How we, with the help of Midjourney, visualized concepts from the world of IT on the Imaginarium maps
Hi all! I would like to share an unusual case from my practice – creating illustrations for Imaginarium game cards using the Midjourney neural network. Our team visualized expressions and concepts from the world of IT. We imagined what an Agile board game might look like, a user path, remote access, brought video surveillance systems to life, and fantasized about the human appearance of a backend, startup, or deadline. There is a new article about this on the LANIT blog. Bonus – under the cut you will find a lot of beautiful pictures and a link to a guide I compiled on working with promts in Midjourney.
It was not without reason that we decided to do something bright and unusual with the participation of artificial intelligence: LANIT has been studying the potential and areas of application of AI for a long time. And with the active development of neural networks, it has become possible to use AI technologies not only in large projects, but also to solve small problems. Over the past year, we, like all users of neural networks, have been learning through trial and error what Kandinsky, Stable Diffusion, Masterpiece, Midjourney and other networks can do. After all, the limit of the capabilities of generative models has not been studied, as their creators also say.
The approaching New Year helped us set a creative task for neural networks, and at the same time for our art department. We decided to give the company’s partners a memorable gift, but one that won’t gather dust on the shelf. So, we decided that it would be an Imaginarium board game, but an unusual one, with a set of cards about the world of information technology.
Data saver, sound systems, captcha and more than 60 cards
We planned a month and a half to implement the idea. During this time, we, the designers, had to draw 65 cards, on each of which we would present a concept or catchphrase from the world of IT. There were also nuances.
We were required to visually represent processes and concepts that certainly cannot be seen in paintings. For example, startup, data keeper, sound systems, deadline.
Cards should be in different styles. Everyone who played “Imaginarium” knows about this.
Here are just some concepts, objects, processes, etc. from the world of IT, which came to the minds of our fellow marketers during a brainstorming session and passed them on as technical specifications.
This list became our starting point.
We worked in tandem with the Midjourney neural network and the Discord bot, in which we wrote the prom. It is popular, and in the public domain you can find a lot of advice on composing promts (text queries to a neural network). In addition, Midjourney was in the lead in the Google rankings for a very long time (although, as can be seen from the second graph, Stable Diffusion already caught up with it in the fall of 2023).
I will mention a few more important technical points. At the start of work, we paid for a subscription to Midjourney with a card from a foreign bank (in the summer of 2023, the cost of a subscription in rubles was about a thousand rubles), since free use of the service was no longer available.
Also, thanks to the subscription, you can add the Midjourney bot to your server in Discord (a communication platform into which a neural network is integrated) and interact with the AI assistant without delays or the need to search for the generated image in the stream of messages of other users.
Colleagues can also connect to the server. However, please note that in order to interact with Midjourney, they will also have to purchase a subscription or create images using your paid profile.
“Obedient and Naughty” promty
At the start, we were full of enthusiasm and thought that we would cope with the task very quickly: after all, we were helped by the almighty AI. In addition, artists, long before the advent of information technology and even more so artificial intelligence, tried to depict on their canvases such difficult concepts as life (the painting “Life” by Pablo Picasso), memory (the work of Salvador Dali “The Persistence of Memory”).
In reality, everything was not easy. The Midjourney neural network turned out to be not such a magical program and also naughty. “Flexibility at minimum wages,” I would say.
If you need to portray a cat flying in space, this program is definitely for you. If you, like us, need to create a picture codenamed “user journey” or Agile, then you will have to work hard here.
The neural network runs on the Discourd bot and was trained on a non-trivial data set, apparently. Therefore, getting what you imagined will be very, very difficult. Of course, it happened that the neuron understood us perfectly, but more often it took several hours, or the whole day, or the whole night to generate one image (preparing additional instructions to finalize the drawing). The neural network resisted and saw the result in its own way, not what we expected from it. And only our team’s artistic and imaginative mindset and creative approach conquered Midjourney. We forced her to “listen” and pushed her in the right direction through routines. During our work, we even developed our own slang, for example, “obedient” and “naughty” promt.
I’ll tell you more about this using examples of creating several cards.
Python developer
It was interesting to work on the image of a Python developer. Having an idea of how difficult this programming language is, we imagined a serious man with a majestic snake. This is the card that Imaginarium players will eventually see.
But it didn’t work out right away. First, the neural network gave us a cute cartoon weirdo upon request.
The next attempt to make the developer more serious resulted in a sullen old man.
And only after a day of battles with Midjourney we achieved the desired result. We also share the inspiration that led to it.
Keep in mind that each word from the associative series contained in the promt changes the picture beyond recognition. A neural network can either come up with the details on its own, create a composition, or destroy the entire idea by starting to be creative in a different direction.
Digital audio
We were lucky with the “digital sound” card: Midjourney guessed what we wanted on the first try.
Smart House
We imagined a technological structure with arms and legs, which rolls down the road on a skateboard, wearing a fashionable baseball cap. It seemed like it could be simpler. However, our AI assistant again flatly refused to obey and stubbornly gave us people (even on skateboards and in baseball caps) against the backdrop of houses, but did not “humanize” the building itself.
When we asked Midjourney to draw houses with eyes, this is what we got:
We had no choice but to abandon the original idea. We decided to make something bright and unusual in 3D style. And, indeed, the result was an eye-catching image, but completely unsuitable for the game (everyone who has played “Imaginarium” knows that nothing specific is drawn on the cards, but only objects and objects that evoke associations with someone or something ).
However, the picture in the middle gave us an idea of how else we could depict a “smart home”. We finalized it and it turned out to be the same picture that we took into our set of cards.
Collection
The illustration for this card was born by chance. The task was different – to draw an ATM. We tried for a long time to depict him in a fantasy form, but everything turned out wrong. As a result, I began to imagine what happens to an ATM during the day. For example, collectors approach him. My imagination pictured supermen in charge of important transactions with banknotes.
I found an image of the person I imagined as a collector, and I really liked the style. In addition, I used a regular photo of an ATM, or rather, only part of the photo with flying bills. And she suggested that Midjourney make a hybrid of the two pictures. Industrial products once again helped direct the boundless imagination of AI in the right direction.
The facial expressions of an emotionally impenetrable person are reflected, which is why we liked the illustration.
Using multiple images to create a hybrid is an effective way to force a neural network to break away from its patterns and produce great results.
Office plankton
“The moment decides everything”… In the case of a neural network, “style decides everything” (at least, a lot). Even at the beginning of working on a set of cards, I found a useful style sheet on one of the sites. It was a huge list of artists and illustrators with examples of how an image could look if their style was applied to the processing.
At first, I naively believed that this would be the case. However, when I used the styles mentioned in the table for the illustrations we came up with, I got a completely different result. When it seemed that the style of the works of Andy Warhol or Tove Jannson was ideal for the card, disappointment awaited. It was necessary to select a style for each of the pictures through numerous trials and errors.
And yet, knowing the styles of artists or animators is very useful when writing prom. For example, turning to the work of Canadian illustrator Jon Klassen helped us when creating the “office plankton” card.
Remote access
We decided to visualize “remote access” in a non-standard way – we placed a photo of the computer in a forest landscape. The style of Claude Monet also turned out to be appropriate.
Sometimes it was difficult for us to choose one image, since Midjourney generated several interesting pictures at once upon request.
Instead of a conclusion
The main thing that we realized for ourselves while working on our “Imaginarium” is that you need to communicate with Midjourney like you would with a baby. However, this is not surprising, because the neuron was born only in February of this year. You will have to adapt to the world of algorithms and give the network tasks at the command level without niceties, and it is better to avoid negations in all areas (but, not, except, without).
The golden rule for creating all industrial items looks like this:
At the end of the prom, the resolution in which the illustration is required is indicated. We ordered images with a resolution of 8k: this value is enough to obtain high-quality images for the web or for printing in small formats.
So, let’s take a look at the pros and cons of working at Midjourney.
Pros of working at Midjourney:
clear interface,
many styles,
There is an option to create a “hybrid” from several images.
Disadvantages of work:
it’s difficult to achieve what you want (especially abstract terms of reference),
it is difficult to control the result.
It would definitely take more than one, two or three months to create these same images by several artists and illustrators. Or there are a lot of illustrators, and, accordingly, a lot of money to pay them. The ability to speak the language of commands and instructions, the “language of the robot,” made it possible to speed up the entire process and complete the work efficiently and on time. This is not easy (after all, there were no professional industrial engineers on our team), but this skill is the future, and without it it will not be possible to move forward.
Finally, as promised, there are many beautiful cards from our set. Brew some tea and get ready to watch.
I hope you found the content useful. Share your commercial graphic works created using neural networks in the comments. And grab a free guide to working in Midjourney.
