LongCat-Video
13.6B-parameter open video model for text, image, and long-form video
What it is
Meituan's 13.6B-parameter open video generation model uses one network to turn text or a still image into video and to extend existing footage, producing 720p 30fps clips that can run for several minutes. The repo ships inference scripts for single- and multi-GPU setups, plus the Avatar variants that animate talking characters from audio input. Researchers and developers who want to generate or study long AI video on their own GPUs can download the weights from Hugging Face and run it locally.