Claude Vision: Analyzing Images, Charts & Visual Documents

What is Claude Vision?

Claude Vision is Claude's multimodal capability that allows it to understand and analyze images. Beyond simple object recognition, Claude can:

• →Read text (OCR) in documents, screenshots, and photos
• →Interpret charts and extract data
• →Analyze technical diagrams (UML, architecture, workflows)
• →Describe images in detail with context
• →Compare multiple images in a single request

Sending an Image via the API

Method 1: Base64

The base64 method encodes the image directly in the request. Ideal for local images.

import anthropic
import base64

client = anthropic.Anthropic()

# Encode the image in base64
with open("sales-chart.png", "rb") as f:
    image_data = base64.standard_b64encode(f.read()).decode("utf-8")

response = client.messages.create(
    model="claude-sonnet-4-20250514",
    max_tokens=1024,
    messages=[{
        "role": "user",
        "content": [
            {
                "type": "image",
                "source": {
                    "type": "base64",
                    "media_type": "image/png",
                    "data": image_data
                }
            },
            {
                "type": "text",
                "text": "Analyze this sales chart. What trends do you observe?"
            }
        ]
    }]
)

print(response.content[0].text)

Method 2: URL

The URL method points to a publicly accessible image.

response = client.messages.create(
    model="claude-sonnet-4-20250514",
    max_tokens=1024,
    messages=[{
        "role": "user",
        "content": [
            {
                "type": "image",
                "source": {
                    "type": "url",
                    "url": "https://example.com/dashboard-screenshot.png"
                }
            },
            {
                "type": "text",
                "text": "Describe this dashboard and identify the visible KPIs."
            }
        ]
    }]
)

Method Comparison

Formats and Limits

Technical limits:

• →Max size: 20 MB per image
• →Images per request: Up to 100
• →Resolution: Automatically resized if too large
• →Tokens: Calculated based on image resolution

Token Calculation per Image

Images consume tokens proportional to their size:

Multi-Image: Comparison and Analysis

Claude can analyze multiple images simultaneously, opening up powerful use cases.

import base64

def load_image(path):
    with open(path, "rb") as f:
        return base64.standard_b64encode(f.read()).decode("utf-8")

response = client.messages.create(
    model="claude-sonnet-4-20250514",
    max_tokens=2048,
    messages=[{
        "role": "user",
        "content": [
            {
                "type": "image",
                "source": {"type": "base64", "media_type": "image/png",
                           "data": load_image("design-v1.png")}
            },
            {
                "type": "image",
                "source": {"type": "base64", "media_type": "image/png",
                           "data": load_image("design-v2.png")}
            },
            {
                "type": "text",
                "text": """Compare these two design versions:
                1. What are the major differences?
                2. Which version better follows UX principles?
                3. Improvement suggestions for the chosen version."""
            }
        ]
    }]
)

Multi-Image Use Cases

OCR and Text Extraction

Claude excels at text recognition (OCR), well beyond simple character reading.

# OCR of a scanned document
response = client.messages.create(
    model="claude-sonnet-4-20250514",
    max_tokens=4096,
    messages=[{
        "role": "user",
        "content": [
            {
                "type": "image",
                "source": {"type": "base64", "media_type": "image/jpeg",
                           "data": load_image("invoice-scan.jpg")}
            },
            {
                "type": "text",
                "text": """Extract the following information from this invoice:
                - Invoice number
                - Date
                - Vendor name
                - Subtotal
                - Tax
                - Total amount
                - List of items with unit price
                
                Return the result as structured JSON."""
            }
        ]
    }]
)

OCR Capabilities

Chart and Diagram Analysis

response = client.messages.create(
    model="claude-sonnet-4-20250514",
    max_tokens=2048,
    messages=[{
        "role": "user",
        "content": [
            {
                "type": "image",
                "source": {"type": "base64", "media_type": "image/png",
                           "data": load_image("financial-chart.png")}
            },
            {
                "type": "text",
                "text": """Analyze this financial chart:
                1. What are the axes and their units?
                2. What is the overall trend?
                3. Identify inflection points or anomalies.
                4. Extract approximate values of key data points.
                5. Summarize in 3 insights for a CFO."""
            }
        ]
    }]
)

Types of Analyzable Visuals

Best Practices

Optimizing Analysis Quality

1. →Use high-quality images — Sufficient resolution so text is readable
2. →Frame the subject well — Avoid overly wide images with important content in small print
3. →Prefer PNG for screenshots — No lossy compression
4. →Orient the image correctly — Claude handles rotation, but correct orientation is better

Optimizing Costs

1. →Resize before sending — A 1000×1000 image is sufficient for most analyses
2. →Compress JPEGs — 80% quality is sufficient for OCR
3. →Limit the number of images — Only send necessary images
4. →Use appropriate resolutions per use case:

Writing Effective Visual Prompts