DALL·E
OpenAI's image maker — follows a written brief well and is the best of them at rendering short text.
Best for
- images with legible words in them
- simple illustrations
- editing an image through conversation
Signature move
Ask for the same image at three different crops.
How it wants to be spoken to
Write full sentences instead of comma lists.
Front-load the subject, then style and composition.
Ask for edits in the same conversation — it keeps the thread.
Keep any text in the image very short.
Why it glitches
None of this is you doing it wrong. These are the shapes it reliably falls into — and the fix beside each.
Won't copy a living artist's style or a real person's face → describe the qualities instead.
Busy scenes lose detail → fewer objects, bigger.
Text longer than a few words still breaks → shorten it.
Composition drifts on edits → restate what must stay the same.
A prompt that works there
A flat vector illustration of [subject doing something], with [two details], on a [colour] background, centred, no text. Then show me the same image cropped for a square social post.
Bring Tay your actual task and she'll shape it for DALL·E.
Practice with Tay