Why your Midjourney images never look like that

The pictures you want to match look calm because someone separated subject, light and camera. Your own attempts often become a chain of “cinematic, ultra realistic, 8k, beautiful, stunning”. The model averages those words and returns a generic advert. Fewer words, in a fixed order, hit more often.

The order that holds

Write the prompt as parts of a sentence, not as a cloud of mood. Each part has one job:

• Subject: who or what, doing something. “A potter lifts a bowl out of the kiln.”
• Place: a specific room, not a genre. “Small workshop with a shelf and a concrete floor”, not “beautiful atmosphere”.
• Light: one source and one direction. “Afternoon light from the left, a hard shadow on the floor.”
• Camera: focal length and framing, the way you would brief a photographer. “50mm, medium-wide, camera at eye level.”
• Material: something you could touch. Clay, linen, wet wood. That replaces “high-end”.
• Ban: what the picture should not be. “No text, no logo, no second person, no studio backdrop.”

Why the adjective list fails

“Cinematic” and “editorial” at the same time pull in two directions: one look is contrast and colour, the other is flat and plain. “8k” and “ultra detailed” rarely change the composition. They only make surfaces louder. When two style words contradict each other, neither wins. Cut them and keep light plus camera.

Parameters such as aspect ratio stay useful, because they fix the frame, not the mood. For a portrait in a feed, `--ar 4:5`. For a landscape frame, `--ar 3:2`. The current parameter names live in Midjourney’s own help. The sentence before them is the part that stays stable for months.

A potter lifts a still-warm bowl out of a small kiln, hands in linen gloves, eyes on the bowl.
Small workshop, a shelf of unglazed cups, concrete floor, one window.
Afternoon light from the left, a hard shadow of the bowl on the floor, the rest subdued.
Photograph, 50mm, medium-wide, eye level, natural colours, visible clay texture.
No text, no logo, no second person, no studio, no pastel filter. --ar 4:5

How to get closer in three runs

Run one: subject, place and light only. Run two: add camera and material, and do not delete what already worked. Run three: a single ban for whatever still bothers you — usually a second person, text in the image, or a look that reads as stock. If you add five new adjectives in run two, you are back at the start.

The same structure works for a product and a portrait. Only the subject changes. The fuller version, with light, references and series, is in the Midjourney Masterclass. For the line under the picture, Social Prompts helps, and so the series looks like you rather than a stock library, Brand Voice with AI. The same idea for text prompts is in 5 prompt mistakes every beginner makes.