Z.AI Vision

Overview

This skill provides Z.AI's GLM-4.6V vision model capabilities for analyzing images and videos through Python scripts. Use it for OCR, UI design analysis, technical diagrams, error screenshots, data visualizations, and video scene understanding.

Quick Start

Prerequisites

1. Install the Z.AI SDK:

pip install zai-sdk

1. Set your API key:

export ZAI_API_KEY='your-api-key'

The API key is required for all vision operations.

Basic Image Analysis

python3 /root/clawd/zai-vision/scripts/vision_analyze.py <image_path> "<prompt>"

Example:

python3 /root/clawd/zai-vision/scripts/vision_analyze.py screenshot.png "Describe this UI"

Basic Video Analysis

python3 /root/clawd/zai-vision/scripts/video_analyze.py <video_path> "<prompt>"

Example:

python3 /root/clawd/zai-vision/scripts/video_analyze.py clip.mp4 "What's happening?"

Capabilities

Image Analysis

OCR / Text Extraction

python3 /root/clawd/zai-vision/scripts/vision_analyze.py doc-scan.jpg "Extract all text"

UI Design Analysis

python3 /root/clawd/zai-vision/scripts/vision_analyze.py ui-mockup.png "Analyze this UI design and list all components"

Error Diagnosis

python3 /root/clawd/zai-vision/scripts/vision_analyze.py error.png "What error is shown and how do I fix it?"

Technical Diagrams

python3 /root/clawd/zai-vision/scripts/vision_analyze.py architecture.png "Explain this architecture diagram"

Data Visualization

python3 /root/clawd/zai-vision/scripts/vision_analyze.py chart.png "What insights does this chart show?"

Video Analysis

Scene Description

python3 /root/clawd/zai-vision/scripts/video_analyze.py demo.mp4 "Describe what's happening"

Note: Video analysis works best with short clips (≤8MB). Videos are processed frame-by-frame.

Parameters

Example with parameters:

python3 /root/clawd/zai-vision/scripts/vision_analyze.py image.jpg "Describe this" \
  --temperature 0.3 \
  --max-tokens 500 \
  --json

Integration with Safe Scripts

When running in the /root/clawd workspace, use clawd-run for safety:

clawd-run /root/clawd/zai-vision/scripts/vision_analyze.py image.png "Analyze"

This provides automatic backups, validation, and timeout protection.

Error Handling

Missing API key:

❌ ZAI_API_KEY environment variable not set

Set it: export ZAI_API_KEY='your-key'

Image not found:

❌ Image file not found: /path/to/image.jpg

Verify the file path.

SDK not installed:

❌ zai-sdk not installed

Install with: pip install zai-sdk

Common Patterns

Pattern 1: Batch Process Multiple Images

for img in /path/to/images/*.png; do
  python3 /root/clawd/zai-vision/scripts/vision_analyze.py "$img" "Describe this image"
done

Pattern 2: Extract and Save JSON

python3 /root/clawd/zai-vision/scripts/vision_analyze.py image.jpg "Analyze" --json > output.json

Pattern 3: Specific Analysis Type

Code from screenshot:

python3 /root/clawd/zai-vision/scripts/vision_analyze.py code.png "Extract the code and explain what it does"

Form field extraction:

python3 /root/clawd/zai-vision/scripts/vision_analyze.py form.jpg "List all form fields and their types"

Brand guidelines check:

python3 /root/clawd/zai-vision/scripts/vision_analyze.py design.png "Check if this follows brand guidelines"

Tips for Best Results

1. Specific prompts: "List all UI components" > "What's this?"
2. High quality images: Better resolution = better understanding
3. Temperature: 0.2-0.5 for factual, 0.7-1.0 for creative
4. Video limits: Keep videos ≤8MB for best performance
5. Handle errors: Always check return codes and error messages

Resources

Scripts

• scripts/vision_analyze.py - Image analysis with GLM-4.6V
• scripts/video_analyze.py - Video analysis (frame-by-frame)

References

• references/API.md - Complete API documentation and examples

When to Use This Skill

Use this skill when you need to:

• Analyze screenshots, photos, or images
• Extract text from images (OCR)
• Understand technical diagrams or charts
• Diagnose errors from screenshots
• Analyze UI designs or mockups
• Describe video scenes
• Process visual content programmatically

For more detailed API information, see references/API.md.