# AI images

**Studio → AI Images.** Describe what you want and generate images that did not exist
before.

## Generate an image

The composer asks **"What would you like to create?"** Describe the image (subject,
style, mood) and choose how many to generate.

The suggestions under the box are starting points for common jobs: website hero, product
spotlight, seasonal artwork, team celebration, announcement banner, abstract pattern.

Generated images land in your media library like any other file: searchable, taggable,
and usable in designs and posts.

### Write an effective description

Be specific about what matters and leave out what does not. Four elements carry most of
the result:

1. **Subject**: what is in the frame
2. **Setting**: where it is
3. **Style**: photographic, illustrated, flat, three-dimensional
4. **Mood and light**: warm, stark, overcast, golden hour

*"A wide, warm photograph of an open-plan office at golden hour, two colleagues talking
at a standing desk, shallow depth of field"* produces something usable. *"Office
picture"* does not.

## Aspect ratios

Pick the shape for where the image is going. The picker names the destination rather
than the ratio:

| | |
|---|---|
| 1:1 | Instagram post |
| 4:5 | Instagram and Facebook feed |
| 9:16 | Stories, Reels, TikTok |
| 16:9 | YouTube, X, LinkedIn |
| 3:2 | Photo, print |
| 21:9 | Banner, cinematic |

## Edit photos with AI

You can also work on images you already have. **Edit with AI** on any file in the
library opens a conversation about that image: describe the change and see it applied.

In the design editor the same capability appears as **Magic eraser & fill**. Paint over
an area and either erase it or describe what should replace it.

When the result is right, **Save as Version** attaches it to the original file and
**Save New** creates a separate file. Use *Save as Version* for corrections and *Save
New* when the result is genuinely a different picture. See
[duplicates](/guide/media/duplicates).

**Revert** removes an edit and everything after it in the conversation.

## Cost

Generating and AI editing spend credits, and generating four images costs four times as
much as generating one. The composer shows the cost before you run it. See [usage and
credits](/guide/admin/usage-and-credits).

## Where generated images can be used

Generated images are ordinary library files, so they work in designs, presentations,
social posts, share links and downloads. What they do not get is a semantic search index
built from a real photograph; instead they carry the prompt that produced them, which is
usually a better description than a caption.

<Callout type="caution">
	Generated images of people depict people who do not exist. If you need a real member
	of your team, use a real photograph. Recognizably synthetic faces in recruitment or
	customer material tend to get noticed.
</Callout>