开源AIGC学习—文生图模型本地运行

1、手动下载
   教程见：https://zhuanlan.zhihu.com/p/475260268
2、git clone方式
   git clone https://huggingface.co/THUDM/chatglm3-6b  /huggingface
虽然也是从huggingface进行git下载，但还是出问题，报Checkout 'https://huggingface.co//huggingface/chatglm3-6b/None' for available files.  
说明本地模型加载有异常，去在线寻找地址不对报错。
3、代码自动下载 手动cp到指定目录
```
   from huggingface_hub import snapshot_download
   repo_id = 'THUDM/CogView2'
   model_dir = snapshot_download(repo_id)
   print(model_dir)

   cp -r model_dir $traget_dir
``` 
   modelscope的也类似

pip install modelscope

```
from modelscope import snapshot_download
model_dir = snapshot_download('damo/multi-modal_chinese_stable_diffusion_v1.0')
print(model_dir)
# /xx/.cache/modelscope/hub/damo/multi-modal_chinese_stable_diffusion_v1
 ``` 
N01:~# cp -r   /xx/.cache/modelscope/hub/damo/ modelscope/damo

pip install torch
pip install opencv-python
pip install opencv-contrib-python
# 依赖安装，当前是 python3 （Python 3.10.12）
```
import torch
import cv2
from modelscope.pipelines import pipeline
from modelscope.utils.constant import Tasks

task = Tasks.text_to_image_synthesis
model_id = '/modelscope/damo/multi-modal_chinese_stable_diffusion_v1'
pipe = pipeline(task=task, model=model_id)
output = pipe({'text': '中国山水画'})
cv2.imwrite('/mnt/d/aigc_result/result.png', output['output_imgs'][0])
# 生成图片到指定路径
```

output = pipe({'text': '中国美食的一种饼类', 'num_inference_steps': 15, 'guidance_scale': 7.5, 'negative_prompt':'模
糊的'})

output = pipe({'text': '中国美食的一种，叫老婆饼，实际是面做的饼类', 'num_inference_steps': 50, 'guidance_scale': 7.
5, 'negative_prompt':'模糊的'})