DreamActor-M1 turns a single photo into a performance
Originally posted on LinkedIn, April 4, 2025.
The rules of filmmaking have changed for good. This week, ByteDance, the parent company of TikTok, released a new image-to-video model called DreamActor-M1, marking another major shift in AI-generated video.
You give it a single image of a person and a reference video of anyone else. The image can show a celebrity, a photogenic face, or a fictional character. The model creates a realistic video in which that character moves, smiles, speaks, and dances exactly like the person in the reference video.
This is not a deepfake or traditional animation. It is AI technology built on a deep understanding of human movement, facial expressions, and timing. DreamActor-M1 is more than a technical demo. It shows a future in which realistic animation can be created without actors, cameras, or massive productions.
In one demo, a single image of Marilyn Monroe moves its lips, smiles, and looks from side to side by following another person’s movements. In another, a cartoon character performs a convincing rap with sharp body movements and precise facial expressions. The same animation can easily be transferred to children, elderly people, or imaginary characters without changing the original reference video.
We are quickly approaching the first films made by combining talented actors with AI characters processed by models. That means fewer actors, smaller crews, and no limit beyond the imagination of the director and creator. I do not know whether that is bad news for the industry, but I am waiting for the first films and series made with this technology. They will push what can appear on screen beyond anything we have seen before.