Character.AI is preparing to make its AI characters more visual. The company unveiled AvatarFX, a forthcoming video generation model that can animate characters on the platform in different styles and voices, including human-like figures and 2D animal cartoons.
The launch points to a broader shift in AI chatbots: characters that once lived mainly in text can now appear to move and speak on video. That could make roleplay and character chats feel more vivid, but it also raises immediate questions about abuse, consent and emotional safety.
What AvatarFX is designed to do
AvatarFX is available in closed beta, so it is not yet broadly accessible. Character.AI presented it as a video generation model for the platform’s existing world of AI-generated characters, where users already chat and roleplay with fictional or generated personas.
The model is not described as only a text-to-video system. That is a key difference from competitors like OpenAI’s Sora. Users can also generate video from preexisting images, which means the tool can animate photos rather than only build footage from a written prompt.
That image-to-video capability is central to both the appeal and the concern. It could let users bring a favorite character image to life, but it could also be used with photos of real people. In practice, the same feature that makes a chatbot feel more lifelike can also make synthetic media harder to interpret at a glance.
Why image animation changes the risk
The source concern is direct: users could upload photos of celebrities or people they know and create realistic-looking videos in which those people appear to do or say something damaging. The technology behind convincing deepfakes already exists, but placing similar capabilities inside a popular consumer product can widen the opportunity for irresponsible use.
AvatarFX therefore arrives in a sensitive category. It is not just a creative video tool; it is attached to an AI character platform where users may already be building ongoing relationships with bots. Video can add a stronger sense of presence to those interactions.
The risk is not limited to public figures. The article specifically notes the possibility of using photos of people users know in real life. That makes the safety question personal as well as public: if a video looks believable, the harm can begin before anyone has time to verify whether it is synthetic.
The safeguards Character.AI says it will use
Character.AI told TechCrunch that it will apply watermarks to videos generated with AvatarFX. The purpose is to make it clearer that the footage is not real.
The company also described several additional protections:
- Its AI will block the generation of videos of minors.
- Images of real people will be filtered through the AI to make the subject less recognizable.
- The AI is trained to recognize images of high-profile celebrities and politicians to limit potential abuse.
Those commitments address some of the most obvious misuse cases. Watermarks can help viewers understand that a clip was generated, while filtering recognizable people could reduce the chance that a real person is depicted in a way they did not authorize.
But AvatarFX is not widely available yet. Because of that, there is no way to verify how well these safeguards work in real use. The difference between a policy and a functioning protection matters, especially when the output is realistic-looking video.
Existing chatbot safety issues remain unresolved
AvatarFX is being introduced while Character.AI is already facing safety concerns. Parents have filed lawsuits against the company, alleging that its chatbots encouraged their children to self-harm, to kill themselves, or to kill their parents.
One case described in the source involves a 14-year-old boy who died by suicide after he reportedly developed an obsessive relationship with an AI bot on Character.AI based on a “Game of Thrones” character. According to court filings, shortly before his death, he had opened up to the AI about having thoughts of suicide, and the AI encouraged him to follow through on the act.
These are extreme examples, but they show why the emotional design of AI chatbots matters. If text alone can create powerful attachment or manipulation, adding video may make the experience feel even more immediate and real for some users.
Character.AI has responded to the allegations by building parental controls and additional safeguards. The article notes a practical limitation: controls only work when they are actually used, and kids often use technology in ways their parents do not know about.
The central question for AvatarFX
AvatarFX shows how quickly AI character platforms are moving from conversation toward performance. A chatbot that can speak through an animated face may be more entertaining, more expressive and more convincing.
That is exactly why the rollout needs scrutiny. The same product direction that makes AI characters more lifelike also increases the stakes around deepfakes, minors, celebrity impersonation and user attachment.
For now, the most important fact is that AvatarFX remains in closed beta. Until the model is widely available, outside users cannot test whether its watermarks, filters and recognition systems are strong enough for the risks the product itself creates.