Skip to main content
Image understanding lets a model read images and return text. You can send screenshots, photos, product images, receipts, charts, or design drafts, then ask the model to describe, compare, classify, or extract information.
For OpenAI-compatible calls, image understanding usually uses chat/completions with both text and image input. Available models depend on the vision labels shown in the model catalog.

​
Use cases

  • Screenshot QA: explain an app screen, console page, error screenshot, or product UI.
  • Document extraction: extract amount, date, vendor, or fields from receipts and invoices.
  • Product and asset review: identify objects, colors, style, composition, and usage context.
  • Chart understanding: summarize a chart trend or explain visual differences.

​
Suggested workflow

1

Choose a vision model

Filter for models that support image input and copy the model name.
2

Prepare image input

Provide an accessible image URL or Base64 image data according to model requirements.
3

Ask for a specific result

Tell the model whether you need a description, comparison, field extraction, or JSON output.
For reliable field extraction, combine image understanding with structured output.

​
API example

The cURL, Python, and Node.js examples for image understanding are maintained in Vision and image generation. They use gpt-5.4.

​
Next steps