The Muse Decides

The Structure

Every prompt starts with a headline and follows the same order: headline, technical details, subject, pose, style, lighting, background, spacial depth. The technical line carries the camera, lens, ISO, aperture, flash sync speed, and focus distance. The lighting line carries the modifiers, the angles, and the power ratio. The background line carries the material and the HEX colors.

The test added more lines as the details demanded them: wardrobe, seating, framing, and aspect ratio.

Test One: The Missing Wardrobe

My first prompt described the woman, the pose, and the light. It said nothing about clothing. Midjourney dressed her in generic black.

I added a wardrobe line: heavy matte velvet in deep oxblood, a narrow band of aged linen at the collar, one gold drop earring. The next batch showed the oxblood velvet, the linen collar, the earring, and the hand at the collarbone, exactly as written.

The model follows what is written. It invents what is not.

Test Two: The Seat

The pose said “seated” and nothing more. A prompt that says seated leaves the seat blank, so I filled it: a black high stool in the center of the room. One line removed one invention.

Test Three: Framing and Aspect Ratio

Framing words and aspect ratio describe the same thing, so they have to agree. A full length standing figure takes a two to three vertical frame. A seated figure takes four to five vertical. A reclined figure on a bed takes four to three horizontal. Every prompt in the series states its ratio in the technical line.

Fifteen results in a row came back as close-ups. [Add what resolved it, with the before and after images.]

Test Four: One Piece of Furniture Per Prompt

I wrote a prompt with a bed and a prompt with a couch. When both furniture pieces appeared in one prompt, the system was confused. Each prompt now names one piece of furniture and its parts: headboard and pillows for the bed, armrest and seat cushions for the sofa.

Test Five: What Sits Between You and the Model

I wrote the prompts in full English with no symbols, then ran them through Grammarly before pasting them into Midjourney. Grammarly converted spelled-out numbers into numerals and symbols, turned “hex color two B two one one A” into #2B211A, and compressed my labeled sections into one paragraph.

Midjourney then showed a default block of text and a set of parameters attached to every job: chaos 5, aspect ratio 4:5, raw, stylize 750, and HD, plus a line of “no” phrases that included “no closeups, no tight cropping.” I had not written any of it. It was a saved default in my settings. Check what travels with your prompt before you judge the result.

The Look

One combination worked flawlessly: an aged canvas backdrop shifting from #2B211A to #3A2C22, a dark hardwood floor fading to near black, a pale yellow cropped graphic tee with orange floral detailing, a light wash denim pleated micro mini skirt, black slingback pumps, a floral urn at frame right, candles at frame left. The dark ground makes the pale yellow and light denim the brightest elements in the frame, and the orange in the florals sits in the same warm family as the canvas.

With the look locked, I change one variable at a time: pose, then angle, then seat. The look stays constant, so each result shows what one line does.