Wrapped Joy Caption alpha 2 node for comfyui from a/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
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
🔥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
-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)
-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.
Make sure your model is well placed as below.
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!