Midjourney v8.2 & v8.1
Midjourney is the image model art directors reach for when the aesthetic has to carry the work. Every native flag is a real control here, so you steer the render with numbers instead of retyping the prompt and hoping.
Midjourney work shared publicly on X by the community. Nothing is cropped to a uniform grid — the shapes are the point. Click any image to open it full size.
Model capabilities
Midjourney trades literal prompt obedience for aesthetic range. Its parameters control how much of that house style lands on your image, and each one moves a specific dial.
Try a prompt
Comic linework, screenprint posters, painterly realism, flat editorial. Midjourney covers all of it from a text prompt. v8.2 is the newest render and v8.1 sits beside it, reading prompts tightly and holding small details, so switching is a genuine creative choice.
--v 8.2 --style raw --ar 3:2
Point Midjourney at an image you like and its palette, linework and texture carry into whatever you generate next. The subject stays yours. Style weight sets how literally Midjourney follows the reference, which is how a set of images becomes one body of work.
--sref your-reference.jpg --sw 400
A 1:2 vertical panorama, a 21:9 banner, an odd 7:4 crop for a layout. Midjourney takes a free-form width-to-height value, and the ratio shapes composition rather than just cropping — the same prompt frames its subject differently at each shape.
--ar 1:2
Stylize decides how hard Midjourney pushes its own taste over your literal words, on a 0 to 1000 scale. Chaos sets how far apart the four returned images land. Weird pushes toward the unconventional; raw pulls the house style back. Together they turn a lucky Midjourney render into a repeatable one.
--s 250 --c 15 --raw
One Midjourney prompt returns a grid of up to four images, not a single answer. That is the model's working rhythm: generate, read four takes, follow the one that surprised you. Quality runs from a rough 0.25 pass to a detailed 2, with an HD tier for finals.
--q 2 --hd
What it is
Midjourney is the image model illustrators, concept artists and art directors use to define what AI illustration looks like. Version 8.2 became the default render in July 2026, with 8.1 alongside it. Both return four images from one prompt.
What separates Midjourney from general-purpose image models is taste rather than resolution. Midjourney applies a strong aesthetic of its own, and its two dozen parameters control how much of it you get. Those parameters began as flags typed into a chat box; here they are ordinary form controls.
Pick the right model
Two models built for different jobs. Choosing by reputation instead of by task is how good prompts produce disappointing images.
| Capability | Midjourney v8 | GPT Image 2 |
|---|---|---|
| Best for | Illustration, concept art, posters, character design | Literal prompts, product shots, images with text |
| Images per prompt | Up to 4 to choose from | 1 per prompt |
| Aspect ratio | Any width:height value | 14 fixed presets |
| Model versions | Both current v8 renders | Single current model |
| Style reference (transfer a look) | ||
| Numeric style controls | ||
| Queue speed choice | ||
| Reliable text inside the image | ||
| Precise edits to an existing image |
Neither column wins outright. Reach for Midjourney when the look carries the work, and for GPT Image 2 when the image must hold readable text or follow a literal instruction.
Where it earns its keep
Four jobs where Midjourney's opinionated look is the feature rather than the obstacle.
Interiors, props and mood in one pass, with enough incidental detail to suggest a world instead of illustrating a sentence. Raise chaos and the four Midjourney images diverge, turning one prompt into four directions to explore.
Flat colour, hard shadows, one dominant hue. Midjourney produces this without being asked twice. Pair a Midjourney style reference with a low stylize value when a client already has a visual language you must stay inside.
Silhouette-first composition, limited palettes and atmosphere hold up at key-art scale. Midjourney's weird and stylize dials get you off the safe centre of a genre without a paragraph of adjectives.
Plain backgrounds, full-body poses and consistent proportions across a cast. A Midjourney style reference plus a reused seed keeps the rendering language fixed while the figure changes.
How to use it
No Discord, no slash commands, no flags to memorise.
Tell Midjourney what you want in plain language, then pick the aspect ratio. Ratio changes composition rather than just the crop, so choose the shape you actually need before polishing the wording.
Raise stylize for a stronger Midjourney signature, lower it to stay closer to your words. Add a style reference to match a look you already have. Anything you leave untouched is simply not sent to the model.
One Midjourney prompt returns up to four images. Pick the one closest to your intent, note its seed, and adjust a single dial at a time instead of rewriting the whole prompt.
Getting more out of it
Most disappointing Midjourney results come from fighting the model instead of using its controls.
Generic results mean stylize is too low. Ignored details mean it is too high, or you want raw mode. Most of the prompt rewriting people do is really a stylize problem.
Two runs of the same adjectives drift apart. A style reference does not. Point Midjourney at an image and let style weight control how closely it follows.
Midjourney frames the same prompt differently at 1:2 and at 3:2. Lock the shape you need, then iterate on wording.
The three Midjourney queue speeds change how long you wait, nothing about the render. The image is identical, so there is no quality reason to avoid the slow queue while exploring.
FAQ
Versions, references, and what Midjourney is not good at.
Ready when you are
Midjourney v8.2 and v8.1, with every native flag as a real control instead of syntax you have to remember.
