Midjourney
The image model with the strongest style — describable, not instructable.
Best for
- editorial and concept imagery
- mood, texture and light
- striking single frames
Signature move
Describe the light last and the camera before it.
How it wants to be spoken to
Describe the picture, not the request. It is not a chat assistant.
Order matters: subject, then setting, then one detail, then lens, then light, then style.
Use --ar for shape and --no to remove things.
Keep the description under about forty words. Adjectives pile up into mush.
Why it glitches
None of this is you doing it wrong. These are the shapes it reliably falls into — and the fix beside each.
Ignores instructions → it renders descriptions; ask for an image, not an action.
Weird hands and garbled text → reframe, crop, or drop the hands.
Every image looks the same → change the light and the lens, not the adjectives.
Too many objects → fewer things, larger in frame.
A prompt that works there
editorial photo of [subject], [setting], [one specific detail], shot on a 50mm lens, soft window light from the left, muted palette of [two colours], negative space on the right --ar 4:5 --no text, watermark
Bring Tay your actual task and she'll shape it for Midjourney.
Practice with Tay