ID 06 · 2022–2023kedi.ai — AI Face Reenactment & Voice Cloning
Software that transfers motion from video onto a still photograph, bringing a portrait to life and giving it a voice.
kedi.ai uses a machine-learning model trained by monolab to transfer motion from video onto photographs, animating a still image and giving it a voice.
For the face animation, we trained an autoencoder model with transfer learning on 400,000 videos from the CelebV-HQ dataset, raised the resolution to 512×512 and added a masked upscaling module built on CodeFormer and GFPGAN. For voice cloning, we trained an acoustic model and a vocoder.
kedi.ai later animated the AI portraits of the Boyner Caddebostan art collection, where visitors chose the emotions on the faces.
In May 2023, HalkTV's morning programme Yeni Bir Sabah used photo animation and voice cloning made with kedi.ai in its report “Sahte videolar nasıl yapılıyor?” (How are fake videos made?).