This ComfyUI workflow turns a plain text prompt into a cinematic video using Lightricks' LTX-2.5, a production-ready diffusion model. The core of the graph is the LTX-2.5 text-to-video node (UUID: 8b4f085c-1bb3-4ecd-aeed-603a8d6d3970), which consumes your prompt, resolution, and generation settings to synthesize a coherent clip. A ResolutionSelector node feeds width and height values that are aligned to the model’s preferred multiples-of-32, and a SaveVideo node encodes the result as a single video file. A MarkdownNote is included in the graph to surface quick tips and resolution guidance right inside ComfyUI.

Under the hood, LTX-2.5 uses a Pixel Diffusion, keyframes-first approach: it establishes strong visual anchors and then diffuses motion between them, helping maintain temporal consistency and cinematic pacing. You can steer fidelity and GPU memory use by pairing the ResolutionSelector with the model’s parameters. For example, 0.5 megapixels at 16:9 maps to 960x544, while 0.3 MP maps to 736x416; both are safe starting points on modest GPUs. When you need extra detail, enable the node’s optional spatial upscaling to boost resolution without re-generating motion from scratch, then finalize with SaveVideo at your target FPS and file path.