Google scholar Enric Corona and his team have developed a tool called VLOGGER that can generate high-resolution videos of people talking based on a single photograph. The tool can animate the videos to accurately depict facial expressions and body movements, making it an "avatar" of high fidelity. The team suggests that VLOGGER could have applications in helpdesk avatars, online communication, education, and personalized virtual assistants. However, it could also be potentially misused to create deepfakes or manipulated videos.

VLOGGER combines multi-modality, large language models, and diffusion techniques to create realistic videos. The team trained a neural network to associate audio with video frames to synchronize the movement and expressions. The tool can be further personalized through fine-tuning and offers control over features like blinking and mouth movements. While VLOGGER provides a way to manipulate high-resolution videos, concerns about potential misuse and the need for detecting deepfakes persist.