A Revolution in Visual Communications: Midjourney for Business
As a CTO, a key part of my job is to communicate complex concepts in ways that are understandable to non-technical people. Effectively incorporating an image that can perfectly convey a complex metaphor is often what makes a presentation succeed. But finding relevant, captivating images to illustrate a point is a time-consuming challenge. Midjourney – the best of the text-to-image generative AI services available – has fundamentally changed not only how I find images, but how I articulate and express my ideas.
One way to understand this transformation is to view Midjourney as a source of visual building blocks. Although my Photoshop skills allow me to refine and sculpt images, Photoshop doesn't offer the raw materials I need. As I'm not an artist, I frequently have to settle for suboptimal images found through Google image searches. However, with Midjourney, I can create these visual building blocks – for the first time ever. This is incredibly empowering and enables me to communicate in ways that I simply couldn't before.
But like all tools, getting the most of Midjourney takes time. Below are some takeaways to help others maximize this tool for work.
Midjourney in action
First off, here are some examples of how I've used Midjourney for work.
- For a presentation on paywalled content, I needed an image of a wall with a large gate. Typically walls have negative connotations, but in this context the wall needed to be positive as it was enabling something new (specifically, a new publisher revenue stream). To emphasize new opportunities, I added a prompt of a person in front of the gate with their arms outstretched.
2. For an idea on engineering scaling, I needed a visual representation of the concept of organic business growth employing natural elements, while avoiding a boring bar chart.
3. For a presentation on AI risks, I needed a political cartoon capturing the fear that chatGPT will be replacing engineers, CTOs, and other knowledge workers, and the only thing that can relate to them is Alexa (another victim of GPT).
Using Midjourney for Business
Here are the most important things I wish I'd known when I started using Midjourney.
Setup – After completing setup, use the /subscribe command and sign up for the Pro Plan ($60/month) so that you can enable /stealth mode so that your images aren't shared publicly. This will also give you 30 fast hours of compute time so that when you're on a roll you won't be slowed down.
Prepping images for presentations – Add white background to your prompts so that you can more easily punch out the background for more flexible copy / pasting. In many cases I needed images to support alpha layers (transparency) so that I could creatively deploy them. (Once the image has a white background you can punch it out in Photoshop to make it transparent).
Aspect ratio – Don’t settle for square images. Figure out if you need a landscape or portrait oriented image before you start prompting, and then append —ar at the end of a query to change the aspect ratio. For instance, if you wanted a 16:9 output you would add --ar 16:9 at the end of your prompt.
Consistent styling – Midjourney can stylize an output in thousands of different ways. Just use the style you want in the prompt, like “alien smelling roses, in the style of Picasso” or “alien smelling roses, illustration style”. If you’re generating multiple images for the same presentation, the pro move is to reuse the same style so that you can keep the visuals consistent. Note that Midjourney version 5 typically outputs photographic quality by default so you’ll need to either use V4 ( --v 4 at the end of a prompt) or specify a style if you don't want photos.
Don't be a perfectionist – Be happy with getting 90% of the way there; it’s shockingly easy to get close to what you want with your first prompt. But the last mile can be frustrating. So you will hit a point when editing your prompt and/or generating variations isn’t getting you any closer to your goal. In that case, see if you can generate a new image in the same style, and then merge the two outputs in Photoshop.
Photoshop skills still needed – For my political cartoon of an engineer commiserating with Alexa, I couldn’t get both the menacing GPT robot and the engineer on the sidewalk all in the same output. So I used the same style, and then separated my prompts to make multiple images (making sure to have a background that was easy to punch out) and then comped everything together in Photoshop.
Iterating on an image – To iterate on an output you need to copy your old prompt, paste it in again, and then edit the text. It’s a bit clunky, but it works. What you can’t do is arrive at an image that you like and then ask Midjourney to add something new to that image. It needs to start over every time. (Technically you can download the good image, re-upload it, and then iterate from that, but the output is not going to be anything like the original. Iterating prompts and generating variations take time. I typically go through 30-50 prompts before settling on an image.
Text – Don’t try and output specific text strings in Midjourney. Though it will try really hard to render it, it just doesn’t understand it yet. Try Adobe Firefly for that.
Photos – To use a photo to “influence” your output, just drop the image into Discord and then right click on it, hit copy link, and then paste that URL into your prompt. Midjourney has put a lot of effort into preventing users from generating deep fakes – which is a good thing – so don’t be disappointed if you try and generate a photo output of yourself and it doesn’t look exactly like you.
Things to be aware of
- With current litigation to do with image copyright, it’s wise to consult legal if you want to use an output beyond an internal presentation.
- Prompts still need work to output diverse groups of people. This is getting better, but you can also be intentional in your prompts by adding things like "many races, men and women, all ages, diverse” to prompts that have people in them.