Multimodal Input with Claude

Claude can process images, PDFs, and other documents alongside text. This unlocks a wide range of tasks that were previously impractical — reading screenshots, extracting data from PDFs, analysing charts, and reasoning about visual content.

Image Understanding

Claude can analyse, describe, and reason about images. Supported capabilities:

• Scene description: Describe what is in an image — objects, people, actions, setting. Useful for accessibility, content moderation, and cataloguing.
• Text in images (OCR): Claude reads text visible in photos — handwriting (reasonably well), typed text in screenshots, signs, and printed documents.
• Chart and graph interpretation: Claude can read bar charts, line graphs, pie charts, and scatter plots — describing trends, reading values, and drawing conclusions. It works well for clearly labelled charts; complex or unlabelled charts produce less reliable output.
• Diagram analysis: Architecture diagrams, flowcharts, and technical schematics — Claude can describe what the diagram shows and reason about relationships.
• UI and screenshot analysis: Describe what a screenshot shows, identify UI elements, explain layouts. Useful for bug reports, documentation, and design review.

Claude is not a pixel-level image analysis tool — it perceives visual patterns and text, not individual pixel values. For precise colour analysis, object detection at scale, or measurement extraction, use dedicated computer vision tools.

PDF and Document Analysis

Uploading a PDF directly is the most reliable way to work with document content:

• Claude reads the document text and can answer questions about it, summarise sections, and extract specific information
• Tables in PDFs are generally extracted correctly — complex multi-column layouts sometimes lose formatting
• For research papers: "What is the main claim? What data supports it? What are the limitations?"
• For contracts: "Extract all payment terms, deadlines, and termination clauses"
• For reports: "Summarise the executive summary, then give me the key metrics from the data section"

PDFs with primarily image-based content (scanned documents, image-heavy slides) may not extract text reliably — Claude will work with what it can see visually.

OCR and Text Extraction from Images

For extracting text from images:

• Screenshots: Claude reliably reads text in interface screenshots. Useful for extracting log output, error messages, or code from a screen capture.
• Handwriting: Claude can read reasonably clear handwriting but struggles with highly cursive or messy handwriting. Always verify extracted handwritten text.
• Printed text in photos: Claude reads printed text in photographs reliably if the image is clear, well-lit, and the text is not too small.
• Mixed image/text: For documents that mix images with text (marketing materials, slides), Claude reads both but processes them differently.

Charts and Diagrams

Prompting strategies for charts and diagrams:

• "Describe this chart — what is it showing and what is the main trend?"
• "What is the approximate value for [category] in this bar chart?"
• "Does this data show a statistically significant trend, based on what you can see?"
• "Explain this architecture diagram — what are the components and how do they connect?"

Be aware that Claude estimates values visually — it cannot read exact pixel values. For precise data extraction from charts, convert to CSV/tabular format using a tool like Tabula (for PDFs) before asking Claude to analyse it.

File Size Limits and Supported Formats

Limits vary by plan. Enterprise and API usage may have different limits. Check Anthropic documentation for current specifications.

Checklist: Do You Understand This?

• Claude reads images: scene description, text (OCR), charts, diagrams, and screenshots
• Upload PDFs directly for document Q&A, summarisation, and extraction — more reliable than pasting text
• Chart reading estimates visually — for precise values, convert data to tabular format first
• Handwriting recognition works for clear writing; verify any extracted handwritten text
• Claude is not a pixel-level image analysis tool — use dedicated computer vision for measurements and precise detection