LingBot-Video: video generator — what it does and what you need to run it

LingBot-Video is an open-source video generator from Robbyant. The authors describe it as a model for video pretraining with a mixture-of-experts architecture, aimed at embodied intelligence tasks. In the repository, the project is tagged with image-to-video, video-generation, and world-models, meaning the core task is turning an image into a video sequence and building world models.

What it does

As described by the author, LingBot-Video tackles the problem of scaling mixture-of-experts for video pretraining. In practice, this means generating video from an image — the image-to-video tag is in the repository. The authors do not provide detailed usage scenarios in the fact sheet, so there is nothing to say about specific limitations on video length, resolution, or motion control.

What you need to run it

The code is written in Python and works through the diffusers library. To run it, you need access to the model weights: the largest file takes up 4.6 GB, and all weight files together — 11.3 GB. The authors do not specify minimum VRAM or system RAM requirements, so it is impossible to estimate from the fact sheet which GPU this will run on.

The license for both the code and the weights is Apache-2.0. This permits commercial use and modification. The code is open source, the repository is available at github.com/Robbyant/lingbot-video, and the weights are at huggingface.co/robbyant/lingbot-video-dense-1.3b. There is also a separate project site: technology.robbyant.com/lingbot-video.

Who it is for

The project is of interest to those working with image-to-video generation and willing to run the model locally via diffusers. The open Apache-2.0 license on the code and weights allows embedding the generator into your own pipelines, including commercial ones.

For those looking for a ready-made service with a web interface, LingBot-Video is not suitable: it is a repository with code and weights, not a cloud product. Also keep in mind that the authors do not publish hardware requirements — before running, you will have to check on your own whether there is enough memory for the 11.3 GB of weights and the generation process itself.

LingBot-Video is a fresh project: the repository was created in July 2026, with the last code change in August 2026. This means the documentation and examples may be incomplete, and the model's behavior may not be sufficiently battle-tested across a wide range of tasks.