Want to join in on the latest TikTok dance trend but don’t have the time or skills to learn the moves? Well, now you can! Researchers from Nanyang Technological University in Singapore and Microsoft have developed an artificial intelligence (AI) model called Disentangled Control for Referring Human Dance Generation in Real World (DisCo) that can create a video of you dancing from just a single still image!
Here’s how it works: DisCo splits the image into three parts – the background, the foreground, and the pose of the person in the shot. The AI then morphs the person into a series of poses, generating individual frames that, when compiled, create a realistic video of that person dancing.
The team trained DisCo on a dataset of around 700,000 generic images of people from TikTok, allowing it to learn about different poses and how to separate foregrounds from backgrounds. The model was further trained on a smaller dataset of approximately 350 dance videos, each 10 to 15 seconds long, to enhance its understanding of human movement while dancing.
According to Tan Wang, one of the researchers involved in the project, DisCo achieves higher levels of realism compared to other similar models, such as the Google and Nvidia-backed DreamPose. Wang suggests that DisCo could be integrated into platforms like TikTok, allowing people to participate in dance trends, even if they can’t dance themselves.