Creating realistic dancing videos from a single photograph is now possible, thanks to the advancements in artificial intelligence (AI).
Developed by a team of researchers at Microsoft and Nanyang Technological University, the AI model called “Disentangled Control for Referring Human Dance Generation in Real World (DisCo)” is trained on TikTok dance trends. Using this technology, a still image of a person can be transformed into a dancing video.
DisCo works by dividing the photograph into three parts: the background, the foreground, and the person’s pose. The AI then takes the person’s pose and transforms it into a series of poses, creating individual frames.
When compiled together, these frames generate a realistic video of the person dancing. With this technology, users can compose dance videos of anyone they want, such as a dancing Elon Musk.
To train DisCo, the team used around 700,000 images of people from TikTok to teach the AI about different poses and how to separate the foreground from the background. They further trained it on a small dataset of about 350 dance videos to deepen its understanding of human movements during dancing.
This DisCo technology could potentially revolutionize TikTok and make it possible for users to generate their own dance videos without learning new choreographies.
However, it also raises concerns about the spread of deepfaked content. Additionally, it has potential applications in the post-production phase of movies and TV shows, allowing studios to add dance routines to actors without hiring professional dancers.
AI technology has been increasingly used in filmmaking, demonstrated by the use of deepfake technology to visually dub a film and change actors’ spoken language. The possibilities are expanding, and AI continues to transform various industries.