LongCat-Video is Meituan's 13.6B-parameter unified model for text-to-video, image-to-video, and video continuation, producing minutes-long 720p/30fps clips with efficient inference.