diff --git a/ppdiffusers/examples/text_to_image/README_sdxl.md b/ppdiffusers/examples/text_to_image/README_sdxl.md
index e75732e7f..05668c369 100644
--- a/ppdiffusers/examples/text_to_image/README_sdxl.md
+++ b/ppdiffusers/examples/text_to_image/README_sdxl.md
@@ -110,6 +110,15 @@ for file_name in sorted(os.listdir(dir_name)):
     image.save("sdxl_train_pokemon_" + file_name + ".png")
 ```
 
+## NPU硬件训练推理
+请参照[tools](../../tools/README.md)进行NPU硬件Paddle安装。
+
+使用NPU进行sdxl微调训练和推理时参考如下命令设置相应的环境变量，训练和推理运行命令可直接参照上述LoRA训练和推理命令。
+```bash
+export FLAGS_npu_storage_format=0
+export FLAGS_use_stride_kernel=0
+```
+
 
 ## Stable Diffusion XL (SDXL) LoRA 训练示例