DreamActor-M1

DreamActor-M1 is an AI image animation research model developed by ByteDance that generates video clips of a subject in motion from a single still photograph. The model applies a combination of techniques designed to preserve the subject's facial identity, skin texture, and appearance attributes across the generated frames while producing temporally coherent motion—meaning the animation flows smoothly without flickering or identity drift between frames. The model supports variable motion styles and intensities, allowing the generated animation to reflect different movement characteristics applied to the same source image. DreamActor-M1 is aimed at AI researchers studying human image animation and identity-preserving video generation, developers exploring avatar animation workflows for digital human or virtual character applications, and creative technologists experimenting with AI-driven motion synthesis from static images. A typical research use involves providing the model with a source portrait photograph and a target motion reference, then evaluating the output video for identity preservation quality, temporal coherence, and motion fidelity. The model's project page on GitHub Pages provides technical documentation, example outputs, and access to the research paper describing the underlying methodology. DreamActor-M1 is a research-stage model and is not packaged as a commercial SaaS product or consumer application. Access to the model code and weights is subject to the specific licensing terms published in the repository, which users should review before use, particularly for any commercial or derivative applications. As a research prototype, it is not designed with a production deployment pipeline, and users without machine learning infrastructure experience may require significant setup effort to run inference. ByteDance has not announced a commercial product based on DreamActor-M1 as of the current research phase.