How to Generate AI Videos Locally with LTX 2.3 in ComfyUI (FREE)
Want to create AI videos on your own PC without paying for every generation? In this tutorial, I'll show you how to use LTX 2.3 with ComfyUI to generate short AI videos locally using text prompts or reference images.
LTX 2.3 is an open-source audio-video generation model designed for high-quality video creation. With ComfyUI's native workflows, you can experiment with text-to-video and image-to-video generation while controlling the settings yourself.
🎬 What Is LTX 2.3?
LTX 2.3 is an AI video generation model that can create videos from text descriptions and animate images. Its ComfyUI workflows let you configure generation settings and build a repeatable local video creation process.
Local generation can avoid per-video cloud generation fees, but you still need suitable hardware, storage, and the required model files. Downloading and running models may require significant disk space and processing time.
🎥 Recommended Settings for a 5-Second Video
For your first test, use these settings as a starting point and adjust them to match your chosen workflow and GPU.
| Setting | Starting value |
|---|---|
| Duration | Approximately 5 seconds |
| Frame rate | 24 FPS |
| Frame count | 121 frames |
| Resolution | Start low, then increase gradually |
| Workflow | Distilled workflow for faster inference |
Note: 121 frames at 24 FPS corresponds to about 5.04 seconds when calculating duration as frame count divided by frame rate. Actual playback duration can depend on how the video is encoded.
⚙️ How to Set Up LTX 2.3 in ComfyUI
- Install or update ComfyUI: Use the official ComfyUI website and make sure your installation supports the LTX 2.3 templates.
- Open the Template Library: Navigate to the video workflows and search for LTX 2.3.
- Choose a workflow: Select Text-to-Video to create a clip from a prompt, or Image-to-Video to animate a reference image.
- Download the required models: Follow ComfyUI's prompts to obtain the model files needed by the selected template.
- Write your prompt: Describe the subject, setting, camera movement, lighting, and actions that should occur over time.
- Configure the video: Set the frame count, frame rate, resolution, and other options available in the workflow.
- Generate the video: Queue the workflow and wait for processing to finish.
- Review and export: Check the output and save the generated video using the workflow's output settings.
📝 Text-to-Video vs. Image-to-Video
Text-to-Video
Describe a scene in words and let the model generate a video based on your prompt. This is useful for cinematic shots, landscapes, concept footage, and creative experiments.
Image-to-Video
Provide a starting image and describe how it should move or change. This can help you create animated scenes while using the reference image to guide the video's visual appearance.
💻 GPU and VRAM Requirements
Local AI video generation can be demanding. The official LTX ComfyUI setup documentation lists a CUDA-compatible GPU with 32GB or more of VRAM for its standard configuration, along with substantial disk space for models and cache.
That does not mean every configuration uses the same requirements. Quantized models, distilled workflows, CPU offloading, and lower-resolution settings may reduce memory demands, but they can affect speed, compatibility, and output quality.
- Start with a lower resolution to test your setup.
- Use a compatible distilled model when faster inference is a priority.
- Check the model and workflow documentation before downloading files.
- Allow extra disk space for model weights, temporary files, and generated videos.
- If you encounter an out-of-memory error, reduce the workload or investigate supported memory-saving options.
Important: Successful generation on a particular GPU is not guaranteed. Requirements depend on the exact checkpoint, precision, workflow, system RAM, and ComfyUI configuration.
💡 Tips for Better AI Videos
- Be specific with prompts: Describe the subject, environment, movement, camera angle, and lighting.
- Test before increasing resolution: A small test can help identify workflow or memory issues before a longer generation.
- Describe motion clearly: Explain what happens during the clip rather than listing only visual keywords.
- Keep early experiments simple: Start with one main subject and a straightforward action.
- Compare workflow settings: Different settings can change motion quality, detail, and generation time.
🔗 Official Resources
- LTX 2.3 ComfyUI Workflow Examples
- ComfyUI Official GitHub Repository
- ComfyUI-LTXVideo GitHub Repository
- ComfyUI Official Website
- LTX 2.3 Model Information
📺 More AI Tutorials from Unreal Fusion
Enjoy AI video generation, ComfyUI workflows, local AI tools, and creative technology? Subscribe to Unreal Fusion for more tutorials and experiments.
If this tutorial helped you, please LIKE 👍, SUBSCRIBE 🔔, and SHARE the video. Thanks for watching!
⚠️ Disclaimer
This article is for educational purposes. LTX 2.3 is open-source software, but model licenses and dependencies may have their own terms. Check the relevant licenses before commercial use. Performance and results vary depending on hardware, model files, and workflow settings.
#LTX23 #ComfyUI #AIVideo #AI #TextToVideo #ImageToVideo #LTXVideo #GenerativeAI