Skip to content
Documentation
Cookbook
Multimodal: Vision, Charts, and Transcription

Multimodal: Vision, Charts, and Transcription

The multimodal (vision) notebooks cover three common workflows: image understanding, structured extraction from images, and agentic loops around images (e.g., a crop tool).

Overview

Claude’s vision capabilities enable you to work with images in powerful ways:

  • Image Understanding: Analyze and describe images, charts, and documents
  • Structured Extraction: Extract data from forms, invoices, and visual content
  • Agentic Vision Workflows: Combine vision with tools for complex multi-step tasks

This section provides 6 practical notebooks with copyable code examples.

Recommended notebooks

1) Getting started: passing images to Claude

Getting started with vision
Pass images (URL) into Claude

2) Prompting for vision quality

Best practices for vision
Prompting patterns for multimodal reliability

3) Focused tasks: charts, slides, forms

Charts, graphs, and slide decks
Charts and slides workflows
Transcribe documents
Typed, handwritten, forms → structured output

4) Agentic: give Claude a crop tool

Crop tool
Add a crop tool for detailed image analysis

5) Vision + tools

Vision + tools is documented in Tool Use:

  • ../tool-use/vision-with-tools

6) Sub-agents (optional)

Using Haiku as a sub-agent
PDF → images → extraction with a sub-agent

Practical tips

  • Clean inputs beat longer prompts; crop/segment when details are dense.
  • Prefer structured outputs (JSON) for extraction tasks, with validator + retry.
  • Handle privacy and retention carefully: images often contain sensitive data.