alt-text-llm 1.0

AI-powered alt text generation and labeling tools for markdown content

alt-text-llm

AI-powered alt text generation and labeling tools for markdown content. Originally developed for my website (repo).

Features

• Intelligent scanning - Detects images/videos missing meaningful alt text (ignores empty alt="")
• AI-powered generation - Uses LLM of your choice to create context-aware alt text suggestions
• Interactive labeling - Manually review and edit LLM suggestions. Images display directly in your terminal
• Automatic application - Apply approved captions back to your markdown files

Installation

From PyPI

pip install alt-text-llm

Automated setup (includes system dependencies)

git clone https://github.com/alexander-turner/alt-text-llm.git
cd alt-text-llm
./setup.sh

Prerequisites

The following command-line tools must be installed:

• llm - LLM interface (install instructions)
• git - Version control
• magick (ImageMagick) - Image processing
• ffmpeg - Video processing
• imgcat - Terminal image display

macOS:

brew install imagemagick ffmpeg imgcat
pip install llm

Linux:

sudo apt-get install imagemagick ffmpeg
pip install llm
# imgcat: curl -sL https://iterm2.com/utilities/imgcat -o ~/.local/bin/imgcat && chmod +x ~/.local/bin/imgcat

Usage

The tool provides three main commands: scan, generate, and label.

1. Scan for missing alt text

Scan your markdown files to find images without meaningful alt text:

alt-text-llm scan --root /path/to/markdown/files

This creates asset_queue.json with all assets needing alt text.

2. Generate AI suggestions

Generate alt text suggestions using an LLM:

alt-text-llm generate \
  --root /path/to/markdown/files \
  --model gemini-2.5-flash \
  --suggestions-file suggested_alts.json

Available options:

• --model (required) - LLM model to use (e.g., gemini-2.5-flash, gpt-4o-mini, claude-3-5-sonnet)
• --max-chars - Maximum characters for alt text (default: 300)
• --timeout - LLM timeout in seconds (default: 120)
• --estimate-only - Only show cost estimate without generating
• --process-existing - Also process assets that already have captions

Cost estimation:

alt-text-llm generate \
  --root /path/to/markdown/files \
  --model gemini-2.5-flash \
  --estimate-only

3. Label and approve suggestions

Interactively review and approve the AI-generated suggestions:

alt-text-llm label \
  --suggestions-file suggested_alts.json \
  --output asset_captions.json

Interactive commands:

• Edit the suggested alt text (vim keybindings enabled)
• Press Enter to accept the suggestion as-is
• Submit undo or u to go back to the previous item
• Images display in your terminal (requires imgcat)

Example workflow

# 1. Scan markdown files for missing alt text
alt-text-llm scan --root ./content

# 2. Estimate the cost
alt-text-llm generate \
  --root ./content \
  --model gemini-2.5-flash \
  --estimate-only

# 3. Generate suggestions (if cost is acceptable)
alt-text-llm generate \
  --root ./content \
  --model gemini-2.5-flash

# 4. Review and approve suggestions
alt-text-llm label

Configuration

LLM Integration

This tool uses the llm CLI tool to generate alt text. This provides access to many different AI models including:

• Gemini (Google) via the llm-gemini plugin
• Claude (Anthropic) via the llm-claude-3 plugin
• And many more via plugins

Setting up your model

For Gemini models (default):

llm install llm-gemini
llm keys set gemini # enter API key
llm -m gemini-2.5-flash "Hello, world!"

For other models:

1. Install the appropriate llm plugin (e.g., llm install llm-openai)
2. Configure your API key (e.g., llm keys set openai)
3. Use the model name with --model flag (e.g., --model gpt-4o-mini)

See the llm documentation for setup instructions and the plugin directory for available models.

Output files

• asset_queue.json - Queue of assets needing alt text (from scan)
• suggested_alts.json - AI-generated suggestions (from generate)
• asset_captions.json - Approved final captions (from label)