Research Project

Add-it: Training-Free Object Insertion in Images With Pretrained Diffusion Models

Training-free method to add objects to real or generated images from a text prompt, leveraging diffusion model attention with weighted fusion and latent blending.

MaskedMimic: Unified Physics-Based Character Control Through Masked Motion Inpainting

A unified controller for physically simulated humanoids enabling diverse motions from intuitive intents across terrains, with applications including text-guided styles and path following.

PlaMo: Plan and Move in Rich 3D Physical Environments

Text-guided planning and physics-based control for humanoid locomotion in complex 3D environments with diverse terrain and obstacles.