Nodes Browser
ComfyDeploy: How Comfyui_JC2 works in ComfyUI?
What is Comfyui_JC2?
Wrapped Joy Caption alpha 2 node for comfyui from [a/https://huggingface.co/spaces/fancyfeast/joy-caption-alpha-two](https://huggingface.co/spaces/fancyfeast/joy-caption-alpha-two) Easy use, for GPU with less 19G, please use nf4 for better balanced speed and result. This Node also took a reference from /chflame163/ComfyUI_LayerStyle and [a/https://huggingface.co/John6666/joy-caption-alpha-two-cli-mod](https://huggingface.co/John6666/joy-caption-alpha-two-cli-mod)
How to install it in ComfyDeploy?
Head over to the machine page
- Click on the "Create a new machine" button
- Select the
Editbuild steps - Add a new step -> Custom Node
- Search for
Comfyui_JC2and select it - Close the build step dialig and then click on the "Save" button to rebuild the machine
Introduction:
Wrapped Joy Caption alpha 2 node for comfyui from https://huggingface.co/spaces/fancyfeast/joy-caption-alpha-two Easy use, for GPU with less 19G, please use nf4 for better balanced speed and result. This Node also took a reference from /chflame163/ComfyUI_LayerStyle and https://huggingface.co/John6666/joy-caption-alpha-two-cli-mod
Please notice🔥:
🔥BF16 need at least 19GB free VRAM to run
🔥NF4 need at least 10GB free VRAM to run
🔥Insufficient VRAM will lead to error in Comfyui
Usage:
-inside comfyui
-click comfyui manager
-Install Via git Url https://github.com/TTPlanetPig/Comfyui_JC2 or inside the folder ./comfyui/custom_nodes run
git clone https://github.com/TTPlanetPig/Comfyui_JC2
-for python_embeded comfyui version, inside folder ./comfyui/custom_nodes/Comfyui_JC2 run
../../../python_embeded/python.exe -m pip install -r requirements.txt
I will assume you have the pytorch ready in your PC, in case you don't, install from here:
pip install torch==2.4.0 torchvision==0.19.0 torchaudio==2.4.0 --index-url https://download.pytorch.org/whl/cu121
# (NOT Recommended if you don't familar with embeded python)
Model preparation:
-clip vision: https://huggingface.co/google/siglip-so400m-patch14-384/tree/main download all files and place in ComfyUI\models\clip_vision\google--siglip-so400m-patch14-384
-LLM: https://huggingface.co/unsloth/Meta-Llama-3.1-8B-Instruct download all files and place in ComfyUI\models\LLM\unsloth--Meta-Llama-3.1-8B-Instruct
-Joy capiton lora: https://huggingface.co/spaces/fancyfeast/joy-caption-alpha-two download all files and place in ComfyUI\models\Joy_caption\cgrkzexw-599808, i will suggest you use huggingface-cli to avoid mistaken on the names.
- Or you can download the models here and unzip it into your folder *
- https://pan.baidu.com/s/1yYRlDKclehSPv-tUVwfVHw?pwd=b84c *
Make sure your model is well placed as below.
Key options:
-
For 3090/4090 use bf16 otherwise nf4. please notice nf4 accuracy is lower than bf16, you can directly figure out.
-
For differnt purpose on the prompts
-
not always longer is better:
-
if enabled, the loaded clip,llm,lora model will not offload
-
if enabled, all the extra option will be effective. if disabled, even you enabled in detail options.
-
Must use together to be affective,
,
Enjoy!