describe-image skill

Overview

This skill describes images using a local AI model to extract visible details, objects, and context. It runs a command-line tool against image files and returns concise, human-readable descriptions for accessibility, indexing, or analysis. It is optimized for privacy and on-premise use since processing happens locally.

How this skill works

The skill calls a local CLI tool that accepts a disk path to the image and a short prompt asking what to describe. The tool runs on a local GPU-backed model and returns descriptive text about the scene, objects, attributes, and any requested focus details. Avoid running multiple instances in parallel to prevent GPU overload.

When to use it

• Generate alt text or accessibility descriptions for images
• Extract scene or object details for indexing and search
• Quickly summarize photo contents for content moderation or triage
• Produce image annotations for datasets or QA checks
• Obtain focused details by asking specific prompts about the image

Best practices

• Provide a concise, specific prompt about what you want described (e.g., "Describe people and emotions").
• Use local absolute disk paths to images to ensure the tool can access files reliably.
• Run requests serially rather than in parallel to avoid overwhelming the GPU.
• Pre-check image formats and sizes; downscale very large images if needed to speed processing.
• Validate outputs for sensitive or ambiguous content before automated use.

Example use cases

• Create accessible alt text for web images by asking for short descriptions.
• Index photo libraries by extracting objects, locations, and visible tags.
• Assist content moderators by highlighting potentially problematic content in images.
• Annotate training datasets with object labels and scene summaries.
• Perform rapid QA on product photos to confirm presence and orientation of items.

FAQ