How to get way better Midjourney results with these steps

I’ve got a Midjourney subscription for quite a while now, but with each new Midjourney update you kinda have to relearn how to write prompts and get amazing outcome.

So what do you do? You YouTube (is that Verb?)

The problem with Youtube (and basically the entire internet for that matter), is clickbait and shitty content.

(I hope short blogposts like this one will make a comeback someday).

So let me give it a shot to use as few words as possible and help you as much as possible, with creating more awesome and better results with Midjourney.

Tip 1/1: Slowly build up your prompt

This is basically the best advice I’ve come across on the internet (yes, I know, I ranted about the internet a few paragraphs back, but in the dirt there are some good tips lying around).

Keep your initial prompt short.

Keep your initial prompt short. Instead of entering the entire 300 character large massive prompt, start super small.

It’s kinda like a conversation, where in each step you check if Midjourney understood your prompt. If so, you continue adding details or changing elements, if not, you alter the prompt.

Why? Because if you enter this massive prompt all at once, and things are off and wrong compared to what you had in mind, there is no way in telling which part Midjourney did not understand.

So, start small.

An example

Let’s take a look at an example. In, let’s say, we need images for a website that sells arm bracelet’s.

I found a reference picture online which contains the exact look I’m going for. So we try and reproduce it.

Prompt 1

A closeup of a female forearm.

This is not what I was going for. The bottom left is alright right, but there a few upper arms in the picture. Clearly the prompt is not clear enough.

Some pictures contain illustrative elements so I will “photograph” in the next prompt. And made it clear that elbow would not be included.

Prompt 2

A close-up photograph of a female forearms and hand, without the elbow.”

This is already better. But the arms look too manly or detailed. I’ve added “smooth skin” and the age into the prompt.

Prompt 3

A close-up photograph of a 20 year old female forearms and hand, without the elbow. Smooth skin.

This is not bad at all. Let’s try and add the clothing next.

Prompt 4

A close-up photograph of a 20 year old female forearms and hand, without the elbow. Smooth skin. She is holding the arm in-front of her body. The woman is wearing a white jacked.

Well this is not at all what we’re going for.

I’ve left out a few prompt attempt, but after about 5 tries, I couldn’t get the position of the arm right.

I decided to add the bracelet first, and work about the clothing later.

Prompt 5

A close-up photograph of a 20 year old female exposed forearm and hand, without the elbow. Smooth skin. She is wearing a 4 small bracelet of turquoise beads with small golden details.

Nice pictures, but not what we’re going for. I’ve added the fact that the people in the picture had to be white.

I knew Midjourney would get this addition, so I decided to throw in some details about the white blazer as well.

Bonus tip: Change the Stylize setting if Midjourney seems to dismiss your instructions.

I reduced the styling from medium to low. Which you can recognize by the –s 50 in the prompt. The lower the styling, the better it listens to your prompt. The higher, the more creative it seems get.

(If decided not to include every step in this blogpost, that would make this blogpost a bit too long. But at some point I included “We only see the forearm. We don’t see her face” in the prompt. Because the results kept showing me part of the face.)

Prompt 6

A super close-up photograph of a 20 year old white female exposed forearm and hand. We only see the forearm. We don’t see her face. On the background we see the off-white blazer she wears. Smooth skin. She is wearing a four small bracelet of turquoise beads with golden details. –v 6.0 –style raw –s 50

Still the bloody face…

The part about “we don’t see her face” is clearly not working. I read some stuff online about negative prompts: telling him what NOT to see. But in this case, this didn’t work.

I removed this “we dont see the face part” and threw in a sentence stating we want to see a blazer/jacket on the background.

If you see a blazer on the background, you can’t see a face, right? 🙂

Prompt 7

A super close-up photograph of a 20 year old white female exposed forearm and hand. Smooth skin. On the background we see the off-white blazer she wears. She is wearing a four small bracelet of turquoise beads with golden details. Shallow depth of field. –v 6.0 –style raw –s 50

Now we’re getting there! No more faces, and a nice close-up of the arm and the bracelet!

This is getting pretty close to what I want, so from this point on I decided to run a few variations on this prompt.

When you run a prompt a second time, you will get different results.

(This has to do with the seed. You can control this if you like by adding –seed 1234 or whatever number behind your prompt.)

After I ran the prompt again for a few times I got this result, which looks pretty great!

Summary

Building up your prompt step by steps works pretty good so far.

This way you know which sentence or which words Midjourney is not getting, misinterpretting or not using at all.

The first step was already not detailed enough. I was getting pictures of a upper arms & forearm, and either too detailed or illustrative. So I added ‘Photograph’ and the age to get the look I was going for.

“We don’t see the face” part didn’t do anything. So I left that out and was able to change it to something clever like “we see her blazer on the background”.

Good luck prompting!