The image does not belong to buddystech.com
Microsoft Research has released VASA-1, an AI model that can turn a single picture and a current audio track²² into a synced animated movie of a person talking or singing. This is a huge step forward.
A system called VASA-1, which stands for “Visual Affective Skills Animator,” makes talking faces for computer characters that look and act like real people. VASA-1 can make lip movements that are perfectly in sync with speech audio from a single still picture and a speech audio clip. It can also capture a wide range of face variations and natural head movements².
One of the main new ideas is a model for creating head movements and facial dynamics that works in a face latent space². One of the most important things about VASA-1² is that it uses movies to create a face latent space that is both expressive and clear.
It is said that Microsoft’s VASA-1 can make movies with a size of 512×512 pixels at up to 40 frames per second and very little latency². In other words, it might be able to be used for real-time tasks like videoconferencing².
The VASA system looks at both a still picture and a speech audio clip² using machine learning. Then, it can make a realistic movie with exact head movements, facial emotions, and lip sync to the sound². It doesn’t copy or fake sounds; instead, it uses an existing audio input that can be spoken or recorded just for that purpose².
Microsoft says that the model is much more realistic, creative, and efficient than earlier speech animation methods². It does look like an improvement over the single-image moving models that came before �.
After everything is said and done, Microsoft’s VASA-1 is a big step forward in AI and high-performance computers. VASA-1 could change the way we communicate with virtual characters because it can make talking faces that look like real people from a single still picture and a voice recording.
