Short answer: in Midjourney V8.1, --sref copies a style, --cref copied a character (and is now legacy), and "iref" is community slang for an image prompt — pasting a whole picture as visual inspiration. Two of those three are first-class in V8.1, and one of them won't even fire on the current model. If you have been pasting old --cref prompts into V8.1 and wondering why nothing changes, this is why.

Let me map all three to what they really are in the current model, in plain terms, with real prompts and real images at the bottom.

Reference	What it copies	Weight param · range (default)	V8.1 status
--sref	Style only — palette, light, texture, medium, rendering. Not the subject.	--sw 0–1000 (100)	Native to V8.1. Holds across a series.
--cref	A character's identity — face and features carried into a new scene.	--cw 0–100 (100)	Legacy flag (Midjourney / Niji 6). Not part of V8.1.
"iref" (image prompt)	The whole image as inspiration — composition, content, and style blended.	--iw (image weight), default 1	Works in V8.1.

Two of those three carry a real flag; "iref" does not. It is the most common point of confusion, so let's untangle the naming before going deeper.

There is no --iref in Midjourney. "Iref" is shorthand people use for an image prompt — you paste an image URL at the start of your prompt (no flag), and the model treats that picture as visual source material. Unlike a style reference, an image prompt pulls from everything: the subject, the composition, the colors, the mood, all at once. You steer how loud it is with --iw (image weight).

In the V5/V6 era, --iw ran roughly 0–3 with a default of 1. V8.1 brought image prompts and image weighting back; the exact ceiling for --iw in V8.1 isn't something Midjourney publishes precisely, so treat 1 as the neutral baseline and adjust from there rather than assuming a hard cap. The mental model that matters: an image prompt is a blend of the whole picture, while --sref is surgically style-only.

A style reference tells V8.1 "render my prompt, but in this aesthetic." It transfers color temperature, lighting behavior, texture, contrast, and medium — the things that aren't what the image shows but how it's painted. Your subject stays whatever your words describe.

You can feed --sref three kinds of value: an image URL, a numeric style code, or a random seed via --sref random. Style weight --sw controls intensity from 0 to 1000 (default 100). At low --sw the style is a whisper; near 1000 it can overwhelm your prompt entirely. The genuinely good news in V8.1: srefs and moodboards are markedly stable — a single reference holds its look across a whole series instead of drifting shot to shot, which is the failure mode older models were notorious for.

Character Reference (--cref) was built for one job: take a person you've generated and put the same face into a new pose, outfit, or scene. Its weight was --cw, scaled 0–100 — at --cw 0 Midjourney copies the face only, at --cw 100 it copies face, hair, and clothing.

Here's the catch for 2026: --cref is a Midjourney / Niji 6-era parameter. It is not part of V8.1's parameter set. Append --cref to a V8.1 prompt and the model simply ignores it — which is exactly the "nothing happened" people keep reporting.

Its successor is Omni Reference (--oref), weighted with --ow — broader than character reference, able to carry a person, an object, a vehicle, or a creature into a new generation, with --ow running 0–1000 (default 100) for loose echo up through hard identity lock. The honest status for V8.1: Omni Reference has not been brought to the model yet — Midjourney has said it is still in training for V8. So as of mid-2026, V8.1 has no native character-lock parameter. The model that nailed style stability hasn't shipped its identity-lock counterpart, and there's no flag you can append to force one.

This is the part that's hard to appreciate until you see it. Below are real V8.1 generations from the AIKIZI gallery. The two Buddhist pieces were produced with the same two style references — https://s.mj.run/SxrlUBCPpb4 and https://s.mj.run/lMrYN1a0H2M — and yet each keeps its own subject. That's --sref doing exactly its job: shared aesthetic DNA, independent content.

Swap the concept words, keep the --sref pair, and the second piece — a Nihonga-flavored composition under a different moon — still reads as a sibling. That cross-image consistency is the entire reason style references exist, and it's where V8.1's stability fix actually pays off.

The same trick works on the other end of the aesthetic spectrum. These two share the references https://s.mj.run/I8ZEy5X1Xco and https://s.mj.run/xVNj3CzP9Lo — one rendered as cybernetic tribal realism, one as a neo-noir engraving — bound by a single style fingerprint at a lighter --sw 300.

One thing worth stating flatly, because it's a recurring myth: a numeric or seed sref (for example --sref 2847193 or --sref random) is a coordinate into Midjourney's latent style space. It is not derived from any picture. That means you cannot point at an arbitrary image you found online and "get its sref number." There's no function that runs an image backwards into a six-digit code — the code came first, the image second.

What you can do is the more useful thing: decode the style of an image into named, human-readable layers, then rebuild it. That's the AIKIZI approach — instead of pretending to recover a black-box number, it reads the image's actual visual DNA (the tokens you see above are real decode output) and gives you a style-match you can paste into any model. Style match and style decode, never "extract the sref of your photo," because that operation doesn't exist.

If you want the deeper mechanics of any single flag, the rest of the cluster goes one layer down.

· Midjourney --sref Style Reference Codes Explained — the full pillar guide.
· How to Use Sref Codes in Midjourney V8 — step-by-step.
· Midjourney --sw Style Weight, Tuned — dialing intensity 0–1000.
· Best Midjourney Sref Codes of 2026 — curated looks to copy.
· The AIKIZI Gallery — thousands of decoded V8.1 images.

Decode A Style →