April 23rd.CharacterAI Corporation today (April 23) announced in a tweet on the X platform the launch of the AvatarFX Model,The ability to make characters in still pictures "talk".
Users simply upload an image and pick a voice, and the platform generates talking, moving images. These images can also display emotions, presenting a stunning sense of realism and fluidity.
According to the company, this is made possible by an advanced AI model called the SOTA DiT-based diffusion video generation model. The model is carefully trained to efficiently generate high-quality video in combination with audio conditioning optimization techniques.
The technological highlight of AvatarFX is its "high-fidelity, time-consistent" video generation capability. Even in complex scenarios with multiple characters, long sequences, or multiple rounds of dialog, it maintains amazing speed and stability. Compared to competitors such as OpenAI's Sora and Google's Veo, AvatarFX doesn't generate video from scratch or based on text, but rather focuses on animating specific images.
This unique workflow provides users with a novel experience, but also poses potential risks. Users may upload photos of celebrities or acquaintances to create fake videos that appear to be real, sparking privacy and ethical controversies.