Introduction to MidJourney
MidJourney is the latest among a slew of powerful AI-based image generation platforms. It provides unprecedented flexibility and power in creating images. Here I will walk you through the very basics of image generation, using a fairly generic prompt: I want to see pictures of knights who are wielding swords and carrying torches!
This tutorial assumes you have already accessed the MidJourney platform through Discord (a free process that takes less than 5 minutes). If you have not gotten access to MidJourney yet, you can follow my other tutorial here. Keep in mind that free access to MidJourney allows you to have the full functionality, but you only get a limited number of image generations (the consensus seems to be ~25).
So let’s begin: a knight with a sword and a torch
We already have something that’s reminiscent of our prompt! MidJourney produces 4 images every time you query it, each generated from a random seed. Let’s say these were in the right direction of what we wanted, but none really did it for us. In that case, we can just press the redo button, and MidJourney will generate 4 more images.
Let’s say I really like the direction of the top right. Let’s see some variations on that idea!
The images are named 1 (top left), 2 (top right), 3 (bottom left) , and 4 (bottom right). MidJourney has commands for producing (V)ariations on an image and for (U)pscaling an image to a higher quality. Let’s click V2 to see variations on the top right.
Each is distinct but with similar colors and elements. We can continue trying to redo to get more variations, or if like one enough, we can upscale it to full resolution. Let’s click “U4” to upscale the bottom right image.
These are the basic of getting an image, navigating variations, and upscaling an output. But let’s keep going to push the boundaries.
As of November 2022, MidJourney released their 4th generation engine, known as “Version 4.” To apply version 4 to image generation, you need to add “--v 4” to the end of your prompt (make sure to use two dashes, a lowercase v, a space, then a 4; improper syntax will result in an error!). Let’s see what our original prompt looks like when using this version.
Already this is COOL! If you look closely, you’ll notice some “AI mistakes,” such as misplaced appendages or armor that overlaps, but in general this is a great start. Let’s look at ways we can refine our prompt to get an image we LOVE.
First, we’ll focus on the content of the image, then later on we will focus on the style. To start, let’s have our knight looking at us.
Good. Now let’s play with where our night is standing. Perhaps we want him in a dungeon?
Or what about in a field in front of a castle?
How about on the moon?
Notice the difference in armor that is brought in depending on the prompt — MidJourney uses other existing images to construct the image after receiving a prompt. Therefore, the prompt of the knight on the moon picked up some white, almost “astronaut-esque” touches, while the knight in the dungeon was very dark with bright flames from his torch. If we want specific outcomes, we need to be more specific:
To finish up today’s example, let’s add a little extra pizzaz, including some stylistic additions which I’ll cover in a later blog.