New technology announced by Alibaba
Animate anyone is a framework for creating videos by applying motion to characters from still images. It allows you to create complex appearance consistency in a single image, effectively control the character's movement through Openpose, and make the video frame transitions incredibly smooth. Alibaba says it will be uploaded to GitHub, but there's no specific release date yet.
Technical explanation
I'm not an expert, so I'll quote ChatGPT's explanation: Pose sequences are encoded using Pose Guider, combined with multiframe noise, and then a denoising process called Denoising UNet is used to generate the final video. I don't understand how this is different from the video generation using MulticontrolNet that was posted on Civit.ai, so that you can show a noise-free demo with high consistency.
Application examples
In the demo, we were able to see the technology applied not only to people but also to cartoon and movie characters, and they showed a scenario where they applied motions extracted from fashion videos and dance videos. If this technology is released as open source, we can expect to see a lot of dancing videos of AI characters on YouTube and TikTok Shorts, but I'm excited about the possibilities of creating content using AI.