Chat

Generating images

Have a picture created from a description, in image mode or right in the chat, and choose its shape.

An image model turns your description into a picture, such as an illustration for a client letter. In new workspaces image generation is on with an image model in the EU; an admin sets which one under Models.

Create an image

Switch on image mode

Click Add in the message box and choose Create image. Below the message box you see which image model works and roughly how many credits a picture takes.

Choose the shape

In the chip, choose the Picture shape: Square, Landscape, Portrait or Widescreen (16:9).

The message field in “Create image” mode with four image formats to choose from and the note that the next message will create an image with Gemini Image.

Describe and send

Watercolour of a small bakery in a Swiss old town in the early morning, warm light, no text.

The picture usually takes about 20 seconds. Image mode stays on until you click Leave image mode on the chip.

You can also write "Create an image of …" in an ordinary chat. A model with tools then creates the picture with your workspace's image model; a model without tools points you to Create image.

Change and download

The image model takes the chat's newest picture as its starting point, including a photo you attached. Write, for example, "the same card, but in blue". Create another version under a picture makes another take from the same description.

Click the picture to open it large. There you find Download image.

Image models

Image modelProcessing
Gemini ImageAt Google in the EU
FLUX.2 ProBy Black Forest Labs, run by Microsoft in the EU. Its content filter often refuses ordinary personal names, for example on business cards.
GPT ImageGlobally at OpenAI, only if your workspace allows the Global region

There is no image model in Switzerland. A workspace that processes only in Switzerland therefore creates no images, and in security mode image generation is off. More in Where your data lives.

Names in a picture

A placeholder cannot be restored inside a picture. In image mode, names therefore go into the picture as you type them; while data protection is active, the message box says so. In an ordinary chat the model sees protected details only as placeholders, and Custodos leaves those out of the picture. If a name should appear in the picture, use image mode. More in Protecting personal data in chat.

When Create image is greyed out

The line under Create image gives the reason:

ReasonWhat helps
The chat runs in security modeStart a chat without security mode, if your workspace allows it.
No image model is allowed or available in the regionAn admin enables one; admins see the link Manage image models.
Your groups do not include image generation, or your group keeps your data in Switzerland (groups on request)An admin changes the group settings.
Credits or budget used upWait until credits are available again. Unlike chat, image generation does not fall back to a cheaper model.

Limits

Up to 4,000 characters per description and up to 6 pictures per minute, image mode and chat together. A picture takes considerably more credits than an answer.

Best practices

  • Describe subject, style, colours and mood, such as "flat illustration, shades of blue and grey".
  • If the picture should contain no text, say so. If it should, give the text word for word.
  • Change one thing per round: "same picture, but evening light".

Next steps

On this page