The Harry Potter novels and subsequent film franchise imagined wall-hung portraits that had the ability to move and speak – and AI could soon give a similar animation treatment to still photos.
The videos created by the upcoming Character.AI AvatarFX won’t hang on a wall, but they will give a still photograph the ability to speak.
Announced in a blog post earlier this week, AvatarFX is an upcoming AI tool by Character.AI that animates still photos, giving people, paintings, fantasy characters, inanimate objects, and more the ability to speak on screen.
The tool, which is expected to roll out to subscribers first in the coming months, animates a still photo with facial expressions, lip movements, and a voice.
The technology mixes a still photo with user-typed instructions for what the avatar should say. The multimodal AI technology uses a flow-based diffusion model, which the company says enables the tool to generate realistic lip, head and body movements.
That’s paired with Character.AI’s existing TSS voice model that generates voices to go from a still image to a talking (or perhaps singing) video.
The company says the AI tool works beyond just animating people – it can also be used to make pets, illustrated characters or even inanimate objects appear to talk. Starting with an image rather than text enables users to have more control over the resulting video, the company says.
The announcement for the upcoming feature was accompanied by a video showing off what the technology can do:
ABOVE: Watch the new technology in action
While AvatarFX is still in testing, the company says that it's implementing a number of safety features to avoid potential pitfalls, like using the AI to create deepfakes.
Character.AI says that images of humans are changed using AI so that the person in the video is not recognizable as the original person. Photos of high-profile politicians and other public figures are blocked from being used on the system at all.
The company also says that it applies a watermark to those generated videos, signaling to the viewer that the footage is a generation and not something that happened in real life. The company also notes that the dialog is run through a safety filter for content that may violate safety policies.
Users must also agree to terms that prohibit things like impersonation, bullying and the use of protected intellectual property. Violators will only have one strike before losing access to the tool.
Character.AI did not offer a specific launch date for AvatarFX other than “in the coming months.”
You may also like…
Learn how to spot an AI-generated photo or browse the best free video editing software.
You must confirm your public display name before commenting
Please logout and then login again, you will then be prompted to enter your display name.
