parameters — --raw

What --raw turns off

Midjourney styles your image whether you asked it to or not. --raw is the flag that stops it, which is why photographic work reaches for it and illustration rarely does.

Every Midjourney image passes through a styling pass you did not ask for. It is the reason a three-word prompt comes back looking composed and graded rather than like a snapshot — and the reason a plain request for a photograph often returns something closer to an advertisement.

--raw turns that pass off.

a photograph of a kitchen window in morning light --raw
  • V5.1+model versions supported
  • offthe default — styling is on unless you say otherwise
  • no valueit is a switch, not a number

What "auto-pilot" means here

Midjourney's own framing: in standard mode it automatically adds its own creative touch, which is what lets simple prompts turn into a wide range of artistic results. Switching to Raw turns off that auto-pilot.

The effect depends entirely on how much your prompt is carrying:

With a…Raw gives you
simple promptmore realistic, photo-like images — the default styling was what made it look "designed"
stylistically detailed promptcontrol — your described look lands instead of being blended with the house one

Who reaches for it

The pattern in real prompts is consistent, and it follows from the above: work that is trying to look like a photograph benefits from removing the styling pass, while work that is already illustration or concept art has no quarrel with it. A prompt aiming at a commercial product shot is the clearest case — nothing about the finish can be left to the model:

An editorial, cinematic, and dramatically lit product scene set in a windswept terracotta desert canyon. skin care product in on the middle of the scene, Place the product prominently on a sun-baked sandstone rock slab in the foreground. near this skincare product, there are grass and flowers looks like very watery In the background, create a glowing, atmospheric haze with gusts of wind blowing fine sand, swirling dramatically through the canyon. Add unusual, gravity-defying balanced rocks, sculptural and monolithic, highlighted by a strong directional rim light that enhances texture and adds a sense of surreal editorial drama. Integrate additional brutalist elements: fine cubes and rectangular blocks of raw white limestone, arranged as an abstract interpretation of a brushed-steel faucet and minimalist sink, clean geometric forms contrasting with the organic terrain. Lighting should be rich, warm, and cinematic, with deep shadows and a glowing amber backlight diffusing through the dust. Shoot the scene as if captured on a medium format camera, using a medium-long lens, in a 9:16 vertical aspect ratio. Hyper-realistic, high-end luxury texture quality. Mood: dramatic, sculptural, atmospheric, editorial luxury. --ar 9:16 --raw

Conversely, a prompt whose whole point is an invented aesthetic is usually happy to let Midjourney's taste contribute, so --raw simply never comes up.

Raw versus low stylize

These get used interchangeably and should not be. They act at different points:

What it does
--rawremoves the default styling pass — a switch
--stylizesets how much latitude is taken with your prompt — a dial, 0–1000

They combine, and the combination is a recognisable idiom: --raw plus a low stylize, usually with a style reference supplying the look that the model is no longer allowed to invent. Both parameters push in the same direction; neither makes the other redundant.

Version support

Raw is compatible with Midjourney versions 5.1 and later, which in practice means every model anyone is still using. It is worth noting what Midjourney says about it on the current generation: for V8.1, the release notes recommend turning Raw on when you want even more prompt adherence from a model already built for it — so this is not a legacy flag being kept alive out of politeness.

Every prompt in the collection is published exactly as it was written, so the presence or absence of --raw is visible per image rather than something you have to infer from the result.