Skip to main content

AI Agents

Making images with AI

Updated October 2, 2026 · 3 min read

Making images with AI

Your agents can make images. Ask for one in plain words, in a chat, in a room, or as part of a task an agent runs, and it appears in the reply.

Asking for an image

Describe what you want the way you would to a designer. The more you say about the subject, setting, style, colours and mood, the closer the first result will be.

  • "Make an illustration of a lighthouse at dawn, soft pastel colours."
  • "Create a banner for our summer barbecue, bright and friendly."
  • "Design a logo for a bakery called Crumb, on a transparent background."

If the image must show exact words, put them in quotes: "a poster with the headline 'GROW TOGETHER'". Short headlines come out best.

Shapes

Images come in three shapes. Say which you want, or let the agent pick:

  • Square (1024 by 1024), for icons, logos and most social posts.
  • Landscape (1536 by 1024), for banners, website headers and slides.
  • Portrait (1024 by 1536), for posters and stories.

Transparent backgrounds

Logos, icons and stickers can have a truly transparent background, so they sit cleanly on any page or slide. Ask for "a transparent background".

Changing an image

Say what should be different: "make it bluer", "same scene at night", "remove the text". The agent edits the most recent image in the conversation and keeps the rest of it as it was.

To change a particular image, use its Edit button. The image goes into your message box; type what should change and send. An edit keeps the image's shape unless you ask for a different one.

You can edit a picture of your own too. Attach a photo and ask, for example, "turn this into a watercolour painting".

Viewing and downloading

Each image has buttons over its corner: Edit, Download and Expand, and on the Agents page Open in editor, which turns the image into a design you can add text and shapes to. Click the image, or Expand, to see it larger without leaving the chat.

Download saves a PNG, which opens everywhere and keeps a transparent background. The small arrow beside it offers two other formats:

  • JPEG, the smallest file. JPEG can't be transparent, so a transparent background turns white.
  • WebP, the original file as it's stored.

Usage

An image counts toward the same weekly usage allowance as your chats, and costs about as much as a few ordinary messages. An edit costs a little more for each picture it starts from. Your usage on the Billing page reflects it.

If there isn't enough usage left for an image, the agent tells you, and nothing is made or charged.

Limits

  • An agent makes up to four images in one reply. Ask for more in your next message.
  • Images count toward your workspace's file storage (see Files and uploads).
  • Some requests are declined by the image service's safety rules. When that happens, nothing is made or charged, and you can reword the request.

Tips for better images

  • Ask for "the picture only, no added text" if you don't want words in the image. Requests like "a social post" or "a banner" often come back as a finished design with headlines.
  • For product photos, ask for an unbranded product so no real brand's logo appears.
  • If the first try isn't right, say what to change rather than starting over. Edits keep what you liked.

Was this article helpful?

Back to AI Agents

Support

Still need help?

Can't find what you're looking for? Start a conversation and our support team will help.

Open your laptop.The work is already done.

Free to start. No credit card required.