Update README.md

README.md

@@ -13,15 +13,19 @@ tags:
 This model is a diffusion model for unconditional image generation of anime style 64*64 face pic.  
 The training set uses [anime-faces](https://huggingface.co/datasets/huggan/anime-faces). This is a dataset consisting of 21551 anime faces scraped from www.getchu.com, which are then cropped using the anime face detection algorithm in https://github.com/nagadomi/lbpcascade_animeface.
 
+Generating multiple pictures at once is prone to broken face. It has been tested that one picture at a time produces the best results and is not prone to broken faces.
+
 ### Usage
 
 ```python
 from diffusers import DDPMPipeline
 
-pipeline = DDPMPipeline.from_pretrained('Chilli-b/test2train_amine_face')
+device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+pipeline = DDPMPipeline.from_pretrained('Chilli-b/test2train_amine_face').to(device)
 image = pipeline().images[0]
 image
 ```
+
 ---
 
 # 中文版
@@ -31,12 +35,15 @@ image
 该模型是一个无条件扩散模型，用于生成尺寸为 64*64 的动漫风格脸部图片。
 训练集使用的是[anime-faces](https://huggingface.co/datasets/huggan/anime-faces),这是一个包含从 www.getchu.com 上爬取的21551个动漫脸，然后使用 https://github.com/nagadomi/lbpcascade_animeface 中的动漫脸检测算法进行裁剪的数据集。
 
+一次生成多张容易出现鬼脸或。实测每次出一张图的效果最好，不容易出现鬼脸。
+
 ### 模型使用
 
 ```python
 from diffusers import DDPMPipeline
 
-pipeline = DDPMPipeline.from_pretrained('Chilli-b/test2train_amine_face')
+device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+pipeline = DDPMPipeline.from_pretrained('Chilli-b/test2train_amine_face').to(device)
 image = pipeline().images[0]
 image
 ```