Two useful features you need to be aware of when starting to use Midjourney are image prompting and the seed parameter.
These can provide some more control and constrain the generator from starting afresh with a new image with each prompt, lending continuity between prompts that can be very useful when homing in on an idea.
As an example, let’s consider the following prompt:
black and white enamelled design of a heron standing in water. beautiful. art nouveau. –ar 1:2 –v 5.1
Of note in that prompt are the “parameters”, which are the terms appearing after the “–” double dashes at the end. Seed is one of the available parameters that can be used, but I haven’t used it here. The two I’ve used here are aspect ratio and version.
“–” appended to the prompt means “use a parameter”
“ar” means “aspect ratio” (the ratio of width to height of the generated image). “v” means “version” as in which version of the Midjourney generator to use. Here, I’ve used version 5.1, which is currently the latest.
After the space, “1:2” means 1 x-axis unit to each y-axis unit, i.e. the image will be twice as tall as it is wide.
However, these results aren’t quite what I was looking for, so let’s try something else.
black and white enamelled tablet. a white heron on a black background. rectangular. flat stylised ceramic design. –ar 1:2 –v 5.1
I like this more.
What I’m going to do now is upscale this option. This gives me the image by itself. Clicking the image, and then “Open in browser” underneath will open a new browser tab with the image in it. I then copy the URL, which is pointing to a location on Midjourney’s servers as I haven’t moved it anywhere else.
This URL allows us to do an image prompt.
Starting as normal by typing “/imagine” I then simply paste in the URL. I can add further terms too in order to refine the prompt based on the image.
This allows us to do some more interesting things. Let’s use that image URL and then some more text:
plaster of paris cast relief. –seed 2745901785 –ar 1:2 –v 5.1
A couple of things to note here.
1. I haven’t told it anything about fairies or women or whatever that hideous ghoulish thing in the bottom-right is. It’s decided those things are close to wherever we are in the latent space.
2. We’re now using the seed parameter. I like to use this in combination with image prompting to give close control. The image prompt will guide the outcome in terms of composition. The seed parameter will define the noise field that is used to generate the result. All Midjourney images are created by generating a field of noise (imagine the black and white static on an old detuned TV) and then sharpening it progressively to pull out details that look like details that have appeared in its learning set. Each generation has a seed number applied to it, and if you find and copy the seed, you can use the noise field that gave that particular result again later.
How do we do this? Being a bot working in Discord, Midjourney makes use of Discord’s UI. To get the seed for a job, you need to ask the bot and you do this by reacting to the job in question.
Selecting this “Add Reaction” button, and then adding the “envelope” reaction will enable you to ask the Midjourney bot to return the seed for a given job by reacting to it with the envelope. You can paste that seed number into future prompts using the format given above.
Using this combination of techniques, I advanced through the following steps, looking for what I wanted:
