parameters — --sw, --sv and --iw
How strongly a reference applies: --sw, --sv and --iw
Adding a reference is one decision; how loud it is is another. --sw is the volume knob, --sv is which algorithm interprets it, and --iw is the same idea for image prompts.
A reference is not a switch. Once you have attached one — a style reference, an image prompt — three separate parameters decide how much of it survives into the final image. They are easy to confuse because two of them start with the same two letters and do entirely different things.
| Parameter | Question it answers | Range | Default |
|---|---|---|---|
--sw | how strongly does the style apply? | 0–1000 | 100 |
--sv | which style-reference algorithm interprets it? | V7: 1–6 · V6: 1–4 | V7: 6 · V6: 4 |
--iw | how strongly does an image prompt apply? | 0–3 (Niji 7: 0–2) | 1 |
--sw — style weight
Style weight is a volume knob on the reference. Any value from 0 to 1000, default 100. Below 100 the reference becomes a tint; above it, the reference starts overriding your description.
a quiet hotel courtyard --sref 3808932062 --sw 400
- 100the default — not the middle of the range
- 1000maximum, ten times the default
- 0reference attached, influence off
Worth registering that 100 sits near the bottom of a 0–1000 scale, so the default is a fairly light touch and there is far more room above it than below. If a reference feels like it is barely landing, the honest first move is a bigger --sw, not more adjectives.
--sv — style reference version
--sv is not a quality setting and not a strength setting. It selects which implementation of the style-reference feature runs, because Midjourney has rewritten it several times and kept the older ones addressable.
| Value | What it is |
|---|---|
| --sv 6 | the default on V7 |
| --sv 4 | the pre-June-2025 V7 model — and the V6 default |
| --sv 3 | V6: an updated take on the "vibey" original |
| --sv 2 | V6: the second iteration |
| --sv 1 | V6: the original algorithm, loosest interpretation |
The practical use is recovery. Style codes from before the V6→V7 rewrite may no longer produce the look they did, and --sv 4 is how you get the old behaviour back without leaving the newer model.
One constraint that bites: --sref random and style codes are only compatible with --sv 4 and --sv 6. Pick 1, 2 or 3 and you are in image-reference-only territory.
Using both at once
--sw and --sv are independent — one says how loud, the other says which algorithm — so a prompt can carry both. Something like this is a normal, non-exotic tail:
wellness --ar 4:5 --v 7 --raw --sw 1000 --sv 6 --sref 1794055294 --s 400
--iw — image weight
The same idea for a different reference type. --iw sets how much an image prompt influences the result relative to your text. Default 1, range 0–3 on V7 and V8.1, and 0–2 on Niji 7.
Note the scale is completely different from --sw: here 3 is the ceiling, so --iw 3 is a strong instruction while --sw 3 is almost nothing. Two parameters, two ranges, no shared intuition.
The incompatibility to remember
Prompts in the collection are published with every parameter parsed into its own field, so you can see which of these actually appear together in practice rather than in theory.