Character.AI steps into AI video generation with AvatarFX

Just weeks after introducing "Parental Insights", a feature designed to give guardians some oversight into their teen"s activity, Character.AI is unveiling a new AI video model named AvatarFX. This development aims to bring the platform"s AI characters to life beyond text, allowing users to generate animated, speaking avatars.

According to information shared by the company, AvatarFX allows users to transform an image into a dynamic character by simply uploading a picture and selecting a voice. The platform then enables these characters to speak, move, and show emotions with what is described as remarkable realism and fluidity.

This is achieved through a "SOTA DiT-based diffusion video generation model", a type of advanced AI model trained on curated data and optimized with specific techniques for audio conditioning and efficient generation. The company claims this technology allows the creation of "high-fidelity, temporally consistent videos at impressive speeds, across longer sequences, even with multiple speakers, multiple turns."

When it comes to video generation, the platform already faces competition from companies like OpenAI with Sora and Google with Veo. However, AvatarFX appears to offer a different workflow focused on animating specific images rather than purely generating video from scratch or text.

While other models excel at creating diverse scenes from prompts, AvatarFX"s reported ability to animate uploaded images, including potentially photos of real individuals, introduces a distinct set of possibilities and concerns.

The potential for this kind of technology to be leveraged for abuse is immediately apparent. Users could theoretically upload photos of celebrities or even people they know in real life to create realistic-looking videos depicting them saying or doing something fabricated.