Last week, a research team from Chinese e-commerce giant Alibaba unveiled their new AI program called Animate Anyone that can create realistic video animations from still images.
Animate Anyone utilizes a novel generative video technique that enables it to animate still images with a high degree of realism while overcoming the unrealistic and hallucinatory effects that older AI models often produced when simulating plausible movement details. This AI program is notable for its ability to consistently preserve appearance details, such as facial features and clothing patterns.
Remarkably, the program allows users to control the animation through motion capture or by using another video as a reference. The resulting animations are impressively convincing, closely resembling real-life movements.
- 5 AI-powered presentation tools to help you at work
- 10 jobs at risk of getting replaced by AI, according to Google Bard
Animate Anyone is not without competition. MagicAnimate, which describes itself as a temporally consistent human image animation using a diffusion model, also provides AI-powered animation capabilities. It has its own appearance encoder and video fusion technique that allows it to generate animations with seamless transitions and enhanced detail retention throughout the frames.
Despite these features, MagicAnimate’s accuracy still requires refinement. Alex Carlier, a former researcher at Meta, tested MagicAnimate and identified several issues such as body shape deformation and poorly animated hands.
The increasing sophistication of AI programs like Animate Anyone and MagicAnimate raises concerns about their potential abuse. With the capability to create convincing fake videos from a single image, these technologies could depict individuals in scenarios or saying things they never did. This advancement makes it increasingly challenging for people, even those aware of AI deepfakes, to confidently identify AI-generated content.