Audience
Researchers and developers in computer vision seeking a tool for generating realistic human animations from static images
About DreamActor-M1
DreamActor-M1 is a state-of-the-art diffusion transformer framework designed to generate realistic human animations from a single image. It offers fine-grained control over facial expressions and body movements, ensuring multi-scale adaptability from portraits to full-body views. It maintains temporal coherence in long videos, even for areas not visible in reference images. Its hybrid motion guidance combines implicit facial representations, 3D head spheres, and 3D body skeletons to achieve detailed animation control. Complementary appearance guidance uses multi-frame references to maintain consistency in unseen regions. A progressive three-stage training strategy optimizes different aspects of animation: starting with body skeletons and head spheres, adding facial representations, and finally fine-tuning all parameters.