Run ComfyUI on MindSpore

The most powerful and modular visual AI engine and application.

ComfyUI lets you design and execute advanced stable diffusion pipelines using a graph/nodes/flowchart based interface. Available on Windows, Linux, and macOS.

🔧 Dependencies and Installation

mindspore	ascend driver	firmware	cann toolkit/kernel
2.7.1	24.1.RC2	7.3.0.1.231	8.2.RC1

git clone https://github.com/zhanghuiyao/comfyui_mindspore.git -b master
git clone https://github.com/mindspore-lab/mindone.git -b master

export PYTHONPATH=$PYTHONPATH:/path_to/mindone:/path_to/comfyui_mindspore

cd comfyui_mindspore
pip install -r requirements.txt

Quick Strat

python main.py --listen 0.0.0.0 --port 9001

# run on ascend 310p
python main.py --listen 0.0.0.0 --port 9001 --force-fp16 --fp16-vae

speedup

# try accelerating your workflow by binding core
taskset -c 0-47 ...

performance

chip	models	image size	infer_steps	sampling cost
910B	flux	512*512	20	~4.3 s
910B	qwen_image	512*512	20	~13 s
310P	flux	512*512	20	~46.6 s