Using Midjourney to Generate Sample Content for QA and Proof-of-Concepts (POC)

Where Are the Good Images At?

Whenever I’m getting to the point where I want to see a component in action or prepping a proof-of-concept for a potential client, one of the things that I almost always get stuck with is finding images. Good images. Images that actually make a component pop. And let’s face it; you can only use placekitten so many times. Ignoring the fact that it doesn’t work for professional situations.

Ideally, you’d want to find a free image stock with relevant images. But nowadays you’ve got services like Midjourney. With a couple of relevant prompts, you can generate images that might reflect the topic you’re after, in the aspect ratio you’re after and potentially even in a colour palette that your intended audience would love. Gone are the days of searching royalty stock photos and paying hundreds.

 

Using Midjourney

Once you’ve signed up, whether for a trial or a subscription, you’ll use Discord to generate your images. As of this writing, the trial service is unavailable, but this is one subscription you easily get your money’s worth. You can do this either in a private message DM with the Midjourney Bot itself or in one of the many appropriate channels.

Here’s what a typical post looks like. Here I’ve entered the command /imagine followed by a series of prompts.

In this case, the entire prompt looked like this:

/imagine eggs in a pan yellow orange red vector illustration flat light --ar 4:3 --v 5

From there, we can choose to upscale one of the provided images, create variations of them or produce a new random set. Prompts such as vector illustration help give you a flat image that might be suitable for a PowerPoint. The beauty here is that you can easily change colours in Photoshop if so desired, but it allows you to get closer to the colours you were after. When I’m happy with my selection, in this case, image #2, I can upscale it and download it.

QA Use Cases

What’s often the case when it comes to QA is not necessarily that an image has the right colour or its topic is fit for purpose. The aspect ratio, however, is often important and sometimes overlooked. Getting images, without having Photoshop, in a proper aspect ratio can be a challenge.

To achieve this, we use the --ar parameter typically followed by the ratio. For example 1:1 or 16:9 or 4:3 . You can learn more about aspect ratios in Midjourney’s documentation.

The beauty about using colourful images, like these or others, is you’re not looking at a bland grey image all over the page. It’s the same old issue where if you’re looking at something mundane and boring and long enough, you’ll miss glaring issues. Keeping images fresh and interesting, your eyes are more keenly aware of what is not working well.

It’s also extremely wise to test a variety of aspect ratios for components that perhaps utilize background images or even an image with an object-fit: cover; style. You really never know what authors might upload into Sitecore, so it’s a good idea to have a variety of sizes and shapes to test with.

POC Use Cases

Proof-of-concepts, while not nearly as common as a good QA session, are a lot more powerful when they contain unique imagery to go along with them. Obviously, a client who gets you to build a site or a proof-of-concept very well may offer imagery. And if they do, I highly recommend you use it. But if they don’t, this is where you can be a bit more creative.

Before utilizing Midjourney though, I recommend taking into account a few things:

• Branding
• Any relevant marketing materials
• Color palette
• Existing photos are more abstract or contain people

These things can only improve your chances if you’ve recognized and mimicked their style. Obviously, if they’re planning on changing their style, this should also be considered.

Full Parameter List

As of version 5 of Midjourney, the complete list of parameters that work with Midjourney commands is as follows:

• --aspect, or --ar - Alter the aspect ratio of images. The default is 1:1, but you can do a wide range.
• --stylize <number>, or --s <number> parameter influences how strongly Midjourney's default aesthetic style is applied to Jobs.
• -quality <.25, .5, 1, or 2>, or --q <.25, .5, 1, or 2> - The level of quality you want it to process. The higher the number, the more time it costs you.
• --chaos <number 0–100> - How random the images are. i.e. How chaotic.
• --no <value> - Useful when exempting items from an image.
• --seed <integer between 0–4294967295> - This is used by Midjourney, but is not a required parameter. It is useful, however when you want to create images varied of the same visual style.
• --stop <integer between 10–100> - Not really sure why you’d use this to be honest. It will stop the image creation at a certain percentage of completion.
• --style <4a, 4b, or 4c> - Use to enable specific Midjourney Model Versions
• --style <expressive, or cute> Switch between versions of the Niji Model Version 5
• --niji - To create a more anime style.

New Parameters

• --repeat <1–40>, or --r <1–40> - Utilize the repeat parameter to basically run the same prompt multiple times.

Using Permutations

As of very recently, Midjourney is now allowing permutations which is a fancy way to say being able to write a single /imagine prompt but getting a variety of results based upon the combinations. For example, if we were to write the following prompt:

/imagine {man, woman} looking at {sunrise, sunset}

This will then generate the following prompts:

• /imagine man looking at sunrise
• /imagine woman looking at sunset
• /imagine man looking at sunrise,
• /imagine woman looking at sunset

This only works when you’re using the /fast mode and not /relax mode. You can read more about Midjourney Permutations here.

Only Getting Started

I’m certainly excited to see where we can take advantage of these types of services to improve the quality of content we produce and how it will benefit clients through using Sitecore. It should be interesting.

---