ComfyUI JoyCaption-Beta-GGUF Node

ComfyUI JoyCaption-Beta-GGUF Node

This project provides a node for ComfyUI to use the JoyCaption-Beta model in GGUF format for image captioning.

中文版说明

Acknowledgments:

This node is based on fpgaminer/joycaption_comfyui, with modifications to support the GGUF model format.

Thanks to the LayerStyleAdvance, I copied the relevant code for extra options from it.

20250802-Update:

Due to the difficulty of installing the GPU version of llama-cpp-python, I have uploaded the model to Ollama. You can now use it by installing Ollama and comfyui-ollama. Model link: https://ollama.com/aha2025/llama-joycaption-beta-one-hf-llava Meanwhile, I've separated the prompt template into a standalone node. For details, please check the ComfyUI JoyCaption-Beta-GGUF.

Usage

Installation

This node requires llama-cpp-python to be installed.

Important:

• Installing with pip install llama-cpp-python will only enable CPU inference.
• To utilize NVIDIA GPU acceleration, install with the following command:

  pip install llama-cpp-python --extra-index-url https://abetlen.github.io/llama-cpp-python/whl/cu124
  

  (Adjust cu124 according to your CUDA version)
• For non-NVIDIA GPUs or other installation methods, please refer to the official llama-cpp-python documentation: https://llama-cpp-python.readthedocs.io/en/latest/

llama-cpp-python is not listed in requirements.txt to allow users to manually install the correct version with GPU support.

Workflow Example

You can view an example workflow image at assets/example.png.

Model Download and Placement

You need to download the JoyCaption-Beta GGUF model and the corresponding mmproj model.

1. Download the models from the following Hugging Face repositories:

   • Main Model (Recommended): concedo/llama-joycaption-beta-one-hf-llava-mmproj-gguf
     • Download the relevant joycaption-beta model files and the llama-joycaption-beta-one-llava-mmproj-model-f16.gguf file.
   • Other Quantized Versions: mradermacher/llama-joycaption-beta-one-hf-llava-GGUF
   • IQ Quantized Version (Theoretically higher quality, potentially slower on CPU): mradermacher/llama-joycaption-beta-one-hf-llava-i1-GGUF

2. Place the downloaded model files into the models\llava_gguf\ folder within your ComfyUI installation directory.

Video Tutorial

You can refer to the following Bilibili video tutorial for setup and usage:

Video