Image generation

The Images tab generates a picture from a description. Type what you want, send, and the image comes back in the page.

Writing a prompt that works

Free image models reward specificity about composition far more than adjectives about quality. In rough order of what actually changes the output:

1. Subject and action — "a border collie mid-jump catching a frisbee"
2. Setting — "on a wet suburban lawn at dusk"
3. Shot and framing — "low angle, shallow depth of field, subject filling the

frame"

1. Style — "documentary photograph" or "flat vector illustration"

Words like "beautiful", "high quality", and "8k" do very little. Naming a lens, a light source, or a medium does a lot.

Where the image goes

Nowhere. The image is returned in the response and your browser holds it. Use the download button to keep it.

That is a deliberate decision, not an oversight. Storing generated images would mean a media directory, a file-serving route with its own ownership checks to get wrong, and a growing pile of data that a database backup silently does not cover. Handing the file to you instead means there is nothing to leak and nothing to lose.

So: if you close the tab without downloading, the image is gone. Reloading the page will not bring it back.

Its own tab, not the chat box

Asking for a picture and asking a question are different requests, and guessing between them from the wording would be worse than a tab. A message containing the word "draw" is usually not a request for a drawing.

Limits

• There is a daily cap per account. It exists because image models on free tiers have

small, shared quotas — one heavy user would otherwise take the feature away from everyone.

• Text inside images comes out unreliably. This is a limitation of the models, not a

setting; if you need a headline on an image, add it yourself afterwards.

• There is no editing, inpainting, or img2img. Generate again with a changed

description.

Not the same as attaching an image

Generating and reading are separate. If you want the assistant to look at a picture you already have, attach it in a normal chat instead — the paperclip in the message box. Unizune AI routes those messages only to engines that can actually see, so an attachment is never quietly ignored.