You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
This repository is the official implementation of ID-Animator.
It is a Zero-shot ID-Preserving Human Video Generation framework. It can generate high-quality ID-specific human video with only one ID image as reference.
Release ID-Animator checkpoints and inference scripts
Release ID-Animator dataset and training scripts
Release ID-Animator SDXL version
Code will be released soon!
Abstract
Generating high fidelity human video with specified identities has attracted significant attention in the content generation community. However, existing techniques struggle to strike a balance between training efficiency and identity preservation, either requiring tedious case-by-case finetuning or usually missing the identity details in video generation process. In this study, we present ID-Animator, a zero-shot human-video generation approach that can perform personalized video generation given single reference facial image without further training. ID-Animator inherits existing diffusion-based video generation backbones with an face adapter to encode the ID-relevant embeddings from learnable facial latent queries. To facilitate the extraction of identity information in video generation, we introduce an ID-oriented dataset construction pipeline, which incorporates decoupled human attribute and action captioning technique from a constructed facial image pool. Based on this pipeline, a random face reference training method is further devised to precisely capture the ID-relevant embeddings from reference images, thus improving the fidelity and generalization capacity of our model for ID-specific video generation. Extensive experiments demonstrates the superiority of ID-Animator to generate personalized human videos over previous models. Moreover, our method is highly compatible with popular pre-trained T2V models like animatediff and various community backbone models, showing high extendability in real-world applications for video generation where identity preservation is highly desired.