My journey in Midjourney

I've been messing around with Midjourney for a couple of months now. I use the phrase "messing around" because that's exactly what the process has felt like.

If you've not come across it yet, Midjourney is an AI-powered image generation tool that allows users to create images by providing text and/or image prompts. You use it via installing the Midjourney bot to your Discord, and adding the bot to your text channel.

My initial testing came out of a curiosity to determine whether I could use Midjourney as a replacement for traditional stock photography in website content or social media posts.

Just for fun, and the purposes of this article, I went back to one of the first upscaled images I created.

/imagine Cats wearing bow ties with a starry night sky

This one I kept for nostalgia. And because cats.

Getting to grips with the descriptive prompts for Midjourney has been an adventure in itself. I've had some rather horrific failures:

Do we remember the "woman laughing at salad" thing from a while back?

The general structure of a prompt is

{image prompt} {description} {parameters}

Where

• image prompt would be the URL of an image that you want to use as a base, although this isn't mandatory.
• description being your description of the image that you'd like it to generate.
• parameters being the settings that you can add to affect how Midjourney creates the image. I won't dive into these in much detail here, but you can see what's available in the Midjourney documentation.

My one piece of advice when it comes to creating the prompts is to experiment. A lot. Try all sorts of different combinations, and don't be shy in getting Midjourney to regenerate the image set that it gives you if it's coming up a bit twisted. Of course, the more random you get with your descriptions, the crazier the results are going to end up.

Here are my thoughts on what you can consider when writing your own prompts:

1. Be clear in your language. The more straightforward and direct you are with your description, the better that the software is going to understand you're requirements. Don't talk to it in the style of Jane Austen or Russell Brand, essentially.
2. Be specific. The more detail you add to your prompt, the more it's going to match your expectations. Mention colours, objects, the environment. You've got to give it as much guidance as you can.
3. Talk about the mood. If you want your image to be of a certain style, then add that into your prompt. For example (and this is one of my favourites) a futuristic cityscape with neon lights creating a cyberpunk atmosphere.
4. Composition. Like any good photographer, you're going to have to think about the composition of your image. You can add things like using the rule of thirds or wide angle shot to your prompt to help guide the software in creating your perfect image.
5. Set your focal point. Again, think like a photographer. What's the focus of your image? Try to define your primary subject in your prompt.

I've actually found, after some considerable messing about, that Midjourney is a great tool for creating images for social posts (particularly Twitter and Instagram). You can adjust the aspect ratio of the images as well, which is a definite plus when posting to different platforms.

By default, Midjourney gives you a 1:1 aspect ratio. You can change this by adding the --aspect 3:2 parameter at the end of your prompt (or whatever ratio that you want to use, I've just added 3:2 here as an example).

Now, let's fast-forward rather rapidly to yesterday evening. I feel like I've satisfied the question "can Midjourney be used to create content for web". So I decided to change it up a bit. I took some stills from one of my favourite films and used them as prompts to create variations in Midjourney v5. The results are astounding.

Here's my first image prompt:

Yeah, that's Rick Deckard setting up the Voight-Kampff before he questions Rachel in Tyrell's tower (guessed what the film is yet?).

Using /imagine with the image as a prompt, a brief description, and the parameters for setting the version and image weight (at the time of writing, Midjourney uses version 4 by default, so we have to explicitly tell it to use version 5).

/imagine [imageurl] voight-kampf, blade runner --version 5 --iw 2

The result was this beauty:

And of course, what kind of fan would I be if I didn't try to recreate this iconic scene...

I'll continue experimenting with what Midjourney v5 can do. So far, I am more than impressed. Maybe another article on this in a month or two where I share my new findings.