LongCatVideo

LongCat-Video is an advanced AI video generation model developed by Meituan, featuring 13.6 billion parameters. It stands out by enabling the creation of high-quality, minutes-long videos at 720p resolution and 30 frames per second. Built on a unified architecture, it supports multiple video-related tasks such as Text-to-Video, Image-to-Video, and Video-Continuation, all within a single model. Its design ensures consistent video quality without color drifting or degradation, making it ideal for generating smooth, long video sequences efficiently.

Key Features:

• Unified architecture supporting Text-to-Video, Image-to-Video, and Video-Continuation tasks in one model.
• Ability to generate minutes-long videos with stable quality and temporal coherence.
• Efficient inference with a coarse-to-fine generation strategy and Block Sparse Attention for fast 720p/30fps video production.
• Multi-reward reinforcement learning (GRPO) training for balanced output quality including text alignment, motion realism, and visual fidelity.

Use Cases:

• Creating dynamic videos from static images for marketing, social media, or storytelling.
• Generating narrative videos or product demonstration clips extending for several minutes.
• Extending existing video clips smoothly, maintaining consistent colors and motion.