Disney researchers have demonstrated what they described as the first photorealistic deepfake at megapixel resolution.
The maximum-resolution videos the scientists could create using the open source DeepFakeLab model were just 256 x 256 pixels in size, while their latest model can generate video with 1,024 x 1,024 resolution.
The model can swap the semblance of two individuals while maintaining the target's facial expressions, although videos indicate it can only generate deepfakes of well-lit individuals looking more or less directly at the camera at present.
The researchers said their goal is to produce deepfakes of sufficient quality for commercial projects, as a less time-consuming and more affordable alternative to traditional visual effects for film.
Deepfakes need far less supervision once the original model is assembled, and can produce video in just hours.
From The Verge
View Full Article
Abstracts Copyright © 2020 SmithBucklin, Washington, DC, USA
No entries found