DreamActor-M1 is a state-of-the-art framework using diffusion transformers for creating realistic human animations from a single image, offering fine-grained control over facial expressions and body movements.
Features
Holistic Control
Utilizes a diffusion transformer based framework for complete control over both facial expressions and body movements, ensuring detailed and dynamic animation.
Multi-Scale Adaptability
Designed to handle varying image scales, from close-up portraits to full-body shots, maintaining high quality and dynamic animation.
Long-Term Coherence
Maintains visual and temporal consistency throughout animations, even over extended sequences, using advanced appearance guidance techniques.
Progressive Training
Employs a staged training approach that progressively introduces complexity, enhancing the model's ability to adapt to diverse animation scenarios.
Hybrid Motion Guidance
Combines implicit facial representations, 3D head spheres, and 3D body skeletons for nuanced animation control, ensuring natural and coherent results.
Complementary Appearance
Maintains consistency in unseen regions by sampling distinct poses and generating multi-frame references, enhancing animation quality.
Image Upload
Users can upload a source portrait with any aspect ratio to serve as the basis for animation.
Video Upload
Users can upload a driving video in a 1:1 aspect ratio to guide the animation movements.
Animation Generation
The app uses state-of-the-art technology to animate the uploaded image based on the driving video's movements, providing users with a lively animated result.