Deepfake technology is about to take a giant leap forward with the introduction of 'Animate Anyone', a new generative video technique developed by researchers at Alibaba Group’s Institute for Intelligent Computing. This advanced technology will allow for the creation of realistic videos from still images, potentially opening the door for more sophisticated deepfakes.
Unlike its predecessors, DisCo and DreamPose, Animate Anyone has crossed the threshold from being a 'janky academic experiment' to being 'good enough if you don't look closely'. The technology works by extracting details such as facial features, patterns, and poses from a reference image. It then creates a series of images where these details are mapped onto slightly different poses, which can be motion-captured or extracted from another video.
While earlier models had issues with 'hallucination', where the model had to invent plausible details like the movement of a sleeve or hair, Animate Anyone has significantly improved on this. However, it's not perfect. The technology still struggles with the eyes and hands, and poses that deviate too much from the original can pose a challenge.
The implications of this technology are both exciting and unnerving. It could potentially allow a malicious actor to create a realistic video of anyone from a single good-quality image. While the technology is currently too complex and buggy for general use, the pace of AI development suggests it won't stay that way for long.
The team behind Animate Anyone isn't releasing the code just yet, but they are actively preparing for public release. As deepfake technology continues to evolve, we may soon find ourselves in a world where seeing isn't necessarily believing.
Made with TRUST_AI - see the Charter: https://www.modelprop.co.uk/trust-ai
Comments