Generating images
Have an image created from a description right in the chat, enlarge and download it, and learn where the image model runs.
With Create image, an image model turns your description into a picture, for example an illustration for a client letter or a title image for a presentation. The image appears right in the chat.
Create an image
Turn on image mode
Click the plus sign Add in the message box and choose Create image. The line below the entry names the image model. If the message box is empty, a Create image chip appears in it, with a note underneath saying that your next message will create an image.
Describe the image
Describe the image and send the message. For example:
Watercolour of a small bakery in a Swiss old town in the early morning, warm light, landscape format, no text.
If you have already typed the description before choosing Create image, the image is created straight away.
Wait for the image
Generating image… appears under the message box, and the image then appears in the chat. Image mode stays on in case you want more images. The cross on the chip (Leave image mode) takes you back to the normal chat.
Enlarge and download
Click the image to enlarge it. The large view has Download image and Close. To download without opening it, use the icon at the top right that appears when you hover over the image. The file name is derived from your description.
Good to know
- Only the description counts. Each image is made from the message you request it with and nothing else. Earlier messages, attached files and brains are not included. To change an image, describe it again in full, with the change you want.
- No regenerating. There is no regenerate arrow under an image. Send the description again.
- Data protection applies here too. If your workspace is set to Pseudonymise, Custodos replaces detected names and identifiers in the description with placeholders before the image model sees them. If it is set to Block, a description containing protected data is refused.
- Credits. An image uses considerably more credits than a normal answer. Admins see the figure per image under Models.
- Stored in the chat. Images stay in the chat. Anyone allowed to read the chat sees them too. When you delete the chat, the images are deleted with it.
Where the image model runs
There are currently two image models. The one your workspace uses is named in the menu under Create image.
| Image model | Processing | Available in the region |
|---|---|---|
| Gemini Image | At Google in the EU (Netherlands) | European Union, EU + Switzerland, Global |
| GPT Image | At OpenAI worldwide, no residency commitment | Global only |
There is no image model for the Switzerland region. An admin decides which model creates images, see Models and regions.
When Create image is not available
The Create image entry stays visible in the menu but is greyed out, and the line below it states the reason.
| Situation | What you see |
|---|---|
| The chat runs in security mode | A note that the feature is not available in security mode, because no image model meets its guarantees. |
| No image model allowed or available in the region | "No image model is available in this workspace." |
| The workspace's credits are used up | "Image generation is paused until the monthly credits reset. Chat keeps working." |
Unlike chat, image generation does not fall back to a cheaper model when credits run out, because there is no cheaper one.
Best practices
- Describe the subject, style, colours and format: "Flat-style illustration, shades of blue and grey, square" gives more targeted results than "a nice picture".
- If you do not want any text in the image, say so explicitly.
- Leave names and personal data out of the description. Pseudonymised names only reach the image model as placeholders.
- If you need several variants, keep image mode on and change one aspect of the description at a time.
Frequently asked questions
The line below the entry states the reason: security mode, no image model in the workspace, or credits used up. The entry is also greyed out briefly while an answer or an image is in progress.
No. Image generation only works from your description. A model can read and describe attached images, but not change them. See Attaching files and images.
Try again. If the provider refuses the description, rephrase it. Failed attempts cost no credits.