MultiAnimate is a novel framework designed for multi-character pose-guided image animation, capable of handling complex interactions and resolving identity ambiguity through a randomized identifier training strategy.
We provide two pre-trained checkpoints to cater to different application scenarios and scalability requirements:
-
epoch=39-step=7000.ckpt (Standard Model)
- Capability: Supports image animation for up to 3 characters.
- Description: This is the primary model extensively evaluated in our paper, optimized for complex human interactions and temporal coherence.
-
epoch=23-step=4200-multi.ckpt (Extended Model)
- Capability: Generalizes to support up to 7 characters.
- Description: To demonstrate the zero-shot scalability of our framework, we extended the Standard Model.
Please refer to our
official GitHub repository for detailed inference instructions, environment setup, and data preparation.
