AI vs Art Directors

If you are reading this, you probably grew up making drawings using colour pencils and crayons. Before you scoff at how ridiculously broad such a segmentation is, consider the radical future that awaits us in this aspect. The generation born right now will probably not go through this process.

I am not an overzealous futurist, but I am as bummed out as the next guy by the fact that we don’t have flying cars like in The Jetsons. But I am a parent, and the experiences I have gone through in the last couple of months with my eight-year-old ‘playing around’ with AI tools makes me think that very soon the expression ‘back to the drawing board’ will be as meaningless as the floppy disk masquerading as the save icon.

At this point, you are probably finding that this is going to be yet another one of those preachy pieces about the breakneck pace of generative AI and how we are all doomed unless we get with the programme. Maybe there will be a little quip at the end about how this article itself was written with the help of ChatGPT. If you have opened up LinkedIn lately, you will notice that everyone and their chowkidar are drooling over the potential of this technology, yours truly included.

I will resist the temptation. My purpose here is to advocate in favour of a very defined and practical use case; that of a digital agency in Pakistan leveraging a specific tool called Midjourney. However, I am more of an enthusiast than an expert, so please forgive me if I make some naive observations.

Midjourney is a generative AI tool for creating visual imagery. As of now, there are two other ‘big’ names in this category: OpenAI’s DALL-E and Stable Diffusion. I have personally used all three and even dabbled with newer offerings from the usual suspects, such as Microsoft’s Bing Image Creator and Shutterstock’s AI add-on. In my tests, none of them comes close to the practicality potential of Midjourney right now. Especially with its newest version, it’s less likely to distort faces and add an extra finger or two, like the other tools often do.

Unlike other platforms, Midjourney runs from inside the popular real-time chat service Discord. If you are a Millennial or younger, you probably know about Discord already. If you are older, think of it as IRC on steroids, complete with ‘rooms’ and public channels where people can interact. Once you subscribe, you get to interact with the Midjourney chatbot and are invited into a bunch of channels for ‘newbies’. In order to enter a prompt, you need to address the bot (or be in a channel where it’s active) and start typing with the command ‘/imagine’. Being on a public, open-to-all style platform like Discord means that you are able to see what other users are prompting the bot to do and try to learn from the process. The community aspect is really helpful!

Once inside, you are ready to play. The free trial account has enough juice to last you a couple of weeks, so don’t worry and start prompting. I recommend starting with something simple, like ‘a cat floating in space’ or ‘construction worker working on a site’. From here onwards, it’s your experimentation that will take you forward.

But I know you (like me) are a Pakistani who worries more about practicality and will need to see if this is worth your while before you go through the motions. Let me attempt to predict some of your questions and answer accordingly.

1. Can Midjourney give me high-resolution images?
Yes, the images are good enough for digital marketing. You can choose your aspect ratio, and you will get the equivalent of 720p, which will work for most use cases. If you want further upscaling, you can go to a third-party tool like Upscale Media and further blow it up, but don’t expect pixel-perfect clarity suitable for print.

2. Can I use it to superpower my presentations?
Yes, we have already used AI-generated imagery from Midjourney to win a couple of pitches, and from the feedback, it seems the recipients were especially impressed with the visuals (though we had not yet revealed that they were made by AI). If you are in the Pakistani agency business, you know that everything needs to have a ‘reference’ in order to be approved. This helps you create those relatively easily, without relying on ridiculously staged stock photos.

3. Can it help me create product-specific ads out of the box?
No, neither Midjourney nor its counterparts can generate product-focused materials with precision yet. Even if you try to feed it specific images of your product so that it can ‘learn’ and integrate them within a scene, it’s bound to make errors in generation. For example, the shape of your shampoo bottle will be inadvertently changed, or it will add an extra button or two to a headphone, making it unusable to use as an ad. Instead, use the tool to create the background scene for your ad and layer on the product manually.

4. What uses can I put this to relatively effortlessly?
In our experiments, the art part of social media graphics was handled pretty well by this tool. It is also great for creating arresting visuals to put on website landing pages or mocking up UI inspirations. Think of this as a panel of varied artists at your fingertips. The more absurd the request, the better this tool will perform versus traditional designers. Think of realistic executions of fantastical situations. For example, the viral photo of the Pope in a bomber jacket was created using Midjourney.

5. What IP considerations do I need to keep in mind?
Just like any emerging tech with massive UGC, the rules are sketchy. With a paid version, you get a commercial license to use the artwork as you may. But of course, if you are specifically using a brand or celebrity’s likeness, it could be problematic. It took us just two prompts to get a perfect image of Iron Man in the streets of Karachi for an awareness campaign we were contemplating for community policing, but we eventually decided against it because of copyright infringement. Rule of thumb: if it doesn’t have a recognisable public figure or product, you are pretty much in the clear. Exercise caution though.

6. Does this mean that visual artists like photographers and designers are obsolete?
I would steer clear of any of these bold proclamations. If anything, I have seen Midjourney help photographers create fantastic visual eye candy in which to place products. It can reduce the expense of props, location and lighting – so long as you have the eye to shoot and place the product seamlessly. You still need expert designers to composite in the necessary imagery and match the brand’s visual tones. It helps visual artists push the boundaries of their craft without relegating them to prompt-writing keyboard warriors.

7. Are there any specific instructions I need to learn to use Midjourney?
The prediction engine is itself pretty intuitive, but you can use specific commands to arrive at the perfect image. However, you need to realise that this model has been trained on millions of images and their metadata, so you need to reference those labels if you want it to mimic something. You can specify camera types, lighting scenes, art genres, and anything else into the prompt and it will serve it pretty accurately. It is useful to specify an aspect ratio with the command (“--ar 2:1” or “--ar 16:9”) to get the right proportions. Use “--stylise 1000” to let the AI go into overdrive and forget reality for a bit.

In a nutshell, despite our rapidly tumbling currency conversion rate, my hypothesis is that digital agencies in Pakistan need to buy these AI tools for their design teams if they are up for them. If nothing else, it will help us reduce some of the grunt work that goes into the process of pitching and realising campaigns. Given how quickly these things evolve, I know fully well that the specifics of this article will not age well. But if it can inspire just one ‘traditional’ designer in an agency somewhere out there to try this tool out, I will consider my effort worthwhile. And that matters, because this piece was NOT written with ChatGPT!

Umair Kazi is Partner, Ishtehari.