Learning Timeline
Key Insights
Quality vs. Compression Trade-off
Using the smallest compression level (Q4) will significantly reduce VRAM usage and generation time, but results will be noticeably lower quality. Expect artifacts, such as distorted hair or finer details, compared to the official 16GB models.
Hardware Requirement Check
Official models require approximately 16GB of VRAM. If you have a GPU with 6GB to 14GB of VRAM, GGUF compression is strictly required to run this workflow locally.
Step by Step
Downloading and Installing GGUF Support
- Navigate to the Hugging Face repository for James 7's HunyuanVideo GGUF models.
- Click the 'Files and versions' tab.
- Select a model file based on your available VRAM (e.g., Q4_K_M.gguf for lower VRAM cards like 6GB).
- Download the file and move it into your ComfyUI directory at: `ComfyUI/models/unet`.
- Open your ComfyUI interface.
- Click the 'Manager' button in the main menu.
- Select 'Custom Nodes Manager'.
- Search for 'GGUF' in the search bar.
- Locate 'ComfyUI GGUF by City96' and click 'Install'.
- Restart ComfyUI to apply the changes.
Implementing the Low VRAM Workflow
- Double-click anywhere on the ComfyUI canvas to open the node search bar.
- Type 'unet' and select the 'Unet Loader GGUF' node.
- Locate the standard 'HunyuanVideo Model Loader' (or similar official loader) currently in your workflow.
- Disconnect the output wires from the standard loader.
- Connect the output of the new 'Unet Loader GGUF' node to the inputs previously fed by the standard loader (typically the Sampler or Model Sampling node).
- If the specific model file does not appear in the loader list, press 'R' on your keyboard to refresh the model list.
- Click the dropdown on the 'Unet Loader GGUF' node and select the downloaded GGUF file.
- Select the original, now disconnected model loader node.
- Press 'Ctrl+M' (or the mute button) to disable the old node and prevent it from using resources.
- Click 'Queue Prompt' to run the generation.
