AI that creates deepfake video from a single photo

Microsoft has introduced an artificial intelligence model with the capacity to break new ground in the field of artificial intelligence. The new tool can produce realistic deepfake videos based on a single photo and audio recording.

Microsoft has unveiled one of the most impressive (even scary) artificial intelligence tools we've ever seen. The software giant revealed an artificial intelligence model called VASA-1 that can create realistic deepfake videos based on a single photo and audio recording.

The new neural network can mimic the movements and emotional expressions of the human face with incredible accuracy. The videos produced in this way look extremely natural and believable. Experts have called it a "scary machine" for deepfake videos.

VASA-1 uses a hidden face space to generate facial dynamics and head movements. Microsoft states that this method is significantly improved compared to previous techniques and gives more realistic results. The resulting studies also confirm this.

Closed for general use for now

The algorithm supports online video rendering at a resolution of 512x512 pixels and a frame rate of 45 frames per second. This makes it possible to interact with the model and chat in real time with realistic avatars. Microsoft currently has no intention of releasing VASA-1 as a commercial product. The company wants the new model to be used as a research tool for now.

Apparently to allay concerns, the company assures that VASA-1 will not be in the hands of users anytime soon. This model could open up new possibilities for the film and gaming industry, be used to develop virtual assistants and customer service applications, and even be useful in the education and healthcare sectors.