Revolutionizing Video Creation: ByteDance’s X-Portrait 2 Technology

IT House reported on November 6 that single-image video driver technology requires only one still photo and a driver video to generate high-quality, “movie-level” video.

ByteDance’s intelligent creative team launches the latest single-picture video driving technology X-Portrait 2. This model retains the ID of the original image, captures and transfers expressions and emotions from subtle to exaggerated, simplifying existing motion capture, character animation and content creation processes.

Different from previous single image-driven methods that rely on facial key point detection, X-Portrait 2 builds an expression encoder model that can self-learn ID-independent movements from a large number of portrait videos through an end-to-end self-supervised training framework. Implicit representation.

Further combining this encoder with a powerful generative diffusion model produces smooth and expressive videos. After training on large-scale high-quality expression videos, X-Portrait 2 significantly outperforms previous technologies in terms of motion performance and ID retention.