourself, swap only the timbre, and say plainly which parts the model touched.

Steal this

  • Sing every part yourself, then change only the sound of the voice.
  • Say which seconds the model touched. Her statement does.

The lesson and the takeaways are our summary. Everything below is theirs, with where they said it.

The setup

Factual
Machines

Her phone, for the footage in one section of the video.

Software

ai.Mogen for the Aftercare vocal. Stable Diffusion on her phone footage, then a frame-by-frame human edit.

Models

ai.Mogen, which began as a chatbot. Stable Diffusion for one section of the video.

As of November 24, 2025. Setups change; follow the links.

How they work

Factual
  1. Her statement says the song and the video were not created by AI. Aftercare is her singing with the timbre swapped to ai.Mogen, then hundreds of hours making it sound right. (Imogen Heap)
  2. For one section of the video, her phone footage went through Stable Diffusion and was then re-edited pretty much frame by frame. (Imogen Heap)
  3. She says she is, to her knowledge, the first musician to train a private singing model for commercial use. (Imogen Heap)
  4. On 24 Nov 2025 she told Billboard she sang every part and put it through her model, and that it took four years and hundreds of hours. (Billboard)