Moodles is a new research experience from Samsung Labs with the ability for researchers to create realistic high-definition deepfakes using a single source photo.
The Samsung Labs team claims they can create portraits from a single image, such as the Mona Lisa, to create life-like deepfakes, as seen in the video below. Deepfakes look absolutely real, and when I say real, they look absolutely real, with realistic head and neck movements and rich facial expressions.
The technology was developed by combining the source photo and a driver’s speed. The driver here is another whose motion is applied to the source image to create the deepfake.
Our training is quite easy. At each step, we sample two random frames from our dataset: source and driver. The abstract of our model explains why the steering wheel, the position of the head and the expression of the face are imposed on the appearance of the frame, which is the reflection of the input image.
The training cues are derived from training episodes where the source and pilot images are from the same video, so our models are in order to match the pilot image.
The research paper titled Megaportraits: One-shot Megapixel Nylon-Avatars. Learn more here.
Google image credit.
You can learn more about Samsung Labs.