ComfyUI-LTXVideo: A Lightweight AI Video Model

We now have an open source (Apache-2.0 license) video generator inside of ComfyUI. ComfyUI-LTXVideo is a set of custom nodes designed for ComfyUI, enabling the use of the LTXVideo diffusion model.

These nodes allow you to create videos from text, images, or existing videos with ease. Whether you're working on short clips or extended scenes, the model offers impressive quality and speed.

LTXVideo's Key Features:

1. Scalable Video Production: Generate extended, high-quality videos with smooth transitions and consistent visuals.
2. Fast Rendering: Optimized for GPUs and TPUs, it drastically reduces generation time without sacrificing quality. Runs efficiently on consumer GPUs like Nvidia 4090, delivering up to 24 FPS at 5-second intervals.
3. Frame-to-Frame Consistency: Ensures seamless motion and eliminates issues like flickering or object distortion.
4. Provides sharp visuals and consistent styles across frames with minimal artifacts.

How to Get Started

1. Access the model and resources via GitHub or the Hugging Face Playground.
2. Update your ComfyUI installation to the latest version.
3. Download the model files and place them in the respective folders (or use ComfyUI Manager to install missing nodes after loading the workflow).
4. Start generating videos using text prompts, images, or video inputs, using sample workflows.

Workflow Highlights

You can find sample workflows on GitHub's page.

• Text-to-Video: Enter detailed prompts to generate cinematic scenes.
• Image-to-Video: Animate static images into dynamic videos.
• Video Extensions: Extend existing videos by seamlessly stitching frames.

Give it a try.

I have made a few generations so far and it's quite good for a free opensource model. Thanks to everyonw involved for releasing it. Sure, it won't be close to proprietary models but it has potential, and even as is can be used for some content creation, especially if we're talking landscapes more than people.