Skip to main content
Back to Blog
AI/MLCloud Computing
4 April 20263 min readUpdated 4 April 2026

Creating Videos with LTX-2.3 Using DigitalOcean GPU Droplets

LTX 2 has already made a significant impact in the field of computer vision, particularly in affordable video generation. The latest version, LTX 2.3, continues this trend with...

Creating Videos with LTX-2.3 Using DigitalOcean GPU Droplets

LTX-2 has already made a significant impact in the field of computer vision, particularly in affordable video generation. The latest version, LTX-2.3, continues this trend with impressive upgrades, making complex and expressive video creation more accessible through open-source tools.

Key Highlights

  • Significant Improvements: LTX-2.3 offers enhanced image-to-video generation, addressing issues like "freezing" and the Ken Burns effect, and improving prompt adherence with a larger text connector.

  • High-Quality Output: This version supports portrait videos up to 1080×1920, with a redesigned latent space that delivers sharp details and cleaner audio alignment, suitable for professional use.

  • Accessible Video Production: By combining open-source software like ComfyUI with scalable cloud services such as DigitalOcean GPU Droplets, even small creators can now perform complex video generation tasks.

New Features and Enhancements in LTX-2.3

LTX-2.3 builds on its predecessor with numerous improvements, particularly in prompt adherence and audio quality. Here's a look at the enhancements:

  • Sharper Details: The updated Variational Autoencoder improves texture preservation, making fine details like hair and text clearer throughout the video generation process.

  • Better Prompt Adherence: With a significantly larger text connector, the model can handle more complex prompts involving multiple subjects and intricate spatial relationships more effectively.

  • Enhanced Image-to-Video Capabilities: Addressing previous weaknesses, LTX-2.3 now offers smoother motion and better visual consistency, resulting in fewer discarded generations.

  • Improved Audio Quality: By refining training data and introducing a new vocoder, audio outputs are cleaner, with reduced artifacts and improved alignment with video.

Running LTX-2.3 on a GPU Droplet

To utilize LTX-2.3, a powerful GPU is necessary. DigitalOcean’s GPU Droplets provide a convenient cloud-based solution. Here's how to set it up:

  1. Create a GPU Droplet: After logging into your account, navigate to the GPU Droplets section, select your desired data center, and choose a GPU type, such as H100 or H200. Ensure you add an SSH key for access.

  2. Access Your GPU Droplet: Use SSH from your local machine to connect to the droplet. Set up your working directory to begin installing ComfyUI.

Setting Up ComfyUI for LTX-2.3

To start using LTX-2.3, download the required model files using Comfy Org's base template. Follow these steps:

git clone https://github.com/Comfy-Org/ComfyUI
cd ComfyUI
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt
cd models/checkpoints
wget [model file URL]
cd ../text_encoders/
wget [text encoder URL]
cd ../loras/
wget [lora files URLs]
cd ../latent_upscale_models/
wget [upscale model URL]
python main.py

After setting up, use the public IPv4 address of your machine to access the interface locally. Download a template link, upload it to ComfyUI, and upon successful setup, you should see an interface ready for video generation.

Final Thoughts

LTX-2.3 marks a significant advancement in open-source video generation, enhancing fidelity and ease of use. Paired with the scalable power of cloud-based GPU resources, it democratizes high-quality video production, enabling individual creators and small teams to achieve results once only possible for large research teams. As infrastructure and models continue to improve, the future of video generation looks increasingly promising and accessible.