LongCat-Video.
LongCat-Video is an open-source video model from Meituan (LongCat team). Windpaint doesn't serve it yet, so there's no model id or price. Request early access to start with the live models.
- Status
- Coming soon
- Task
- Text to video, Image to video
- Input
- Prompt, optional start frame or video
- Output
- Clip at 720p, 30 fps
- Developer
- Meituan (LongCat team)
- Parameters
- 13.6B
- Licence
- MIT
- Released
- October 2025
What the model does.
Paraphrased from the official model card.
LongCat-Video is a 13.6B video model from Meituan. One model does text-to-video, image-to-video and video continuation, where it extends a clip you already have.
It was pretrained on continuation, which its card says lets it build videos minutes long by extending clip after clip. It generates 720p at 30 fps, working coarse to fine in time and resolution.
Coming soon to Windpaint.
LongCat-Video isn't served, so it has no model id, no price and no request example here. When it goes live, this page gets its tiers, prices and request fields like the live models have.
Want this model? Request early access. We're adding models as we add capacity.
Is LongCat-Video licensed for commercial use?
Licence: MIT.
Yes. MIT allows commercial use, including running the model behind a paid API, with no revenue or user caps.
- The model weights are released under the MIT License. The README adds that the licence does not grant any rights to use Meituan trademarks or patents. LongCat-Video repo
- The README says developers and downstream users are responsible for complying with applicable laws, including data protection, privacy and content safety, and that nothing in the model card alters or restricts the MIT terms. LongCat-Video repo
This is our reading of the licence, not legal advice. Read the licence text before you rely on it.
LongCat-Video hardware requirements.
Published by Meituan (LongCat team) for running the weights yourself. These are the developer's figures, not Windpaint's.
- Published VRAM
- Not stated
- Applies to
- No VRAM figure in the README. Single- and multi-GPU commands; FlashAttention-2 by default.
- FlashAttention-2 is enabled in the model config by default; the config can be changed to FlashAttention-3 or xformers once installed. LongCat-Video repo
- The README gives single-GPU and multi-GPU (context parallel, 2-GPU example) commands for each demo. LongCat-Video repo
LongCat-Video native resolution.
What the developer's own code and examples use.
- The README says it generates 720p, 30fps videos using a coarse-to-fine strategy along the temporal and spatial axes. LongCat-Video repo
- Official text-to-video demo: the base pass is 832 wide by 480 high, 93 frames, written at 15 fps. A refinement-LoRA pass then produces the 720p output, written at 30 fps (15 fps if
spatial_refine_onlyis set). LongCat-Video repo, run_demo_text_to_video.py - The long-video demo starts from a 93-frame clip and chains 11 continuation segments of 93 frames, each conditioned on the last 13 frames. The script comments this as a 1 minute video. LongCat-Video repo, run_demo_long_video.py
Recommended settings for LongCat-Video.
Steps, guidance and defaults from the developer's card and code, for running the weights yourself.
- Standard pass in the T2V demo:
num_inference_steps=50,guidance_scale=4.0, 93 frames. LongCat-Video repo, run_demo_text_to_video.py - Distilled pass with the bundled cfg_step LoRA:
num_inference_steps=16,guidance_scale=1.0,use_distill=True. LongCat-Video repo, run_demo_text_to_video.py - 720p refinement: load the bundled refinement LoRA, enable block sparse attention, 50 steps. LongCat-Video repo, run_demo_text_to_video.py
- The video continuation demo conditions on
num_cond_frames=13and takes resolution480por720p. LongCat-Video repo, run_demo_video_continuation.py - The official demos use this negative prompt: "Bright tones, overexposed, static, blurred details, subtitles, style, works, paintings, images, static, overall gray, worst quality, low quality, JPEG compression residue, ugly, incomplete, extra fingers, poorly drawn hands, poorly drawn faces, deformed, disfigured, misshapen limbs, fused fingers, still picture, messy background, three legs, many people in the background, walking backwards". LongCat-Video repo, run_demo_text_to_video.py
LongCat-Video limitations.
As the developer states them.
- Usage considerations: the model "has not been specifically designed or comprehensively evaluated for every possible downstream application". Developers should account for performance variation across languages and assess accuracy, safety and fairness before sensitive or high-risk use. LongCat-Video repo
- In the official demo scripts, 720p/30fps output comes from a second refinement pass over a 480p, 15 fps first pass. LongCat-Video repo, run_demo_text_to_video.py
Questions.
Can I call LongCat-Video on Windpaint today?
No. LongCat-Video isn't served yet, so it has no model id, no price and no entry in the catalog endpoint. There's no date for LongCat-Video.
What is LongCat Video Avatar?
A separate Meituan model for audio-driven character video. It supports audio-text-to-video, audio-text-image-to-video and video continuation, with single or multiple audio streams. Version 1.5 was released on May 21, 2026.
Where is the LongCat Video GitHub?
github.com/meituan-longcat/LongCat-Video. It has inference scripts for text-to-video, image-to-video, video continuation, long video and interactive video, on one GPU or across several.
Is LongCat-Video open source?
Yes. Its weights are licensed MIT, an OSI-approved open-source licence.
Can LongCat-Video be used commercially?
Yes. MIT allows commercial use, including running the model behind a paid API, with no revenue or user caps. Its licence is MIT. That's our reading of the licence, not legal advice; check the licence text before you rely on it.
How do I ask for LongCat-Video?
Request early access at app.windpaint.ai/register. We're letting people in gradually while we add models and capacity.