# Image captions (/docs/enhance/captions)





For learners who are blind or have low vision, an image without a description is a hole in the content. This step generates **image descriptions (alt text)** for the figures in your book — context-aware descriptions that actually describe the image, informed by the surrounding content.

## Decisions to make [#decisions-to-make]

* **Which images carry meaning?** Diagrams, illustrations, and charts need descriptions. Purely decorative elements do not — a description would only add noise for screen reader users.
* **Is the description accurate and useful?** AI-generated descriptions are a strong starting point, but you know the pedagogical intent of each image. Review them, especially where the image teaches something.

## Configure and generate [#configure-and-generate]

Captions need your book's pages laid out first — if [Storyboard](/docs/convert-pdf/storyboard) isn't finished yet, the stage tells you so instead of letting you run it. Once it's ready, open the **Image Captions** stage and set:

* **Grade Level** — Early, Middle, or Advanced — so descriptions match your audience's reading level.
* **Custom Instructions** (optional) — steer tone, focus, or vocabulary. Autofill drafts a starting point from the book's title, language, and summary.

A live sample preview on the right shows how a description reads before you run anything. When you're ready, select **Run Captions**.

<img alt="The Image Captions stage before it has run, showing Grade Level and Custom Instructions settings alongside a sample captions preview" src="__img0" />

## Review and edit [#review-and-edit]

Once captions are generated, every image becomes a card you can filter (**All** / **Captioned** / **Decorative**, each with a count), search by caption text or image ID, and jump between pages if the book has several. Select an image to open it full-size; select its caption text to edit it in place, with `Esc` to cancel and `Cmd`/`Ctrl`+`Enter` to save.

Marking an image **Decorative** is how you act on the first decision above — it excludes that image from screen readers entirely and clears the need for a caption. A small dot next to each image ID shows whether its caption came from the AI or from a manual edit, and the version picker lets you preview and restore an earlier draft.

<Callout title="A good image description">
  Answers the question: what would a learner miss if they could not see this image? Not more, not less.
</Callout>