sdxl

0. Start docker 启动命令可参考: README.md

1. Prepare model

# download stable-diffusion-xl from huggingface
git clone https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0

2. Prepare dataset

3. Prepare scripts

# Ubuntu system
apt update
apt install -y libgl1-mesa-glx

pip install einops toml imagesize opencv-python voluptuous transformers==4.36.1 diffusers

4. Convert scripts to musa

5. Train

6. Inference

python sd_inference.py \
    --model_path ./stable-diffusion-xl-base-1.0 \
    --batch_size 2