Photo credit: Disney Research
Deepfakes have been quite popular among internet users, as many hobbyists have been face swapping many celebrities in famous movies. Disney Research has developed an algorithm for fully automatic neural face swapping in images and videos. In other words, this is the first method capable of rendering photo-realistic and temporally coherent results at megapixel resolution, making it suitable for the big screen. Read more for a video and additional information.
This was made possible using a progressively trained multi-way comb network and a light- and contrast-preserving blending method. When compositing the generated expression onto the target face, the blending strategy is adapted to preserve contrast as well as low-frequency lighting. Lastly, a refinement strategy was incorporated onto the face landmark stabilization algorithm to achieve temporal stability, crucial for high-resolution videos.