Currently viewing the tag: "photo-realism"

In celebration of International Women’s Day, I decided to use AI image generation to create some beautiful, photo-realistic portraits of women from around the world, in traditional dress.

Those of you who know me personally will know that I worked as a Freelance Graphic Designer for many years before becoming a Learning Technologist. Whilst I no longer work as a Graphic Designer, I do still keep my ear to the ground in the Graphic Design communities where there has been such a mixed reaction to AI image generation. It has been really interesting to watch the reactions over the last six months, as AI image generation has improved so much in such a short time. I, like many others, see it as an amazing and powerful tool that can work with a digital artist to produce pieces of work in a fraction of the time.

Limitations at the beginning

I used deepai.org, a free online text-to-image website that doesn’t require any login or registration, to play around and explore this new medium. For those simply wanting to type a keyword or two and see the result, it is really good fun. You could waste hours of your life just typing in different keywords and seeing what you get. It’s just so much fun creating weird and wonderful images! Here was one of my first attempts. I’d just asked for a sunflower, I wasn’t expecting a little Panda face peering out from the middle. I quickly learned what I’d done wrong and was determined to get better control over the results.

Sunflower illustration with panda eyes in the middle. AI Image Generation

Writing Prompts: There is a skill to it!

I went back to the Graphic Design community blogs and YouTube videos where I’d seen absolutely stunning results, with futuristic and surreal city-scapes and weird fantastical creatures. Most of the designers I saw are using a platform called Midjourney. Midjourney offers a free trial and then a monthly subscription of just $10 a month for their cheapest plan. These AI artists, and yes I will call them artists despite how controversial that is, are using and sharing specific prompts that, through trial and error, they have found work really effectively to achieve certain visual effects.

It is quite well accepted that AI can’t do hands and often can’t do faces particularly well. I often see lions with 6 legs. You end up with some fairly disturbing images sometimes. The image below was created when I asked for a scene with The Queen of England. You’ll see in this example what I mean when I say it can’t do faces. (The little furry, three-eared creature, with no eyes, was supposed to be Paddington 😞).

AI image generation. Image of the Queen with a blur where her face should be.

Harnessing the power of AI

Despite using a free AI image generator, which states very clearly on its homepage NOT to expect photo realism, I was amazed by these results. I was blown away by the quality of all the images that came out, and the ones I’ve omitted from my gallery below, I’ve only done so because they looked a bit too airbrushed.

The prompt I generally used went as follows; (X is the nationality)
Create a portrait of a traditional X woman, clear facial features, cinematic, 35mm lens, f/1.8, accent lighting, global illumination"

Why don’t you give it a go and try a different nationality? I’d love to hear how you got on.

A much longer version of this prompt was originally shared on Reddit and I took it from a YouTube video. You can watch if you want to understand more about what some of those elements are in the prompt — https://youtu.be/KXCVBu4btUk. (Photographers reading this will already have recognised some of those terms).

How AI image generation works

You may be looking at the images, wondering who these people are and whether they want AI using their faces. Well, you may be surprised to find out that none of these women are real people. They do not exist. You will not find these faces anywhere on the internet. Of course, coincidently, they might happen to look like someone in the world, but the faces, along with the rest of the image, are created by AI.

For example, if I want to paint a picture of a horse, I don’t have a horse to look at, so I’ll find a number of images on the internet to observe the proportions, the face shape, the mane, etc. I look at lots of different images from different angles to get a good idea of what it looks like. Then I’ll do my painting based on what I’ve observed. Similarly, AI will look at thousands of images on the Internet based on the specifics you’d put in your prompt. It then uses that information to create a brand new, original (Royalty Free*) image just for you. If you don’t like it, you can just tweak your prompt and it’ll make you a brand new, original image.

*Check the T&Cs of the platform you are using

Moving forward with AI

AI is here to stay whether we like it or not. I hope that we can appreciate it for what it can do for us, and embrace the technology. I am all for technology that can save us time and AI image generation certainly does that. Does it replace the artist? No, not necessarily. As you have seen in the examples, there is a skill, and you do have to learn how to get the best results. I look forward to seeing the images get better and better.