Skip to content

Pictures from the assistant ​

The assistant can not only read pictures (see The AI assistant on the side → Images) but also draw them: an illustration for a post, a strip for an e-mail, the title picture of a presentation or a picture for a greeting card or invitation. Drawing is done by the image model your instance has set up. On killBottleneck Cloud it is on; on your own server you connect your own service (see On your own server). Where the instance cannot draw, the drawing buttons do not appear at all.

New image: write what should be in the picture, choose what it is for and a style. Painting is preselected.

Draw a picture in the conversation ​

Just tell the assistant what you want: "Draw an illustration: a small team in a workshop planning an order" or "Make me an e-mail banner with an autumn mood". The assistant writes a detailed description for the image model itself, and a picture card appears below the answer. Drawing starts right away and usually takes a few seconds. On the card you can download the picture or redraw it (same brief, a new result).

Only ask for text in the picture when you really want it ("…with the words Merry Christmas"). The model handles a short caption, but now and then a letter comes out wrong; then click Redraw. Pictures with text are only available where the instance has them turned on.

New image in Office ​

Without a conversation, you can draw a picture in the Office panel (the documents icon in the assistant's header), on the Pictures tab, with the New image button:

  1. What should the image show? A few words, e.g. "a carpentry workshop in winter, warm light". The assistant turns them into a detailed English description, so the result keeps the look and the details.
  2. Text in the image (optional, only where the instance supports it), e.g. "Merry Christmas".
  3. What for: this sets the shape of the picture.
    • Presentation (landscape),
    • E-mail (strip): a wide strip for the top of an e-mail,
    • Post (square),
    • Story (portrait).
  4. Style:
    • Painting (warm light): the default, a painted scene with warm light,
    • Illustration (as in presentations): flatter and lighter,
    • Photo,
    • Cartoon,
    • Watercolour,
    • Pencil sketch,
    • 3D.
  5. Draw: the finished picture opens straight in the viewer.

The form stays filled in after drawing, so you can easily try the same description in another style. The Draw button stays inactive until you write a description; a style is always selected.

The Pictures tab and the viewer ​

The Pictures tab holds all your pictures from the assistant, newest first, with a search by title and description. Click a picture to open the viewer on the left next to the panel. The arrows page through the pictures, ⛶ makes the viewer full screen and Esc closes it. The top bar of the viewer has these buttons:

  • Download: the JPEG to your computer,
  • Redraw: same brief, a new result (a picture used in a Page cannot be redrawn, because the page would lose it; draw a new one instead),
  • Keep for good: pins the picture, so the clean-up leaves it alone,
  • Delete: a picture used in a document or a page cannot be deleted. Remove it there first.
The viewer: arrows to page through, download, redraw and keep for good. At the bottom you see when the clean-up would delete the picture.

Clean-up after 90 days ​

Pictures you don't use anywhere are deleted automatically after 90 days. The viewer shows when ("Unused — will be deleted on …"). The clean-up never touches pictures that are:

  • used in an Office document or in a Page, including their previous version,
  • set to Keep for good.

When you delete a conversation, its unused pictures are deleted too.

Pictures in presentations, cards and invitations ​

When a presentation is made in Office, the assistant draws a title picture for it on its own and, where it helps, up to 3 pictures for content slides. A greeting card, invitation or poster gets one picture. A certificate has no picture. If you don't want pictures, say so in the brief ("no pictures").

In the Documents panel, a document has a Pictures section. For each picture you can Redraw or Remove it, and add a picture to another slide with your own description. The PowerPoint and the PDF then download with the pictures.

Illustrations in Pages ​

When writing a Page, the assistant can add up to 3 illustrations, usually to the opening and to parts about people, places or products. The illustration description matches the look of the page (light or dark, colours). Tables, numbers and charts stay in HTML and don't belong in pictures. The illustrations also open from the shared /p/… link.

A banner in the morning summary e-mail ​

Everyone turns on the Morning summary by e-mail for themselves in Notification settings. It is a short summary of the day: what is burning, what is due today and this week. At the top of the e-mail there is a picture of the day (by season and weekday), one for the whole instance. If the picture fails, the e-mail arrives without it. The e-mail only comes on days when you have open work.

On your own server ​

On your own server you connect your own image service with the standard OpenAI Images API (POST …/images/generations): OpenAI itself or a compatible service. In .env:

bash
KB_IMAGE_PROVIDER=openai
KB_IMAGE_TOKEN=sk-…                      # the service's API key
KB_IMAGE_URL=https://api.openai.com/v1   # default; for another service its …/v1 address
KB_IMAGE_MODEL=gpt-image-1-mini          # default
KB_IMAGE_TEXT=1                          # optional: allow pictures with text

The service must return a JPEG. For gpt-image models the app asks for it itself. If the service returns another format, the card says so right away ("…returned a format other than JPEG"). Without KB_IMAGE_TOKEN drawing stays off. All variables are in the reference Environment variables → Drawing pictures.

Next ​

Open source under the MIT license · © Tengo s.r.o.