Video2StyleGAN: We present Video2StyleGAN, a video editing framework capable of generating videos from a single image. Our framework can take a driving video and transfer its global and local information to a reference image. The rotation and translation of the head are derived from the co-driving frames.

Reference Image

Reference Image

Reference Image

Reference Image

Reference Image

Comparisons with alternate methods: Comparison of our method with baseline and other state-of-the-art methods trained on video data and keypoints. Our method generates videos at high resolution (1024 x 1024) and does not require training on videos. Notice the quality of high frequency details that are preserved by our method.