LHM (Large Animatable Human Reconstruction Model) is a model for rapid, high-fidelity animatable 3D human reconstruction from a single image, decoupling geometry, appearance and deformation with a multimodal transformer architecture that encodes human body positional features and image features through attention.
It preserves clothing geometry and texture in detail, producing realistic 3D avatars via 3D Gaussian splatting, with a head feature pyramid encoding scheme that enhances face identity preservation and fine detail recovery, and it generates animatable models in seconds without post-processing, supporting real-time rendering, pose adjustment and export of 3D mesh files such as OBJ. Trained on large-scale video datasets with an image reconstruction loss, it generalizes well to diverse real-world scenarios without relying on synthetic 3D scans or computationally intensive optimization.
LHM is suited to virtual reality, gaming, e-commerce and entertainment where detailed animatable avatars are needed quickly, and it offers an integrated Gradio interface for local visualization and interactive pose manipulation.
Get more likes & reach the top of search results by adding this button on your site!
Get top updates in AI to your inbox every weekend. It's free!