跳转到主要内容

​
Gemini 文档理解

Gemini 模型支持理解 PDF 文档和其他文档格式，并可以按照指定格式输出结构化结果。 
POST /v1beta/models/{model}:generateContent

​
PDF 文档理解

curl "https://crazyrouter.com/v1beta/models/gemini-2.5-flash:generateContent?key=YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "contents": [
      {
        "role": "user",
        "parts": [
          {"text": "总结这份文档的核心内容"},
          {
            "inlineData": {
              "mimeType": "application/pdf",
              "data": "JVBERi0xLjQKMSAwIG9iago..."
            }
          }
        ]
      }
    ]
  }'

​
格式化输出

使用 responseMimeType 和 responseSchema 控制输出格式：

​
JSON 格式输出

cURL
curl "https://crazyrouter.com/v1beta/models/gemini-2.5-flash:generateContent?key=YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "contents": [
      {
        "role": "user",
        "parts": [
          {"text": "从这份简历中提取关键信息"},
          {
            "inlineData": {
              "mimeType": "application/pdf",
              "data": "JVBERi0xLjQK..."
            }
          }
        ]
      }
    ],
    "generationConfig": {
      "responseMimeType": "application/json",
      "responseSchema": {
        "type": "object",
        "properties": {
          "name": {"type": "string"},
          "email": {"type": "string"},
          "phone": {"type": "string"},
          "education": {
            "type": "array",
            "items": {
              "type": "object",
              "properties": {
                "school": {"type": "string"},
                "degree": {"type": "string"},
                "year": {"type": "string"}
              }
            }
          },
          "experience": {
            "type": "array",
            "items": {
              "type": "object",
              "properties": {
                "company": {"type": "string"},
                "position": {"type": "string"},
                "duration": {"type": "string"}
              }
            }
          },
          "skills": {
            "type": "array",
            "items": {"type": "string"}
          }
        }
      }
    }
  }'

​
响应

{
  "candidates": [
    {
      "content": {
        "parts": [
          {
            "text": "{\"name\":\"张三\",\"email\":\"zhangsan@example.com\",\"phone\":\"138xxxx0000\",\"education\":[{\"school\":\"北京大学\",\"degree\":\"计算机科学硕士\",\"year\":\"2020\"}],\"experience\":[{\"company\":\"某科技公司\",\"position\":\"高级工程师\",\"duration\":\"2020-2024\"}],\"skills\":[\"Python\",\"Go\",\"Kubernetes\"]}"
          }
        ],
        "role": "model"
      },
      "finishReason": "STOP"
    }
  ]
}

​
表格数据提取

Python
response = model.generate_content(
    [
        "提取文档中所有表格的数据，以 JSON 格式返回",
        {"mime_type": "application/pdf", "data": pdf_data}
    ],
    generation_config=genai.GenerationConfig(
        response_mime_type="application/json"
    )
)

import json
tables = json.loads(response.text)

​
文档对比

Python
with open("v1.pdf", "rb") as f:
    v1_data = f.read()
with open("v2.pdf", "rb") as f:
    v2_data = f.read()

response = model.generate_content([
    "对比这两个版本的文档，列出所有修改内容",
    {"mime_type": "application/pdf", "data": v1_data},
    {"mime_type": "application/pdf", "data": v2_data}
])
使用 responseSchema 可以确保模型输出严格符合指定的 JSON 结构，非常适合数据提取和自动化处理场景。
大型 PDF 文档会消耗大量 Token。建议对超过 50 页的文档进行分段处理。