LTX-Video Sequential Prompt Relay | 8GB VRAM Optimized

This workflow is built to handle smooth, multi-segment sequential actions in LTX-Video using prompt scheduling. I specifically optimized the context chunking to prevent CUDA memory deadlocks on hardware like an RTX 3070 with 8GB of VRAM and 16GB of system RAM, allowing you to push longer frame runs without crashing your system.

⚠️ Required Node: ComfyUI-PromptRelay

Important Installation Note: This node currently has issues downloading directly through the standard ComfyUI Manager. You will likely need to install it manually from GitHub.

How to install manually:

• Go to your ComfyUI/custom_nodes directory.

• Open your command prompt/terminal in that folder and run: git clone https://github.com/kijai/ComfyUI-PromptRelay (or your specific fork/URL).

• Alternatively, download the .ZIP directly from the GitHub repository, extract it, and place the extracted folder inside your custom_nodes directory.

• Restart ComfyUI.

🧠 Model Requirements

• LTX-Video 2.3: If you are running an 8GB VRAM card, it is highly recommended to use the GGUF or FP8 quantized versions of the model to avoid out-of-memory errors.

📝 How to Use the Relay Node

• Input length: Input your desired segment length and use the +add button to add each new segment to the string.

• Format your prompts: Write your prompts using the starting frame as the key (example: "0": "your first prompt", "85": "your second prompt"). Each segment will begin exactly on those numbered frames.

• Manual adjustment: You can use the slider to manually adjust the segment lengths in the timeline after they have been added.

• Match your frames: Make sure the total frame count in your sampler matches the total frames spaced out in your relay node.