Verified against Midjourney · 2026-07-20
Write a natural-language editorial portrait brief for Midjourney V7
A connected natural-language portrait brief — subject, pose, setting, light, and lens folded into one flowing description rather than a comma-stacked tag list — matched to how V7's parser actually reads a prompt, not how V6 rewarded keyword stuffing.
The prompt
Ready to copy — highlighted parts are example details you can swap.
SUBJECT
a weathered fisherman in his sixties with sun-creased skin, a grey beard, and a faded navy work jacket
POSE AND EXPRESSION
standing at the bow of his boat, hands resting on the rail, looking toward the horizon, an expression that reads as caught mid-moment rather than posed for a camera, unless the subject line above specifically calls for direct eye contact.
SETTING
a foggy harbor at first light, moored boats fading into the mist behind him
LIGHT
soft directional light from a low winter sun, cutting sideways through the fog. Name the actual source and its direction rather than a mood adjective on its own — Midjourney renders a described source ("low winter sun through fog") far more convincingly than a floating quality tag like "cinematic lighting" that names no physical cause.
CAMERA AND LENS
Shot on a Canon EOS R5 with an 85mm f/1.2 lens, shallow depth of field, the harbor behind him falling into soft blur.
MOOD
quiet, contemplative, documentary realism rather than a styled fashion shoot
SKIN AND MATERIAL HONESTY
Real skin has visible pores, uneven tone, and asymmetry; real fabric creases where a body actually bends it. Do not let the description drift toward smooth, airbrushed, or symmetrical — if that plastic look shows up in the grid, it is Midjourney's own beautification bias reasserting itself, not something this brief asked for.
WRITE THIS AS ONE CONNECTED DESCRIPTION
Combine the subject, pose, setting, light, and lens lines above into a single flowing paragraph, in that order, the way you would brief a photographer standing next to you — not as a list of disconnected fragments separated by commas with no relationship between them. V7 was retrained specifically to parse this kind of connected phrasing and use the relationship between clauses — this light, falling on this subject, doing this pose, in this setting — rather than treating each comma-separated fragment as an isolated, unrelated instruction competing for attention with every other fragment.
WHAT TO LEAVE OUT
Do not stack generic quality boosters — "8k," "hyperrealistic," "masterpiece," "highly detailed," "award-winning photography." These did real, measurable work nudging V5 and V6's keyword-matching behavior; on V7 they mostly do nothing useful, and stacking several at once can actively push the image toward an over-processed, over-sharpened look that fights the specific, grounded description already written above instead of reinforcing it.
IF THE FIRST GRID READS TOO SAFE OR TOO SIMILAR ACROSS ALL FOUR
Raise chaos on the next run rather than rewriting the description — chaos controls variation between the four grid images, not how literally any single one follows the text, so a same-feeling grid is a chaos problem, not a wording problem.
PARAMETERS
--ar 4:5 --stylize 150 --v 7
OUTPUT
The collapsed one-paragraph description above, followed directly by the parameter line, ready to paste into Midjourney as a single continuous prompt.Customize
Optional — swap in your own details for the highlighted parts above.
Why this works
V7 was retrained to parse a full natural-language brief the way a human photographer would read a shot list, not a flat bag of isolated tags — this is the single largest prompting-behavior change from V6, and it is the reason this brief is written as connected clauses in a fixed order (subject, then pose, then setting, then light, then lens) rather than a comma-separated pile. On V6, stacking quality tags like "8k, hyperrealistic, masterpiece" reliably nudged output quality upward because the model was substantially keyword-matching; on V7 those same tags do measurably less and can actively fight the model's already-strong default realism by pulling it toward an over-processed, over-sharpened look nobody asked for. Writing the light as a named physical source with a direction, rather than a mood adjective floating with no cause, matters because V7 conditions on causal relationships between clauses — "low winter sun cutting through fog" tells it where the shadow falls and how the fog scatters that specific light, while "cinematic lighting" names an aesthetic category with no physical information to render from, so the model falls back on its most statistically common interpretation of that phrase instead of the scene actually described. The skin-and-material-honesty instruction targets a real, well-documented default bias: left unguided, Midjourney's aesthetic training pulls portraits toward smoother, more symmetrical, more airbrushed skin than the photographic references the prompt is actually describing, because that beautified look is heavily overrepresented in the images the model was trained to consider "good." Naming the failure mode explicitly — visible pores, uneven tone, asymmetry, creased fabric — gives V7 a concrete target to render instead of letting its own aesthetic bias quietly override a documentary brief with a beauty-filter one. Separating chaos from stylize in the guidance at the end closes the most common Midjourney parameter confusion directly: a grid of four images that all look too similar to each other is a chaos problem, since chaos governs variation between the four outputs in one grid, not how far any single image drifts from the literal prompt — which is stylize's job. Telling a user to raise chaos rather than rewrite the prompt when the grid feels repetitive saves an entire wasted generation cycle spent second-guessing wording that was never the actual lever.
What you get back
A single editorial-looking portrait matching the described subject, pose, and light — expect real photographic variation between the four grid options (skin texture, exact framing, micro-expression) rather than four near-identical copies, with the fog and lens character carrying through consistently.
Verified against
Midjourney v7 · 2026-07-20
Changelog
- 2026-07-20 — Initial publish, written and verified against Midjourney v7 as a connected natural-language brief.
Need this built into your business?
If a prompt isn't enough — what Scult builds, built and maintained for you — that's Scult's day job.
EXPLORE WHAT SCULT BUILDS
