What is comfyui_rebels_lingbot?

This project adds custom nodes to ComfyUI so you can run LingBot-Video, an AI model that turns text, or text plus an image, into short video clips, on a regular home GPU. It was built and tested on an RTX 3070 with 8GB of video memory and 16GB of system RAM, so it targets people without high end hardware. Two versions of the model are supported. The smaller one, at 1.3 billion parameters, fits entirely inside 8GB of video memory. The larger one, a 30 billion parameter mixture of experts model, is loaded from a compressed GGUF file and streams only the parts it needs into memory as it works, so it can also run on the same 8GB card. A separate text understanding model, Qwen3-VL, handles reading your prompt. To save memory, it runs on the regular processor rather than the graphics card, and is unloaded before the video model starts, so the two never compete for the same memory at once. A key detail is how prompts work. LingBot-Video was trained on structured JSON captions rather than plain sentences, with separate fields for things like scene description, camera movement, and lighting. The project includes a node that builds this exact format for you, so instead of writing a normal sentence you fill in structured fields. Getting the lighting fields right matters a lot: leaving them unset tends to make videos come out too dark. Setup involves cloning the project into ComfyUI's custom nodes folder, installing a handful of Python packages, and downloading the model weight files into the right folders. The 30 billion parameter model needs an extra configuration file copied into place. The project documents specific settings that matter, including a fixed output resolution, a maximum clip length, and recommended guidance values that differ between text to video and image to video modes. The documentation also lists known limitations. Turning on a memory saving decode option can cause colors to shift over longer clips, so it is off by default. Dark or lifeless video output is described as almost always caused by using plain text prompts instead of the structured format the node provides.