What the --exp parameter actually does in Midjourney V8.2

A parameter with no documentation, tested across photographic and illustration prompts from 5 to 100

Aug 07, 2026

∙ Paid

Some Midjourney parameters get an announcement, a docs page, maybe a tutorial video. Then there’s --exp, which got mentioned in Office Hours and pretty much left at that. Go looking in the official documentation today and you’ll find nothing except a pointer back to the V7 announcement.

The official documentation didn’t clearly say what exp stands for. Experimental is the reasonable guess, and honestly it fits, because using it feels exactly like that.

This is one of the most elusive parameters Midjourney has. You use it, you like the look, you keep using it. The mechanism stays a black box. Which means the only real way to learn what it does is to try it and watch what changes.

--exp got revamped for V8.2, and I covered the announcement in Creativity AI #86. This article is my actual testing of it.

What we knew from V7

I wrote a quick overview of --exp back in V7, and most of the basics still apply:

Defaults to 0, accepts whole numbers from 0 to 100
Decimals get rejected, so --exp 0.5 won’t run
Midjourney suggested testing 5, 10, 25, 50, and 100
The visual jump from 5 to 50 is big, from 50 to 100 much subtler
It works alongside --stylize, --p, --seed, image prompts, --sref, and --oref
The description at the time: more detailed, dynamic, creative, tone-mapped
Push the value high and prompt adherence falls apart

That last one is the trade-off that has followed --exp since the beginning. It’s said to favour creativity and aesthetic quality over strictly following what you asked for. Keep that in mind, because it shows up again below.

My V7 advice was to stay at 25 or lower. Let’s see if that still holds a full model generation later.

Test one: a photographic scene

I started with an explorer and his cat discovering a hidden treasure chamber in a cave. Photographic, V8.2.

No Exp

A weathered explorer kneels beside his cat as they uncover a hidden treasure chamber, torchlight revealing mountains of gold, astonishment reflected in their eyes, photorealistic adventure film still, rich amber tones --ar 16:9 --profile pakcpbq --v 8.2

At --exp 0 the image already looks good. No surprise there, since V8.2 is a quiet win for photographic images even before you touch anything.

Exp 5

At 5, slightly more detail in the face and the surroundings. The expression is a little more intense than the version without it.

Exp 10

At 10, noticeably more detail, especially the surroundings. Look at the torch. The cat has shifted position too. In this particular case it also made the image slightly less realistic, which I didn't like.

Exp 25

At 25, not much difference from 10. More detail again, but nothing dramatic.

Exp 50

At 50, the prompt starts drifting. The man is now holding some kind of lighting device, an electric or elaborate torch that I never asked for.

Exp 100

And at 100, the expression goes further, and one cat becomes two. Two cats. From a prompt with one.

So the old warning still stands. Max it out --exp and you get low prompt adherence.

Test two: illustration

Different image type, same ladder. A cute corgi celebrating a birthday in a Parisian apartment.