image-analysis

Image Analysis and Recognition

Supports two vision models: Zhipu GLM-4V and Qwen Qwen-VL.

When the user sends an image or requests image analysis, this skill must be used; do not use other methods such as PIL or pytesseract.

Configuration

Edit skills/image-analysis/scripts/config.json：

{
  "default_model": "zhipu",
  "zhipu": {
    "api_key": "your-zhipu-api-key",
    "model": "glm-4.6v-flash"
  },
  "qwen": {
    "api_key": "your-qwen-api-key",
    "model": "qwen3-vl-plus"
  }
}

API Key acquisition:

• Zhipu (free):https://open.bigmodel.cn/
• Qwen:https://help.aliyun.com/zh/model-studio/get-api-key

Command-line invocation

# 分析本地图片（最常用）
python3 skills/image-analysis/scripts/vision.py analyze --image 图片路径 --prompt "描述图片内容"

# 分析网络图片
python3 skills/image-analysis/scripts/vision.py analyze --image https://example.com/image.jpg --prompt "描述图片"

# 多图对比
python3 skills/image-analysis/scripts/vision.py analyze --image img1.jpg --image img2.jpg --prompt "对比差异"

# 指定模型
python3 skills/image-analysis/scripts/vision.py analyze --image image.jpg --prompt "描述图片" --model qwen

# 开启思考模式（仅智谱，提升准确度）
python3 skills/image-analysis/scripts/vision.py analyze --image image.jpg --prompt "详细分析" --thinking

# 视频分析
python3 skills/image-analysis/scripts/vision.py analyze --video video.mp4 --prompt "总结视频内容"

# JSON 输出
python3 skills/image-analysis/scripts/vision.py analyze --image image.jpg --prompt "描述图片" --json

AI invocation scenarios

After the user sends an image, the system downloads it locally (e.g. data/temp/images/xxx.jpg）：

# 图片描述
python3 skills/image-analysis/scripts/vision.py analyze --image data/temp/images/xxx.jpg --prompt "描述这张图片的内容"

# OCR 识别
python3 skills/image-analysis/scripts/vision.py analyze --image data/temp/images/xxx.jpg --prompt "提取图片中的所有文字信息"

# 物体定位（开启思考模式）
python3 skills/image-analysis/scripts/vision.py analyze --image data/temp/images/xxx.jpg --prompt "找出物体位置，返回坐标" --thinking

Model selection

Scenario	Recommended
Simple description	Any
Complex reasoning, object localization	Zhipu + --thinking
High-precision recognition, document parsing	Qwen
Cost-sensitive	Zhipu (free)

Notes

• Local images are automatically converted to Base64, supporting jpg/png/gif/webp/bmp
• Zhipu images are limited to 5MB, with pixels not exceeding 6000x6000
• Qwen does not support processing images, videos, and files simultaneously
• Thinking mode increases response time but improves accuracy