$ ~/registry/skill/microsoft-image-vision

SKILL

image-vision

Name: image-vision
Author: microsoft

Analyze images, extract text, and answer visual questions with LLM vision models.

Stars

★ 9

Source

GitHub

Updated

2026-07-09

// securityLow risk

Prompt only, no code
Open source

Running security audit…

Credentials
Network egress
Code execution
Data access
Supply chain

// install

Copy the install command and let the AI configure it · recommended for beginners

Please install the "image-vision" skill from askskill:
1. Download https://raw.githubusercontent.com/microsoft/amplifier-bundle-skills/main/skills/image-vision/SKILL.md
2. Save it as ~/.claude/skills/image-vision/SKILL.md
3. Reload skills and tell me it's ready

// download

Download SKILL.mdInstall manifest ↗

// examples

Image content analysis

Prompt

Please analyze this image and list the key objects, scene context, visible text, and the likely theme it conveys.

Expected

A structured image description including main elements, visible text, and a theme assessment.

Text extraction from image

Prompt

Extract all visible text from this image in reading order and mark any unclear parts.

Expected

OCR text output organized in order, with notes on uncertain or low-confidence sections.

Compare differences between images

Prompt

Compare these two images and summarize differences in layout, color, text, and object changes, highlighting the most obvious ones.

Expected

A clear comparison of differences, useful for reviewing version changes or design revisions.

// docs

Image Vision Analysis

Overview

Analyze images using state-of-the-art LLM vision models. Use the provided scripts for most tasks - custom code only needed for advanced scenarios.

Workflow Decision Tree

First time using this skill?

→ Read setup.md for one-time environment and API key setup

Simple image analysis (most common)

→ Use "Quick Start" canned scripts below

Batch processing or multi-turn conversations

→ Read patterns.md for advanced patterns

Something failing?

→ Check setup.md for troubleshooting

Quick Start (Use Wrapper Scripts)

ALWAYS use the wrapper scripts - they handle venv setup automatically:

# Simple analysis (auto-creates venv on first use)
./vision-analyze.sh <provider> <image_path> <prompt>

# Robust analysis (auto-fallback if provider times out)
./vision-analyze-robust.sh <image_path> <prompt> [timeout_seconds]

The wrapper scripts automatically:

Create venv if it doesn't exist
Install required SDKs
Use venv Python (no manual activation needed)
Handle errors gracefully

Example usage:

# Analyze a UI screenshot (Anthropic Claude)
./vision-analyze.sh anthropic screenshot.png "Describe any UI bugs or issues you see"

# Extract text (Google Gemini - fastest)
./vision-analyze.sh gemini document.jpg "Extract all text from this image"

# Robust analysis with auto-fallback (tries Gemini → Anthropic → OpenAI)
./vision-analyze-robust.sh photo.png "Describe this image in detail"

# With custom timeout (default is 60 seconds)
./vision-analyze-robust.sh large-image.png "Analyze this" 120

Advanced: Direct Script Usage (Not Recommended)

If you need to call the Python scripts directly, you MUST use the venv Python:

# ❌ WRONG - uses system Python, will fail
python examples/anthropic-vision.py image.png "prompt"

# ✅ CORRECT - uses venv Python
./.venv/bin/python examples/anthropic-vision.py image.png "prompt"

For agents: Always use the wrapper scripts to avoid setup issues.

Provider Comparison

Provider	Model	Best For	Speed	Cost
Anthropic	claude-sonnet-4-5	Latest, balanced quality/speed	Fast	$$
Anthropic	claude-3-opus	Highest quality (older)	Slow	$$$
Anthropic	claude-3-haiku	Fastest, simple tasks	Very Fast	$
OpenAI	gpt-5	Latest flagship model	Fast	$$$
OpenAI	gpt-4.1	High-volume production	Fast	$$
Gemini	gemini-2.5-flash	Latest, excellent balance	Very Fast	$
Gemini	gemini-2.5-pro	Large images, best quality	Medium	$$
Azure	(deployment-based)	Enterprise, compliance	Varies	Varies

Supported Image Formats

JPEG/JPG - Most common
PNG - With transparency
GIF - Static or animated
WEBP - Modern format

Max sizes:

Anthropic: 5MB per image
OpenAI: 20MB (auto-resizes)
Gemini: Varies by model (1.5 pro handles very large)

Common Use Cases

# UI/UX Analysis - High-level layout and spacing
./vision-analyze.sh anthropic app-screenshot.png \
  "Analyze this UI for accessibility issues and suggest improvements"

# Bug Identification (use robust for auto-fallback)
./vision-analyze-robust.sh error-state.png \
  "What's wrong with this interface? Describe any visual bugs."

# Content Moderation
./vision-analyze.sh openai user-upload.jpg \
  "Does this image contain inappropriate content? Yes or no, and explain."

# Document Understanding (Gemini is fastest)
./vision-analyze.sh gemini invoice.png \
  "Extract the total amount, date, and vendor name from this invoice"

# Design Review - Layout, color, hierarchy (not typography details)
./vision-analyze-robust.sh mockup.png \
  "Provide design feedback on this mockup. Consider layout, color hierarchy, and spacing."