MidJourney: Does Prompt Order Matter?

This page is just an exploration of prompt order, to determine whether order matters to MidJourney's algorithm.

The current conclusion is... We don't know! :).

Setting Subject Lighting

Hypothesis

Different orders of setting / subject / lighting will show a consistent difference in the final image if order matters.

Setup

We'll start with a 2 standard prompts as recommended in the prompt-craft channel:

• Setting :: Subject :: lightning, any other modifiers
• Setting
  • setting is a jungle
  • setting is a desert
• Subject
  • Oswald McHoolahan is standing
  • Patricia Hughes is sitting
• Lighting
  • Sunbeam coming through the leaves
  • Dusk lighting

Let's see how it does with various orders!

We'll hold stylize, seed, and sameseed constant.

Commands used

/imagine prompt:setting is a jungle:: Oswald McHoolahan is standing:: sunbeam coming through the leaves:: --stylize 625 --seed 1 --sameseed 5

/imagine prompt:setting is a jungle:: sunbeam coming through the leaves::  Oswald McHoolahan is standing:: --stylize 625 --seed 1 --sameseed 5

/imagine prompt:Oswald McHoolahan is standing:: setting is a jungle :: sunbeam coming through the leaves:: --stylize 625 --seed 1 --sameseed 5

/imagine prompt:Oswald McHoolahan is standing:: sunbeam coming through the leaves:: setting is a jungle :: --stylize 625 --seed 1 --sameseed 5

/imagine prompt:sunbeam coming through the leaves:: setting is a jungle:: Oswald McHoolahan is standing:: --stylize 625 --seed 1 --sameseed 5

/imagine prompt:sunbeam coming through the leaves:: Oswald McHoolahan is standing:: setting is a jungle:: --stylize 625 --seed 1 --sameseed 5

/imagine prompt:setting is a desert:: Patricia Hughes is sitting:: Dusk lighting::  --stylize 625 --seed 1 --sameseed 5

/imagine prompt:setting is a desert:: Dusk lighting:: Patricia Hughes is sitting::  --stylize 625 --seed 1 --sameseed 5

/imagine prompt:Patricia Hughes is sitting:: setting is a desert:: Dusk lighting::  --stylize 625 --seed 1 --sameseed 5

/imagine prompt:Patricia Hughes is sitting:: Dusk lighting:: setting is a desert::  --stylize 625 --seed 1 --sameseed 5

/imagine prompt:Dusk lighting:: setting is a desert:: Patricia Hughes is sitting::  --stylize 625 --seed 1 --sameseed 5

/imagine prompt:Dusk lighting:: Patricia Hughes is sitting:: setting is a desert::  --stylize 625 --seed 1 --sameseed 5

Results

In the first set, it does seem that prompt order mattered - while the images are very similar, they seem to emphasize different pieces of the image. The results appear to focus on the items in order - the results starting with "Setting" focus on the setting, the results starting with "Subject" seem to have a larger subject, and the results starting with "lighting" have a greater focus on the beam of light.

In the second, it appears less prominent - though might still be there. There at least does seem to be a difference in the images, though it's not as obvious (to my eye at least) as the first set associated to the first prompt.

Overall I'd say this is still inconclusive - while it may show (at least in these two cases) a difference based on input prompt order, it doesn't quite show how it impacts it, and might be due to other variables at play.

Setting subject lighting	Setting lighting subject	Setting subject lighting	Setting lighting subject
Subject setting lighting	Subject lighting setting	Subject setting lighting	Subject lighting setting
Lighting setting subject	Lighting subject setting	Lighting setting subject	Lighting subject setting

Cat Dog

Hypothesis

Well what about a more simple test... cat and dog? Pretty opposite. This should be a clearer indication.

If the theory holds true, putting cat first will give a more catlike dog, and vice versa.

Setup

Pretty straightforward - we'll try both cat and dog, and hold stylize, seed, and sameseed constant.

Commands used

/imagine prompt:cat :: dog:: --seed 10 --stylize 625 --sameseed 15 --uplight

/imagine prompt:dog :: cat:: --seed 10 --stylize 625 --sameseed 15 --uplight

Results

As you can see below... It's not a huge impact. Perhaps this is more a selection issue - maybe the "seed noise" inherently is more cat or dog. More testing necessary.

cat :: dog :: seed 10, sameseed 15

dog :: cat :: seed 10, sameseed 15