How to Write Midjourney 8.2 Prompts: Structure, Parameters, and Examples
A practical Midjourney 8.2 prompt structure: six description fields, where parameters go, how style references divide the work, and templates you can run in Genvizu today.
Genvizu Team5 min read
Midjourney V8.2 has been the default version since July 24, 2026, and its parameter surface is the same one V8.1 used. That is good news for prompt writing: nothing you learned on 8.1 is wasted, and this guide is a structure rather than a list of version-specific tricks.
Two articles, two jobs. For release dates, what changed against V8.1, credit costs, and Genvizu's integration details, read Midjourney 8.2 Is Live. This page is the writing side: how to build a prompt that produces the image you actually described.
Describe the image first, put parameters last
Midjourney's Prompt Basics recommends clear, concise phrases covering subject, medium, environment, lighting, color, mood, and composition. Parameters belong at the end of the prompt, not scattered through the visual description.
A reliable formula:
[subject and defining features] + [medium or photography type] + [environment]
+ [composition and viewpoint] + [lighting] + [color and mood]
+ [material or essential detail] + [parameters]
For example:
solitary coastal hotel on a Mediterranean cliff,
editorial architectural photography, wide establishing view,
late-afternoon side light, chalk white walls and faded cobalt blue,
quiet summer atmosphere, restrained textures --ar 16:9 --stylize 150
Unlike generic padding such as "masterpiece, best quality, 8K," this prompt says what the viewer sees, how the camera frames it, and why the image carries a specific atmosphere. V8.2's stronger default aesthetics do not rescue a prompt that never states the subject.
Control the image with six fields
| Field | Question to answer | Example |
|---|---|---|
| Subject | What is the visual focus? | a cliffside hotel, three orange cats |
| Medium | Photo, illustration, or render? | film portrait, colored-pencil illustration, clay render |
| Environment | Where is the subject? | a foggy station, a greenhouse on the Moon |
| Lighting | Where does the light come from? | overcast diffusion, backlit rim light |
| Color | What is the palette and contrast? | faded sky blue, warm gray with burgundy |
| Composition | What shot and viewpoint? | symmetrical wide shot, overhead view, negative space on the right |
One or two meaningful details per field is usually enough. When a prompt gets long, delete adjectives that do not change your acceptance criteria instead of adding synonyms. In Genvizu the prompt is trimmed at 1,023 characters on a word boundary, so a bloated brief loses its tail rather than its filler.
Separate the job of text from the job of a style reference
The gallery example Hazy Blue Tranquility pairs a short mood description with an --sref style reference:
Airy Hazy Blue --sref 4250870612
Soft, delicate illustration style bathed in faded sky blues and muted whites.
Gentle minimalist aesthetic, soft edges, nostalgic and tranquil mood.
The lesson is the division of responsibility: text defines content, palette, and mood, while the style reference supplies a broader visual language. If the subject has to be accurate, state its identity, quantity, and composition in text. A style code will not preserve them for you, and that has not changed in 8.2.
How the parameters behave
--arsets the aspect ratio. Choose it from the final placement: 16:9 for many covers, 9:16 for vertical content. Genvizu supports 1:1, 4:3, 3:2, 16:9, 3:4, 2:3, and 9:16; a value outside that set falls back to whatever the panel has selected.--stylize(or--s) allows more aesthetic interpretation as it rises. Start low when the composition has to be exact. Because 8.2 changed the default look, re-check the value you used to trust on 8.1.--chaos(or--c) is for exploration. Lower it when you are producing a controlled series.--quality(or--q) accepts 1 or 4. Genvizu sends 1 unless you specify.--rawreduces default styling and improves prompt adherence — useful for products, architecture, and tightly art-directed briefs.--iwsets image weight from 0 to 3 when you supply a reference image, controlling how strongly it pulls the result.--sreftransfers style;--swweights it. Neither locks a person's identity or a product's construction. In Genvizu these two stay in the prompt so Midjourney parses style codes natively, and duplicates are stripped — paste three--srefvalues and only the first survives.--seedis not extracted into an API parameter by Genvizu, and whether the upstream parses a--seedleft in the prompt is unverified. Do not rely on it to reproduce a composition, and never across model versions.
Two Genvizu-specific behaviors are worth knowing. Resolution comes from the panel selector, not from a flag: --hd pasted into a prompt is stripped so that billing always matches the output you receive. And the version comes from the model you pick, so leave --v out of the prompt entirely — select Midjourney 8.2 or Midjourney 8.1 in the picker instead.
Three templates you can adapt
Realistic photography
[subject] in [environment], [photography type], [lens and shot size],
[pose or action], [key-light direction and quality], [palette], [mood],
natural skin and material detail, clean background --ar [ratio] --raw
Brand editorial creative
[product or person], premium editorial campaign,
[geometric composition], [primary brand color] with [accent color],
[hard or soft light] creating [shadow shape],
ample negative space for layout, no readable text --ar [ratio]
Illustration and concept design
[scene and action], [specific illustration medium],
[line or brush treatment], [limited palette], [design era or visual language],
[viewpoint], [mood], coherent shapes --ar [ratio] --stylize [value]
Generate a baseline first, then test composition, lighting, and style one variable at a time. Every Midjourney run in Genvizu returns four images at 23 credits for 1K or 35 for 2K, so a single baseline already gives you four comparisons before you change anything.
Midjourney 8.2 prompt FAQ
Do I need to write --v 8.2?
No. V8.2 is the default version on Midjourney, and in Genvizu the version comes from the model you select in the picker. A --v flag in the prompt just consumes characters from your 1,024-character budget.
Should I write prompts in Chinese or English? Genvizu's Midjourney models require English prompts. Either language is fine for organizing the idea, and English keeps photography, material, art-medium, and camera terms consistent. A practical workflow: write the brief clearly in your own language, convert it into concise English, then verify counts, directions, colors, and parameters by hand. The Midjourney Prompt Helper does that conversion for you.
Why can adding more detail make the result worse? Long prompts tend to contradict themselves — asking for minimal negative space and a background filled with decoration in the same sentence. Rank your requirements as subject, composition, lighting, color, and keep only the details you can actually check in the final image.
Do V8.1 prompts still work on 8.2?
Yes. The parameter surface is unchanged, so old prompts parse. What shifts is the rendered look, which mostly means re-baselining --stylize and re-testing style codes before you regenerate a whole set.
Build a cleaner Midjourney 8.2 prompt
Enter your subject, aspect ratio, and style direction, get back a clear English structure with the parameters in the right place: Open the Genvizu Midjourney Prompt Helper. For version facts, credits, and migration notes, see Midjourney 8.2 Is Live.
