A fully local running Qwen3.8 27b Assistant workflow in ComfyUI, built for those who care about privacy. You can use it to expand simple text, analyze images or videos, and use MiniMax H3 official skills to restructure prompts. It can also act like a normal chatbot. The ability depends on the model you choose., unlimited version also supported. Use it with caution and responsibility.
Check the tutorial video for instructions before using this workflow:
Description
FAQ
Comments (17)
I watched the video, and downloaded the workflow and found some of the nodes, but they are in.... Mandarin?
If you could provide a list of everything you'd need and links to HugginFace/GIT it would be very helpful
I'm just passing through, I did not attempt installing this. I skimmed through the video and noticed it shows a GitHub page. However they do not show the "main" branch - they show an English version branch. It may be that you need to go back and make sure you install from the English branch.
thanks , you need to install the node version I provided. the original node doesn't have an English Version. https://github.com/noname19921992/comfyUI-llama-TE-English/tree/main
@altoiddealer yes. should use the English version I provide
custom_nodes:
https://github.com/noname19921992/comfyUI-llama-TE-English/tree/main
Local Qwen GGUF LLM Models:
Place the model files under ComfyUI/models/LLM directory
https://huggingface.co/unsloth/Qwen3.8-27B-GGUF/tree/main
https://huggingface.co/JonathanColetti/Qwen3.8-27B-Uncensored-GGUF/tree/main
https://huggingface.co/huihui-ai/Huihui-Qwen3.8-27B-abliterated-GGUF/tree/main
This works great! Probably the trickiest part is selecting and installing the wheel, but towards the end of the video he walks you through it, pretty easy. Thanks <3
I'm really trying to run this... but I am getting errors with the comfy set up... currently the error is
Exception Message:** AttributeError: 'Llama' object has no attribute 'token_eot'
I've asked others for help... and they say
"The actual crash happens because the comfyUI-llama-TE-English custom node is trying to call a function called token_eot() (End Of Turn) inside the llama-cpp-python library. Your logs show you have version 0.3.35 installed, and it seems that version (or the way it was built) doesn't have that specific attribute"
ha. it seems you have an old vesion of llama cpp wheel. please uninstall it first and then follow the instructions in the video tutorial to reinstall the wheel
I got it to work!
Its actually pretty cool!!!!
My troubles were related to a pretty messy comfy folder, and at the end of the day, I just started over from scratch. Followed the Video, and got it working.
It worked and pretty easy.
work like a charm. so convenient to have this in Comfy.
Thank you for your Workflow.
A couple of questions: Is a prompt generation time of 4-5 minutes normal for the MinimaxH3 output?
4060TI / 16BG, 64GB Ram.
I'm using the Q4_uncensored model.
Is it possible to chain the 'Prompt Expansion' output into the 'Free Chat group' so I only have to type my prompt, and it generates my Minimax output in one go?
Thank you !
I think the speed is a bit slower than normal, and for your last question. this node might suite you : chflame163/ComfyUI_Qwen_H3_Prompt: Use local Qwen3.8-27B and official MiniMax-H3 Skills in ComfyUI to generate H3 prompts. Runs a standalone local llama-server , inference operates fully offline with no internet access.
is there any safetensor version of this gguf is too slow bro
...not yet, this is for gguf only...
THank you for this, it works well at prompting... But after it has run, i have to restart comfyui. After a prompt is generated, I use the prompt, and the GPU RAM goes to 100% and stays there, and the render doesn't happen. Something sticking around from this?
did you try right click to clean the vram?