Stable Diffusion WebUI使用python调用WD 1.4标签器（tagger），获取并处理

Stable Diffusion为秋叶大佬的绘世2.2.4
webUI api后台：http://127.0.0.1:7860/docs
  数据获取结果，python代码在文章尾部

1. 1girl: 0.9883618950843811, 98%
2. solo: 0.9468605518341064, 94%
3. horns: 0.9203381538391113, 92%
4. braid: 0.7536494731903076, 75%
5. brown_hair: 0.7361204624176025, 73%
6. sensitive: 0.7181869745254517, 71%
7. looking_at_viewer: 0.6558270454406738, 65%
8. long_hair: 0.6555134654045105, 65%
9. portrait: 0.5619801878929138, 56%
10. hair_ornament: 0.5276427268981934, 52%
11. lips: 0.5271897912025452, 52%
12. realistic: 0.47530364990234375, 47%
13. brown_eyes: 0.44382530450820923, 44%
14. fur_trim: 0.44058263301849365, 44%
15. red_hair: 0.4004508852958679, 40%
16. upper_body: 0.39194822311401367, 39%
17. mole: 0.35748565196990967, 35%
18. general: 0.2813188433647156, 28%
19. questionable: 0.004140794277191162, 0%
20. explicit: 0.0005668997764587402, 0%

复制代码

使用/tagger/v1/interrogate，先使用get方法获取model模组有十多个，然后把json_data提交上去就可以了。记得把图片转码为base64。本文章仅用于测试，请仔细阅读api docs，model和threshold按照需求调整即可

1. import requests
2. import base64
3. from collections import OrderedDict
4. from PIL import Image
7. url = 'http://127.0.0.1:7860/tagger/v1/interrogate'
8. image_path = 'D:/code/image/6.jpg'
9. model = 'wd14-convnext'
10. threshold = 0.35
12. #确认照片为上传照片
13. image = Image.open(image_path)
14. image.show()
16. # 将图片转换为Base64字符串
17. with open(image_path, 'rb') as file:
18.     image_data = file.read()
19.     base64_image = base64.b64encode(image_data).decode('utf-8')
21. # 构建请求体的JSON数据
22. data = {
23.     "image": base64_image,
24.     "model": model,
25.     "threshold": threshold
26. }
28. # 发送POST请求
29. response = requests.post(url, json=data)
31. # 检查响应状态码
32. if response.status_code == 200:
33.     json_data = response.json()
34.     # 处理返回的JSON数据
35.     caption_dict = json_data['caption']
36.     sorted_items = sorted(caption_dict.items(), key=lambda x: x[1], reverse=True)
37.     output = '\n'.join([f'{k}: {v}, {int(v * 100)}%' for k, v in sorted_items])
38.     print(output)
40. else:
41.     print('Error:', response.status_code)
42.     print('Response body:', response.text)

复制代码

来源：https://blog.csdn.net/weixin_42375356/article/details/130904119
免责声明：如果侵犯了您的权益，请联系站长，我们会及时删除侵权内容，谢谢合作！