In Midjourney V8.1, --sw (style weight) controls one thing: how strongly a --sref reference asserts its look over the rest of your prompt. The range is 0 to 1000 and the default is 100. At --sw 0 the reference is effectively switched off; push the number up and the reference takes more and more authority over color, texture, lighting and finish — until, near the ceiling, it can overrule the subject you described.
That is the whole mechanic. The interesting part is the judgement: where on the 0–1000 dial a given reference actually wants to sit. A distinctive, high-contrast reference asserts itself plainly at moderate weight; a softer aesthetic needs more weight to register at all. Below, two real V8.1 families — one rendered at --sw 300, one at --sw 500 — make the difference concrete.
Style weight does not change what is in the frame — it changes how much of the reference's visual identity survives into the result. Think of it as a tug-of-war between two forces: your prompt text (the subject, the scene) and your --sref (the aesthetic). --sw is the rope's center mark.
Low weight lets the prompt lead and the reference tint. High weight lets the reference lead and the prompt becomes the subject the reference is applied to. The sweet spot is rarely the extremes — it is wherever the reference's signature reads cleanly without erasing your idea.
| --sw value | Reference authority | When to reach for it |
|---|---|---|
0 | None — reference cancelled | Sanity check: prompt-only output, no style applied. |
100 (default) | Light tint | A nudge of the reference's mood while the prompt clearly leads. |
300 | Clear, balanced | The reference reads unmistakably, the subject stays legible. |
500 | Dominant | The reference drives palette, finish and lighting across the whole frame. |
1000 | Overpowering | Reference saturates everything; prompt detail can get steamrolled. |
One variable at a time. To read what --sw is doing, hold the prompt and the references constant and move only the weight. Change the prompt and the reference together and you have learned nothing about --sw — you have just made a different image.
Here are two real V8.1 series. The first sits at --sw 300: a cyber / tech-noir family built from two stacked references. At 300 the reference asserts its indigo-and-graphite palette, its engraved texture and its hard chiaroscuro plainly — but the subject (the figure, the pose, the gear) stays in command of the composition.
The second series runs at --sw 500: a Buddhist / Gongbi family built from a different pair of references. At 500 the reference is no longer a tint — it is the author. Mineral-pigment color, gilded filigree, sinuous gongbi linework and the auratic halo flood the entire image. The subject is now something the style is happening to, rather than the other way round. That is the felt difference between 300 and 500: at 300 the reference advises; at 500 it governs.
Two honest caveats. First: the cyber family above is its own reference set and the Buddhist family is another, so this is a comparison of weights across two looks, not the same reference dialed from 300 to 500. To isolate --sw exactly, run one reference set and one prompt, then re-roll at 100 / 300 / 500 / 800 and read your own grid. Second: 300 and 500 are not magic constants. A reference with a loud, distinctive signature (the engraved cyber set) already reads strongly at 300; a subtler reference might need 400–500 to register the same presence. Match the weight to the reference, not to a rule of thumb.
Both examples above stack two references on a single --sref flag. This is where a lot of confusion lives, so be precise: --sw is one weight applied to the blended style of everything on that flag — not a per-reference slider. When you write --sref A B --sw 300, Midjourney first fuses A and B into a combined aesthetic, then applies 300 to that fusion. There is no way to say "60% of A, 40% of B" through --sw alone.
The payoff of stacking shows up as consistency. Each family above — the cyber series and the Buddhist series — was generated from the same reference pair across multiple prompts, and every member lands in the same visual world. That is the real reason to weight references deliberately: a fixed reference set plus a steady --sw is how you produce a coherent series, not just a single lucky frame. If one reference is overpowering the blend, the fix is not --sw — it is dropping or swapping that reference, or balancing the set with a second one that pulls the other way.
On the reference URLs: the s.mj.run links above are the real references behind these images, but they are Midjourney-hosted and personal — treat them as provenance, not as permanent public assets. They can expire, and they encode someone's private uploads. The transferable lesson is the look and the weight, which you can rebuild from your own references.
Notice that every prompt above also carries --stylize 1000 (you can write it --s 1000). New users routinely conflate the two. They are not the same control, and turning one will not do the job of the other.
| Dial | Governs | Range / default | Needs a --sref? |
|---|---|---|---|
--sw | How strongly your reference overrides the prompt | 0–1000 / 100 | Yes — meaningless without one |
--stylize (--s) | How much of Midjourney's own aesthetic flourish it adds | 0–1000 / 100 | No — works on any prompt |
--stylize is the model's built-in "make it prettier / more artistic" slider. Low values follow your prompt literally; high values give Midjourney room to embellish color, composition and detail in its house style — whether or not you supplied a reference. The pool images run it at the maximum, --stylize 1000, which is why they read so finished and painterly even before the reference lands.
The clean mental model: --stylize sets how much Midjourney editorializes; --sw sets how much your reference gets to veto that editorializing. A high --stylize with a low --sw gives you Midjourney's flourish loosely flavored by your reference. A high --sw with a high --stylize — the configuration in these examples — gives you a heavily-stylized image firmly pinned to the reference's identity. They stack; they do not substitute.
A repeatable routine beats guessing:
| Step | What to do |
|---|---|
| 1. Anchor | Lock one neutral prompt and your reference set. Change nothing else. |
| 2. Sweep | Render at --sw 100, 300, 500, 800. Compare the four grids side by side. |
| 3. Read the trade | Find the point where the reference reads clearly but your subject is still intact. That is your weight for this reference. |
| 4. Separate the dials | If it looks over-cooked rather than over-referenced, lower --stylize, not --sw. |
| 5. Lock the series | Keep the reference set + weight fixed and vary only the prompt to build a coherent body of work. |
Distinctive references find their level early — the cyber set sings at 300. Softer references climb higher before they assert. Either way, the number is a conversation with one specific reference, not a global setting you carry between projects.
Working backwards from a look you admire? You cannot read a numeric --sref or a precise --sw out of a finished image — that information is not recoverable from pixels. What you can do is decode the image's style DNA — palette, lighting, finish, the named aesthetic layers — and rebuild it with your own references at a weight you choose. That is style-matching, not extraction, and it is the honest version of "how was this made."
· Midjourney style codes explained — the complete --sref reference, syntax to FAQ.
· How to use sref codes in Midjourney V8 — the hands-on walkthrough.
· sref vs cref vs iref — pick the right reference parameter for the job.
· Best Midjourney sref codes (2026) — sourcing and testing codes that fit your work.
· Browse the decoded style gallery — thousands of real V8.1 looks to style-match against.
· Explore the aikizi lab — tools for reading and rebuilding visual style.