Multi-person image animation improves motion and identity matching
WeLike2Party! In-Context Motion Transfer for Multi-Human Image Animation
Computer Vision and Pattern RecognitionArtificial Intelligence
Summary
Animating people in images to make them move like in a video is hard when several people interact and block each other. The researchers created a new method called WeLike2Party that copies motion directly from videos without needing to find body poses or shapes. They also built a big fake video dataset to train and test their method, helping it keep who’s who and how they move even when people overlap. Their system makes it look like multiple people in a picture are moving realistically together.
What this means in practice
- •For visual effects artists: Create realistic animations of multiple interacting people from a single image and a driving video without relying on body pose extraction.
- •For game developers: Generate lifelike multi-character movements in game scenes by transferring motion directly from videos to character images with preserved identities.
Authors
Sangeyl Lee, Seunghyun Shin, Seungho Park, Wooseok Jeon, Hae-Gon Jeon
Abstract
Human image animation aims to transfer motion from a driving video to subjects in a reference image. Despite remarkable progress in video generation, achieving high-fidelity animation of multiple interacting subjects remains a challenge. Many existing approaches rely on explicit motion representations such as 2D skeletons or parametric body meshes and struggle to preserve identity-motion binding under inter-person occlusion. To address this limitation, we propose WeLike2Party, a multi-human animation framework built on direct in-context video conditioning without explicit pose or mesh extraction at inference. We further introduce Reference Asymmetric RoPE Conditioning to preserve fine-grained appearance details, and Identity Binding Supervision to associate each reference identity with its intended motion trajectory. To support cross-identity training, we construct MotionTwin, a large-scale synthetic dataset comprising 14.4K cross-identity video pairs with shared subject and camera motions, totaling 84.3 hours of photorealistic video. We additionally present MotionTwin-Bench, a cross-identity benchmark specifically designed to evaluate subject-level visual fidelity and identity-motion binding. Extensive experiments on MotionTwin-Bench and real-world videos demonstrate that WeLike2Party outperforms recent state-of-the-art methods in subject-level visual fidelity, identity-motion binding, and overall perceptual quality, particularly in multi-person interactions with substantial occlusion.