What --sref does: style references and style codes

--sref is how you say "like this, but not of this". It captures the visual character of an image — colour, medium, texture, lighting — and applies it to whatever your text prompt describes. What it deliberately does not do is copy the objects or people in the reference.

--sref <image URL>
--sref <style code>
--sref random

• V6+model versions supported
• --swcontrols how strongly it applies
• 0–1000that control’s range, default 100

Two things can follow the flag

This is the part that confuses people, because the same parameter takes two quite different arguments:

A code is not a compressed image. It addresses a style that already exists inside Midjourney, which is why codes are portable and image references are not: paste someone's --sref 3808932062 into your own prompt and you get their look. Paste their image URL and you get a broken reference unless the file is still online.

That portability is the reason this collection treats codes as first-class — every code that appears on more than one image gets its own page, so you can see one style across different subjects.

Everyday snapshot, shot on Svema color film --ar 3:2 --v 7 --raw --sref 3808932062 --p n59emtm 72csxc4 --s 50

--sref random

Write random instead of a code and Midjourney picks one for you — then rewrites the prompt with the code it chose, so the result is reproducible after the fact. Rerolls and variations keep it.

The exception is worth knowing: with permutations, --repeat, or V8.1 draft mode, each image gets a different code. That makes --sref random --repeat a style-browsing tool rather than a generation tool.

For browsing deliberately there is the Style Explorer on the Explore page — search it in plain words like "photographic" or "anime", try a code against your current prompt, and like the ones worth keeping.

Mixing references

You can pass more than one, separated by spaces, and you can mix codes with images:

a still life of pears --sref 117928474 2167081065

In Discord each reference can also carry its own weight using the double-colon syntax, which is how you make one dominate:

--sref URL1::2 URL2::1 URL3::1

Note that two codes in one prompt is two styles being blended, not one compound style. This collection indexes them individually for that reason — a page grouped by the pair would bury both.

What to write in the text prompt

With a reference doing the styling, the description's job changes: name the subject, not the treatment. Midjourney's guidance is to keep the text simple and avoid style words that fight the reference.

The failure mode when you ignore this is not an error — it is a blander image, because your adjectives and the reference are both specifying the look and the model averages them.

Old codes that stopped working

The style-reference feature was rebuilt between V6 and V7, and codes from before the change may no longer produce the style they used to. Two ways back: add --sv 4 to use the older V7 model, or switch to V6. Both are covered in style weight and reference versions.

If you are archiving prompts, this is the argument for recording the model version alongside the code — a bare --sref with no --v is a style that quietly means something different a year later.

Practical constraints

Every image in the collection publishes its complete prompt, so the style codes are there to copy. They are the one part of somebody else's prompt that works verbatim in yours — unlike a profile code, which belongs to its author.