LongCat-Video

LongCat-Video is a 13.6B-parameter foundational video generation model developed to excel in Text-to-Video, Image-to-Video, and Video-Continuation tasks. It supports efficient and high-quality generation of long videos (minutes-long) without color drifting or quality degradation, marking an initial step toward world models.

Key features:

• Unified Architecture: A single model framework natively supports all three tasks (Text-to-Video, Image-to-Video, Video-Continuation) with consistent performance across each.
• Long Video Generation: Pretrained on Video-Continuation tasks, enabling stable generation of extended videos.
• Efficient Inference: Generates 720p, 30fps videos quickly using a coarse-to-fine generation strategy and Block Sparse Attention, optimizing temporal and spatial axes.
• Performance: Achieves results comparable to leading open-source and commercial models, validated via multi-reward Group Relative Policy Optimization (GRPO) on internal and public benchmarks.
• The model is released under the MIT License and is open-source. Citation and licensing terms are available in the repository. Note that developers must adhere to applicable laws and evaluate model suitability for specific applications.

---

The model is a component of the video generation pipeline, consisting of:

• Text encoder: ~5.7B parameters,
• Transformer: ~13.6B parameters,
• VAE: ~127M parameters,
• LORA: ~630M and ~802M parameters.

Total: ~20.1B parameters

---

The 'VRAM requirements' line is calculated based on the size of the weights in the specified precision, with an additional +10% coefficient for inference:
  model_parameters_count * bytes_per_parameter * 1.1 / 1024^3 
where model_parameters_count is number of model parameters, bytes_per_parameter is number of bytes used to store one parameter. For 16 bit = 2, 8 bit = 1, 4 bit = 0.5.

Announce Date: 24.10.2025
Parameters: 13.6B
VRAM requirements: 10.7 GB using 4 bits quantization, 21.3 GB using 8 bits quantization, 42.7 GB using 16 bits quantization
Developer: Meituan
License: MIT

Use our pre-built public endpoints for free to test inference and explore LongCat-Video capabilities. You can obtain an API access token on the token management page after registration and verification.

Rent your own physically dedicated instance with hourly or long-term monthly billing.

We recommend deploying private instances in the following scenarios:

• maximize endpoint performance,
• enable full context for long sequences,
• ensure top-tier security for data processing in an isolated, dedicated environment,
• use custom weights, such as fine-tuned models or LoRA adapters.

Contact our dedicated neural networks support team at nn@immers.cloud or send your request to the sales department at sale@immers.cloud.