Wan-Move - An open-source motion-controlled video generation framework developed in collaboration with Alibaba, Tsinghua University, and others.

What is Wan-Move?

Wan-Move is an open-source motion-controlled video generation framework developed by Alibaba Tongyi Labs and other institutions. It achieves high-quality video motion control through latent trajectory guidance. Its core highlight is that it requires no architectural changes to existing image-to-video models, enabling fine-grained point-level motion control and generating 5-second, 480p videos with motion control quality comparable to commercial systems. Wan-Move provides the MoveBench benchmark, which includes a large-scale sample and high-quality trajectory annotations, allowing for the evaluation and comparison of motion control capabilities across different methods.

Main functions of Wan-Move

• High-quality motion controlIt can generate 5-second, 480p videos with motion control quality comparable to commercial systems, meeting the needs of high-quality video creation.
• Potential trajectory guidanceIt generates a spatiotemporal feature map along the trajectory by propagating the features of the first frame, without the need for an additional motion module, and can be seamlessly integrated into existing image-to-video models.
• Fine-grained point-level controlIt supports precise region-level motion control for each element in the scene, enabling highly customized video effects.
• Benchmark test MoveBenchIt provides large-scale, diverse, and long-duration video samples and high-quality trajectory annotations for evaluating and comparing the motion control capabilities of different methods.
• Open source and ease of useThe code, model weights, and MoveBench are all open source, allowing users to quickly get started with video generation and motion control experiments, lowering the barrier to entry.

The technical principle of Wan-Move

• Potential trajectory guidanceWan-Move uses latent trajectory guidance to represent the motion of objects as dense point trajectories and maps these trajectories into a latent space. By propagating the features of the first frame along the trajectory, aligned spatiotemporal feature maps are generated, enabling fine-grained control over motion during video generation.
• No additional modules requiredThe framework achieves motion control without requiring architectural changes to existing image-to-video models (such as Wan-I2V-14B) or the addition of additional motion encoders, and has excellent compatibility and scalability.
• Large-scale trainingThrough large-scale training data and optimization, Wan-Move can generate high-quality 5-second, 480p videos with motion control capabilities comparable to commercial systems, ensuring both visual appeal and motion accuracy.
• Fine-grained point-level controlThe movement of objects is represented by dense point trajectories, allowing users to perform precise region-level motion control on each element in the scene, achieving highly customized video effects.
• Benchmark test MoveBenchTo verify and evaluate the effectiveness of motion control, Wan-Move provides the MoveBench benchmark, which includes a large sample size, diverse content categories, long-duration videos, and high-quality trajectory annotations, providing a standardized testing platform for research and development.

Wan-Move's project address

• Project official website: https://wan-move.github.io/
• Github repositoryhttps://github.com/ali-vilab/Wan-Move
• HuggingFace model libraryhttps://huggingface.co/Ruihang/Wan-Move-14B-480P
• arXiv technical paper: https://arxiv.org/pdf/2512.08765

Application scenarios of Wan-Move

• Video creationUsers can generate videos with specific motion effects by defining the motion trajectory of objects. This is applicable to fields such as animation production, special effects design, and creative short videos, helping creators quickly realize complex motion scenes.
• Advertising and MarketingIn advertising videos, Wan-Move can be used to generate dynamic product displays, brand stories, and more, attracting viewers' attention through precise motion control and enhancing the appeal and impact of the advertisement.
• Video editingIt supports editing the first frame of a video and applying those changes to the entire video. It also supports motion copying and camera motion control, helping video editors quickly adjust and optimize video content.
• Education and TrainingIn the field of education, Wan-Move can generate teaching videos with dynamic demonstration effects, such as physics experiments and biological animations, which help students better understand and learn knowledge through vivid motion displays.
• Game developmentIn game development, Wan-Move can be used to generate animation effects in games, such as character movements and scene changes, thereby enhancing the visual effects and user experience of the game.
• Virtual Reality (VR) and Augmented Reality (AR)Wan-Move can generate dynamic video content that matches virtual environments or augmented reality scenes, providing users with a more immersive experience.