Midjourney v6 vs v7 for Realistic Portraits: What Changed

An independent musician is building the cover for an album built around a fictional 1970s jazz saxophonist. They camp out in a baritone sax, a smoke-filled club, and a face that has to read as a real photograph, not a wax figure. Their v6 prompt, saved from a year ago, now returns something smoother, shinier, and slightly uncanny in v7. They assume the model got worse. It did not; the instructions got stale.

Upgrading the version flag is not a prompt upgrade. v6 and v7 want different kinds of sentences. Copy your old prompt into the new model and you will get a portrait that looks over-retouched, because the words you used to fight v6’s softness now fight a model that was already sharper.

The Symptom: Waxy Skin and Glassy Eyes in v7

The tell is predictable. Running a v6-era portrait prompt under --v 7 tends to produce skin with no pores, eyes that reflect nothing, and hair that reads as a single surface. The composition is fine and the lighting is plausible, which makes it more frustrating: it is close to right, which is the hardest kind of wrong.

Two forces cause it. First, vocabulary drift: words like 4k, photorealistic, highly detailed face, and sharp focus pushed v6 toward a clean render, and v7 already leans clean, so the prompt doubles down. Second, default behavior: v7 ships with more stylization baked in and a stronger instinct for a polished editorial look. You have to pull it back toward a specific photograph instead of nudging it toward realism.

Same Saxophonist, Two Prompts

Both prompts below target the same square cover. One is written with v6 reflexes; the other is tuned for how v7 treats faces, film, and light.

1970s jazz saxophonist portrait, photorealistic, realistic face,
highly detailed, ultra realistic, 4k, sharp focus, cinematic
lighting, album cover, professional photography --ar 1:1 --v 6

Editorial portrait of a 50-something baritone saxophonist, three-quarter
view, head tilted toward the horn. One warm tungsten practical light
from the upper left, deep shadow on the right cheek. Kodak Portra 400,
shot on an 85mm lens at f/1.4. Visible skin texture, faint sweat sheen
on the brow, uneven stubble. Smoke haze catching the light. 1970s
record-sleeve framing, subject slightly off-center, no text.

--ar 1:1 --style raw --stylize 150 --v 7

The v6 prompt lists adjectives. The v7 prompt describes a photograph that already exists in someone’s memory: a camera, a lens, a film stock, one light source, and a specific person. The model has more to hold onto, and the adjectives that were making skin plastic are gone.

What v7 Changed Under the Hood for Faces

You do not need the internal changelog to write good prompts, but three observable shifts explain the rewrite.

• Detail is now assumed, not requested. Asking for ultra realistic, 4k, sharp focus no longer buys realism; it buys gloss. Spend those tokens on texture instead: pores, stubble, sweat, wear.
• Lighting reads as a physical source. v7 renders a named light position and color temperature more faithfully than v6 did. “One warm tungsten practical from the upper left” beats “cinematic lighting,” which now maps to a generic grade.
• Realism needs a reference point. Naming a film stock and a lens communicates the exact grain and depth of field you want, and v7 responds to that specificity. This is also where character reference features help when the same face has to carry across a whole sleeve.

None of this makes v6 obsolete. Magazine-editorial work, fashion spreads, and glossy product photography can still favor v6’s house look. The version choice depends on the job, which is why the parameters below matter as much as the prose.

Parameter Deltas for Portrait Work

The pattern across the table is consistent: turn down the model’s stylistic opinion and turn up the physical specificity. For a wider tour of the version flag itself, the v7 prompt guide covers the flags this build adds.

Rebuilding Your Own v6 Prompt for v7

Do not start from scratch. Convert the prompt you already like with this four-step rewrite, then iterate one variable at a time.

1. Swap adjectives for nouns. Replace photorealistic, 4k, detailed with the camera, lens, and film stock you want the result to look like it came from.
2. Name one light. Delete every generic lighting word and write a single source with a direction and a temperature.
3. Describe texture on the subject. Skin, hair, fabric, and moisture belong in the sentence about the person, not in a trailing keyword list.
4. Add --style raw and lower --stylize. Re-run the same seed range and compare before you touch anything else.

A reusable skeleton that applies the same logic to any realistic portrait looks like this:

Editorial portrait of {{subject, age, expression, pose}}.
{{one light source: type, direction, color temperature}}.
{{film stock}}, {{lens and aperture}}, {{depth of field}}.
Visible {{skin texture and other physical detail}}.
{{environment and atmosphere}}.
{{framing note}}.
--ar {{ratio}} --style raw --stylize {{100-250}} --v 7

Generate three variants of the saxophonist with that skeleton, then change one line at a time. If skin goes waxy again, you reached for an adjective like flawless or sharp focus; delete it and describe texture instead. If the cover starts to look like a generic stock portrait, your stylize is too high or your lighting sentence lost its direction. Both fixes are one edit away, which is the point of writing v7 prompts that read like a camera brief.

Sources & Notes

Observations here are based on public Midjourney v6 and v7 documentation and hands-on use through 2025. Version behavior changes with model updates, so treat the parameter ranges as starting points and re-test. The saxophonist album is fictional and used for illustration only. Internal references: Midjourney v7 Prompt Guide, Cinematic Portrait Prompts, Midjourney Style Parameter Explained.