A style reference moves the look of one image onto another without moving its contents. Midjourney defines it in one sentence and then documents the syntax, which is where most guides stop reading and start inventing weights.
Values below are quoted from the official page; the --sw 125 tiers elsewhere are somebody’s testing ladder.
What a style reference is, in Midjourney’s words
The definition is one sentence: “A Style Reference is a way to capture the visual vibe of an existing image and apply it to your new Midjourney creations. It doesn’t copy objects or people, just the overall style—like colors, medium, textures, or lighting—helping you achieve a consistent visual theme.”
Two conditions sit beside it: style references work with “Midjourney versions 6 and later”, and “you need a text prompt in combination with your Style Reference”.
The page also separates a request from a description: “the look of this image but a dog” against “detailed portrait of a dog” — instruction words, the same rule the Midjourney prompt guide gives generally.
How to attach one: the web panel and --sref in Discord
On the website
Pick or upload the image from the panel behind the image icon in the Imagine bar, then “drag and drop your image from the uploads library into the Style Reference section”. The lock icon pins it for later prompts, and only the Style Reference slot changes the look on its own. The quick-start guide covers that bar.
In Discord
“To use a Style Reference in Discord, start by adding the --sref parameter to the end of your prompt, then pasting your image URL.” Several images are separated by a space. The URL must already be online; accepted endings are .png, .gif, .webp, .jpg and .jpeg.
Style weight: --sw, and its two published numbers
The style weight parameter --sw “allows you to control how strongly the style of your reference influences your new image”. It arrives with exactly two figures: a range of “any value between 0 and 1000” and a default of --sw 100.
Two details matter. In V7, “style weight has more impact when used with style codes than with images”, so one value pushes a code further than a picture. And --sw is not compatible with Moodboards.
Weights per image with ::
Discord weights several references separately: --sref URL1::2 URL2::1 URL3::1. That is the page’s only example; the multiplier rules live on Multi-Prompts & Weights.
Style codes, --sv and the Style Creator
A style code replaces the uploaded image: “Each numerical code produces a unique visual style, and can be used by adding --sref <code> to the end of your prompt.” --sref random picks one, and “random” turns into a real code on submit. A prompt that carries a code holds it through “rerun/reroll or variations”, while permutation, --repeat and V8.1 Draft prompts give “each image … a different style code”.
--sv chooses the algorithm behind them. V7 ships six versions, --sv 6 by default and --sv 4 as “the old V7 sref model (prior to June 16, 2025)”; V6 ships four with --sv 4 as its default. Two constraints follow: --sref random and style codes work only with --sv 4 and --sv 6, and --sv is incompatible with Moodboards too. V8.2 keeps the feature, as the Midjourney V8.2 guide notes.
What the Style Creator actually does
The Style Creator builds a code from the images you pick and the ones you skip, and it is “exclusive to midjourney.com”. Previews spend GPU time, run on “Midjourney version 7” and get cheaper with --draft. Expect noise early: “Most styles stabilize after 5–10 rounds”, and later rounds bring only small changes. Sessions cannot be reopened.
Worked example: three --sref prompts with documented values
Every value below is from the official page.
Case one: one image at the default weight
a glass greenhouse at dusk, rows of orchids, soft ambient light, muted palette --ar 16:9 --sref URL1 --sw 100URL1 stands for a live image URL. --sw 100 is the documented default, written out so the value is visible; omitting it changes nothing.
Case two: an older style code on --sv 4
moss-covered stone bridge over a slow river, overcast light, muted greens --ar 3:2 --sref <code> --sv 4The page warns that “older style codes may not produce the same styles anymore” after the V6-to-V7 update, and gives two fixes: “add --sv 4 to your prompt (uses the old V7 model) or switch to V6”. Since --sv 6 is the V7 default, the flag appears only when you go back on purpose.
Case three: two references with per-image weights
ceramic teapot on a linen cloth, soft window light --ar 4:3 --sref URL1::2 URL2::1 --sw 100Discord syntax: the multiplier follows each URL, ::2 counting double against ::1. In V7 the same weight acts differently on codes and images, so compare before picking a number — and add a Moodboard and --sw drops out.
What the official page does not publish about style reference
- A recommended
--swvalue per subject; the page gives the range, the default, and nothing between. - An
--svversion list for V8.2: the tables cover V6 and V7, and V8.2 points to a page with no sref table. - A cap on references per prompt: the four-image limit belongs to the Edit Model, not to
--sref. - A numeric range for the
::multipliers, plus any effect a reference has on GPU cost. The cost table is by task type, and the Midjourney API guide shows how to log it per job.
Frequently asked questions
Does a style reference copy the subject? No. The page says it “doesn’t copy objects or people”, only colours, medium, textures and lighting.
What is the default --sw? --sw 100, on a documented range of 0 to 1000.
How do I reuse one look across many prompts? Keep the style code and pass it as --sref. Reruns, rerolls and variations hold it; permutation, --repeat and V8.1 Draft jobs do not.
Can my own uploaded image become a style code? No: “You cannot create a style code based on an uploaded image.” Attach it as a reference each time, or build one in the Style Creator.
Why does an old style code look different now? The feature changed between V6 and V7; --sv 4 restores the old V7 model, and V6 is the documented alternative.