Images your bot makes

Member·5 minutes to read

What this does for you

Your bot can draw a picture for you and put it in its answer. You describe what you want, and the bot asks an image model to make it. The picture lands in your conversation as a file you can open, download and share, and it is always marked as made by AI.

The bot does not make videos yet. It can make still pictures and diagrams only.

Before you start

The bot needs an image model your organisation allows. Image models run outside the EU today, so an organisation whose residency policy keeps processing in the EU has none (see What can go wrong).
Each picture is charged to the run that made it, so it counts against the run's, bot's or routine's spending limit.

Steps

To ask for an image:

In the chat, ask the bot to make the picture and say what should be in it. Be specific: the image model sees only the description the bot writes for it, not the rest of your conversation.
Making a picture costs money, so the bot asks you first. Approve the request to let it go ahead. A rule you saved to always allow a tool never covers image generation: you approve every picture.
The picture appears in the bot's answer, with the bot's short description of it underneath. Click the picture to see it larger, or Download to save it.

A single answer can carry at most two pictures, and your organisation has a daily number of pictures it can make.

The synthetic mark

Every picture the bot makes carries a Generated mark in its corner. Hover over it, or read it with a screen reader, and it says This image was generated by AI. The mark is part of the picture's record, not a setting: a picture without it is never written to your conversation. A diagram the bot draws from its own description carries the same mark, and it explains that the diagram was drawn by the app rather than by an image model.

The provenance line

Under the picture, click How was this made to see where it came from:

Row	What it tells you
`Made by`	The model that made the picture, by its display name.
`Service`	The service that ran it, and whether the picture was made on this deployment or `Processed outside the EU`.
`When`	The time it was made, in UTC.
`From`	A short fingerprint of the exact description the model was given (`prompt` and twelve characters). The description itself is not kept on the picture, but the same description always gives the same fingerprint.
`Size`	The picture's width and height in pixels.

The panel ends with Generated by AI.

A worked example

Asking for a picture for a slide:

Make me a picture for the title slide: a lighthouse on a rocky coast at dusk, flat illustration style, no text.

The bot asks to make one picture. You approve it, and the answer shows the lighthouse with the Generated mark in its corner and the bot's description under it. How was this made shows the model that made it, Processed outside the EU, the time and the prompt fingerprint.

What can go wrong

When the bot cannot make a picture, it tells you in a sentence why it did not make the picture. The ones you may see:

Message	What it means	What to do
`No image model is available under this organisation's residency policy.`	Your organisation's data-residency rules leave no image model the bot may use. Image models run outside the EU today, so an organisation set to keep processing in the EU has none.	Ask an administrator whether the residency policy or the model allowlist should allow one.
`A single answer can carry at most two pictures.`	The answer already has two pictures or diagrams.	Ask for the next picture in a new message.
`This organisation has made all the images it can make today.`	The daily number is used up.	Try again tomorrow.
`There is not enough of this run's budget left to make a picture.`	The spending limit on this run, bot or routine is nearly reached.	Raise the limit, or ask in a new message.
`The image provider would not make that picture. Try describing it differently.`	The image model refused the description.	Describe the picture differently.
`The image provider could not be reached just now. Nothing was charged.`	The image model did not answer.	Try again in a moment.