InfinityStar achieves state-of-the-art performance on image and video generation benchmarks, outperforming all autoregressive models by large margins and even surpassing diffusion competitors like HunyuanVideo. It is approximately 10x faster than leading diffusion-based methods. The framework is also capable of producing industrial-level 720p videos, setting a new standard for quality in its class. InfinityStar's unified modeling approach enables it to handle a wide range of generation tasks with ease and efficiency.
The InfinityStar project provides a comprehensive workflow for training and finetuning the model, covering data organization, feature extraction, and training scripts. The framework is implemented in Python and uses FlexAttention to speed up training. The project also includes a demo website for users to play with InfinityStar and generate videos. The demo website showcases the framework's capabilities and provides a fun way to interact with the technology.
