A new AI model called Infinity generates realistic talking characters. It is based on a video diffusion transformer that has been trained with audio input. According to the developers, this is the first model of its kind. Users can enter their scripts and get videos of animated characters speaking the text. It can process different languages, animate paintings or sculptures, and even sing. However, there are still weaknesses with animals, cartoons and the representation of well-known personalities. The current version is only an intermediate stage, as the model is still being actively trained, the developers explained on Hacker News. More information and sample videos can be found in this blog post.