Turn an image into a description
Use AI to describe image content in natural language, with a summary of the main subjects, objects, actions and visible scene.
Common uses
Quick image summaries
Understand the main content of an image without manually describing it.
Accessibility support
Generate a starting point for image descriptions and alt text.
Image collections
Create descriptive text that can help organise or review large sets of images.
Research and analysis
Use automatic descriptions as one additional way to inspect visual content.
How does image description work?
A vision-language model analyses the image and generates text based on the visual patterns it recognises.
The model attempts to describe the most relevant visible content rather than listing every pixel or object.
Descriptions are generated automatically and may contain mistakes or omit details.
Limitations
Descriptions may be incomplete
Small, subtle or background details can be omitted.
The model can make mistakes
Objects, actions or relationships may occasionally be described incorrectly.
Context can be misunderstood
The model only has access to the image and may not understand the broader real-world context.
Generated descriptions are not factual verification
A description represents the model's interpretation of the image and should not be treated as independent evidence that an event or claim is true.
Language
Descriptions are currently generated in English.
Frequently Asked Questions
What does the image description include?
The description generally focuses on the main subjects, objects, actions and visible setting.
Can I use the description as alt text?
It can be a useful starting point, but important accessibility text should still be reviewed by a person and adapted to the context in which the image appears.
Why did the description miss something?
Vision-language models prioritise what they consider visually important. Small or less prominent details may be omitted.
Can the description identify people?
The tool describes visible content but is not designed to establish the real-world identity of a person.
Can the description be wrong?
Yes. Automatically generated descriptions can contain errors, so important claims should be verified independently.