How AI image generation is going to change the world

Matthijs Langendijk
7 min readSep 21, 2022

--

You might have seen the news already, where an AI-generated image was submitted to a digital art contest — and won! It sparked a big debate about whether AI-generated images should be considered for such competitions. It showed the first sparks of a world where images like these will play a big role. And that doesn’t just apply to art, it’s also going to transform basically every other aspect where images, pictures and graphic design come into play. Here’s why.

PS. all of the images you’ll find in this blog are AI-generated.

Everybody can do it

Honestly, the most impacting factor, at least from my perspective, is that anybody can start making digital art and images. And they can start that today. It doesn’t take much training besides understanding the parameters you’re working in, let alone having the need for years of experience and of course a keen eye for design. Just enter the prompt and you’re off to the races.

To show you an example of this, I’ve been playing with a tool called ‘Midjourney’. This tool, similar to DALL-E by OpenAI, is able to generate images from just a simple text description. In the case of Midjourney, they have exposed a Discord bot that you can write your prompts to. The bot then takes a few minutes at most, and your image is ready. Let me show you some examples of the images I’ve created, along with the texts I’ve asked the bot to create for me.

prompt: a pizza on a plate, a single slice with cheese pull, basil, restaurant, table, Italy
prompt: a full-scale yacht, jeff bezos, biggest ever, billions

Looks pretty realistic right? If I would have told you these were pictures I had taken over the weekend, would you have believed me? I think you would have. And that’s showing the true power of tools like Midjourney. It’s incredibly easy to generate images, and they are extremely realistic. The only real limitation you have is really your imagination.

It shows how easy it is, even for someone with as bad a design eye as me, to create realistic-looking images. Because the AI uses the text I’ve given it, you’ll definitely get more accustomed to what is important to the algorithm. For example, giving it the prompt ‘Rembrandt’ will give you a picture closer to the style of Rembrandt. So there are definitely some tricks that you can apply to get the results you want. But it definitely doesn’t require years and years of painting experience, and a lot of talent, as you would need for the real deal.

Time and money saving

Say you’re working on a big marketing campaign that requires you have tens or even hundreds of different images of the product you’re selling. Before AI image generation, you would need to rent a studio, dress it with a good background, lighting and ofcourse pay for the staff as well. Then all the images created get thrown on a bunch and you pick and edit the ones you like the most. It’s a both time consuming and expensive process.

How cool would it be if you can instead create most of that image with AI? The lighting, background and place where you want to highlight your product can all be generated by AI. Then all it takes is for you to place your product in there with some clever image editing and you’re ready for the campaign. The freedom to try as many things as you can think of, without having to spend a big amount of money changing the scene in a studio for every small change; AI is going to revolutionise the way these types of images are getting created.

Applications for AI generated images

Images are everywhere. You wake up with an image of that piece of news you’re reading, you stare at an image you have on your phone’s lock screen many times a day. The list is very long where there are images in play.

Stock Images

The state in which AI image generation finds itself now, makes stock photography the most interesting place. The AI currently has quite a hard time generating very specific content. Yes, you can ask it (the AI) for a boat or pizza like I did earlier. But you won’t know what it’s exactly going to look like until the image has been generated. It makes for a more creative process, but also makes it harder to get exactly what you want. That’s why stock photography is the most ideal place to start here; since you’re not working with very strict requirements and can simply generate generic images that can be used for multiple use cases.

prompt: a living room focused on the big television in the middle, stock photograph, centered

Interior Design

Given that you can ask the AI for basically anything, you can also ask it to design the interior of a house for you. Imagine you are thinking of getting a new kitchen, but don’t really know what style you like yet. You might have an idea of the materials or colours though. And that’s exactly what you ask the bot to create for you. The images generated can then be used while you’re actively shopping for a kitchen. Just show it to the salesperson and they’ll point you to the kitchen that best fits your desires. Here are some examples:

prompt: a kitchen with fridge, cooking station, pans, industrial, stainless steel

Game Character Design

Something that the AI also excels at is generating people — especially characterised. For the programmers under us, it is often very difficult to get a good design going for the game you’ve been putting many hours in. You have the creativity in your mind, but don’t have the means to make them come to life in a design. That’s exactly where the AI can come into play as well. Simply ask it for the character that you have in mind, tweak the prompt until you get the result you like — and use it in your game.

prompt: a short male dwarf, red beard, wielding an axe, 2d art, character creation, octane render

It’s not always great yet

The state of AI image generation today is definitely still not where it can be. I can best show that with some example images that I tried to create where I tried to include the branding of some companies. In this case, I tried getting a restaurant image for Domino’s, Burgerking and Mcdonald's. You’ll see from the images that, while it does understand what the brand is, it’s not giving me those exact results. It shows that the AI does need some more work, or maybe specific tweaking, to work for those scenarios too.

prompt: a burger restaurant, burgerking
prompt: a burger restaurant, mcdonalds
prompt: a pizza restaurant, Domino’s brand, Italian

But it’s also just really cool though

Boy is this stuff cool though! Your imagination can really run free! I’ve always had so many interesting and creative thoughts in my mind that I could never make into reality. I know what I want it to look like, but I don't have the creative hands to actually make it happen. And now I can, and so can you!

Let me just end this blog by showing you some insane images that are all generated by Midjourney. It should show you the true power of AI image generation, and why it’s definitely going to take over the world.

prompt: a museum hall with pieces of art displayed along the wall and on the floor
prompt: cute house next to a blue lake, Pixar, ultrarealistic, aquamarine, aqua, aquablue, wooden house, old house, old door, volumetric lighting, octane lighting
prompt: an old grandpa reading a book in the living room of a house
prompt: white male with grey hair, old age, 85, rumpled, faded skin, portrait
prompt: Arabic male, short black curly hair, round face, neckbeard, black moustache
prompt: pale woman, dark brown hair in a ponytail, very slim glasses, kind face, smiling, Romanian
prompt: a dark blonde haired male, Jewish nose, blonde eyebrows, grey eyes, high forehead, curled hair, pale skin, hamster cheeks, small beard, slim face
prompt: steampunk world, skyscrapers, train tracks, wheels, ultrarealistic, hyperrealistic, volumetric lighting, octane lighting, unreal engine, high detail

--

--

Matthijs Langendijk
Matthijs Langendijk

Written by Matthijs Langendijk

Developer from the Netherlands. Working as a Solution Architect at Triple. www.linkedin.com/in/matthijs-langendijk/

No responses yet