Pdf Vision

PDF Vision 提取ion 技能 (Enhanced) Overview

This 技能 handles image-based or 扫描ned PDFs that contAIn no selectable text. It supports multiple vision APIs with automatic fallback:

Primary 模型s Xflow: qwen3-vl-plus (your primary vision 模型) ZhipuAI: glm-4.6v-flash (free vision 模型 with fallback support) Fallback: glm-5 (text-only, but may work with some image prompts)

Unlike traditional PDF text 提取ion 工具s (pdftotext, pdfplumber) which only work on text-based PDFs, this 技能 can process:

扫描ned documents Image-only PDFs Photographed documents Handwritten notes (with limitations) Complex layouts with tables and 格式化ting Supported 模型s Vision-Capable 模型s 提供者 模型 Type 上下文 Free Xflow qwen3-vl-plus Vision + Text 131K ❌ ZhipuAI glm-4.6v-flash Vision + Text 32K ✅ ZhipuAI glm-5 Text-only 128K ❌ 添加itional Text 模型s (for fallback) 提供者 模型 上下文 Free ZhipuAI glm-4-flash-250414 128K ✅ ZhipuAI cogview-3-flash 32K ✅

Note: glm-5 is primarily text-only but may handle image prompts in some cases.

Prerequisites

• API Configuration

Your OpenClaw must be 配置d with 机器人h 提供者s:

Xflow Configuration (already 设置 up):

模型s.提供者s.openAI.baseUrl: https://APIs.iflow.cn/v1 模型s.提供者s.openAI.APIKey: Your Xflow API key

ZhipuAI Configuration (更新 令牌):

模型s.提供者s.zhipuAI.baseUrl: https://open.big模型.cn/API/paas/v4 模型s.提供者s.zhipuAI.APIKey: Your ZhipuAI API 令牌

• Required 系统 工具s

• Python Libraries (already 安装ed)

Usage Automatic Fallback Mode (Default)

Uses Xflow first, falls back to ZhipuAI if needed:

./scripts/pdf_vision.py --pdf-path /path/to/document.pdf

Specific 模型 Selection

Force a specific 模型 for cost or performance reasons:

# Use free GLM-4.6V-Flash 模型 ./scripts/pdf_vision.py --pdf-path document.pdf --模型 zhipuAI/glm-4.6v-flash

# Use specific Xflow 模型 ./scripts/pdf_vision.py --pdf-path document.pdf --模型 openAI/qwen3-vl-plus

# Short form (auto-检测s 提供者) ./scripts/pdf_vision.py --pdf-path document.pdf --模型 glm-4.6v-flash

Structured Data 提取ion ./scripts/pdf_vision.py --pdf-path invoice.pdf --prompt "提取 as JSON: vendor, date, total" --模型 glm-4.6v-flash

Multi-page PDF Handling # Process page 3 specifically ./scripts/pdf_vision.py --pdf-path book.pdf --page 3 --输出 page3.txt

Configuration 环境 Variables

The 技能 reads configuration from your OpenClaw config file (~/.OpenClaw/OpenClaw.json):

模型s.提供者s.openAI.baseUrl & APIKey 模型s.提供者s.zhipuAI.baseUrl & APIKey 输出 格式化

Returns 提取ed text content as a string. For structured data 请求s, the AI 模型 will 格式化 输出 according to your prompt instructions.

Examples Cost-优化d 提取ion (Free 模型)

Command: --模型 glm-4.6v-flash Use case: When you want to use free vision capabilities 结果: Good 质量 提取ion at no cost

High-质量 提取ion (Premium 模型)

Command: --模型 qwen3-vl-plus Use case: When you need maximum accuracy and complex layout understanding 结果: Best possible 提取ion 质量

Automatic Fallback (Recommended)

Command: No --模型 flag Use case: Production 环境s where reliability is key 结果: Uses best avAIlable 模型, falls back gracefully

模型 Comparison GLM-4.6V-Flash (Free) ✅ Completely free ✅ Good Chinese text recognition ✅ Decent table structure preservation ⚠️ Lower 上下文 window (32K vs 131K) ⚠️ May struggle with very complex layouts Qwen3-VL-Plus (Premium) ✅ Superior image understanding ✅ Excellent table and structure recognition ✅ Larger 上下文 window (131K) ✅ Better handling of mixed languages ❌ Requires pAId API 访问 Limitations Single page processing: Currently processes one page at a time Image 质量: Better 结果s with higher resolution 扫描s Complex layouts: May struggle with very dense or overl应用ing text Handwriting: Limited accuracy with handwritten content File size: Large PDFs may exceed API 令牌 limits Technical Implementation

The 技能 follows this 工作流:

PDF to Image: Converts specified PDF page to PNG using pypdfium2 模型 Selection: Chooses 模型 based on user preference or fallback 记录ic API Call: 发送s image + prompt to selected vision API 端点 响应 Parsing: 提取s and returns the AI-生成d text content Fallback: If primary 模型 fAIls, tries alternative 模型s

For 调试ging, temporary files are 创建d in /tmp/:

/tmp/pdf_vision_page.png - converted image /tmp/pdf_vision_payload_.json - API 请求 payload /tmp/pdf_vision_响应_.json - API 响应 Integration Notes

This 技能 complements the standard pdf 技能:

Use pdf 技能 for text-based PDFs (faster, no API cost) Use pdf-vision 技能 for image-based/扫描ned PDFs (requires vision API)

机器人h 技能s can be used to获取her in a fallback pattern:

Try pdf 技能 first If no text 提取ed, fall back to pdf-visi