Artificial Intelligence and Imagery: A Personal Journey
(This is an image of Queen Ólöf of Denmark holding her son Fróði.)
Usually, when considering your AAR, the words come first. To those who do not see themselves as visual artists, take heart: this new way of making images starts with words too. Word choice is paramount.
For some of us, putting out an AAR without illustrations or artwork is like taking your baby out in public without a diaper or clothing. Many of us treat our AARs just like babies: caring for them; seeing them grow; and hoping they prosper. So dressing them fashionably is part of that caring ritual. But what to do if you aren’t artistically inclined when it comes to visuals, hate searching the web for public domain images, and don’t have a budget for fancy software or to pay for images?
Certainly, from the title of this, you can glean the answer. But do you really want to deal with the controversies and ethics of Artificial Intelligence?
Those who don’t will find other solutions.
However, as someone who manages an AAR inspired by graphic novels, it is essential for me to approach the use of AI in an ethical way. I don’t have a budget to pay artists (as some have done, even if it is a relatively small amount) or to pay for software licenses. Currently, I use the free version of
krita for complex editing of images. However, as a Mac user, I find its internal tools work just fine for the basics. And there are free online options too. (More on that later.)
But this is not about software. It’s about solutions for creating images. So back to the start of this post. You have probably already figured out that the first image is from AI. Currently, this is an image I don’t plan to use in the Paradox forums, so we can deconstruct it a bit here. That opening image was constructed using
Bing, the very first AI platform that I started using about 11 months ago. However, that first image was made with the DALL·E 3 image generator, a much more advanced AI than Bing’s original DALL·E image generator. That image was also made with the Assistance of
Microsoft’s Copilot AI, which just means I had an interaction with an AI platform that was in the form of a conversation rather than just punching instructions into a prompt. The key parts of getting this result were the instructions to Copilot: a Norse woman in a royal family holding a baby in the 8th Century. In addition, the instructions also asked for the image to look like a Romantic era piece of art painted by Danish artist J. L. Lund. Sometimes the AI doesn’t give you everything you want. This prompt was actually for a different project and I didn’t like the results, but I saved this image and when I needed something for Queen Ólöf, I thought this fit perfectly. Here’s
a link to see some of the prompting.
So the first lesson is that working with AI can be hit and miss, and sometimes you have to try different methods and prompts to get what you’d like. But sometimes you will also get it on the first pass.
Some may already be questioning the ethics of my use of the style of J. L. Lund in these instructions. As Lund has passed away long ago and his works are all in the public domain (unless you own the physical copies) the generally accepted standard is that I am not hurting a living artist by borrowing this style. As someone working with Danish characters and an AAR based around the Danes, I have found Lund’s style is one that evokes an aesthetic that many associate with the Norse and Vikings. Also, I have used Lund’s works and credited them both in my AAR and in other projects. Usually, I link to the artwork on the AI platform which often gives some transparency to how the work was created. (But more on linking and Bing, a bit later.) The big debate around AI is how the work of living artists and writers are being used without compensation in the creation of new art, images and stories. Personally, I do not agree with this practice and if I feel a platform is stealing from others, I discontinue using it. However, some have done tests to push AI to take intellectual property and it is possible to do this easily on some platforms. Personally, I find that unethical. Some may say it is part of satire or fair comment, so I suppose context is everything in these discussions, but in general I don’t think artistic property should be reused without permission and without compensation for living writers and artists. Just understand though it is difficult to discern sometimes which platforms are the best, ethically.
That important tangent was probably too much for many folks but it needed to be said.
Let me take you back to my early stumbling about trying to find images for
my AAR.
The
first chapter of
Lost Seasons of the Danes features two images from Bing.
One is a photorealistic portrait of King Harald “Wartooth” and the other has become the logo of sorts for the AAR: a woodcut of a Norse funeral pyre.
These images were created with the original DALL·E image generator and you can compare them to the image at the top of this post to make your own comparisons regarding visual upgrades in DALL·E 3. I still love these images, but Bing is no longer the only platform I use and for good reasons.
One of several drawbacks to Bing is that although it appears that it stores your images, after a while it only stores the thumbnails of those images. Some might say: what do you want for free? And I would agree. If you use Bing, you should download everything to a hard drive. I discovered this the hard way. One part of using AI is you will likely create a lot of images that don’t match what you need immediately. That’s why I like storing them to sort for other projects later. But storage in the cloud online is easier than taking up space on your hard drive. If you try Bing, don’t make this mistake. Other platforms have proved to be more secure, even for those using the free versions. Bing also offers links to your images so folks may see the prompting and construction. But again, as time goes on, it only keeps those links current for some images, not all. This is frustrating.
One more aspect of Bing is you can easily use Microsoft’s
Designer, an online digital editor, to alter the images you create. Designer had an eraser feature that was good. For me, this was needed because the AI always insists on creating horned helmets for Vikings. However, the eraser doesn’t work as magically as those you may see on television. You might highlight what you wanted erased 20 times and still it would not be completely gone. This was another built in frustration for folks using the free version. You could only perform so many special effects functions in 24 hours. So it might take days to fix an image that wasn’t just right.
Now, Microsoft has discontinued the eraser function altogether in Designer. At least for the time being.
This is why, although I still use Bing and Copilot in a pinch when all else fails, I really need to be pushed in that direction. Sorry to say they are last on my list.
The primary platform I prefer now is
Playground. Playground is great for creating portraits, and ultimately, a lot of my AAR is built around those. Playground uses its variant of the Stable Diffusion image generator. Playground also allows the use of reference images. Bing allows that now, but other platforms had that feature before Bing. So if I have a public domain image that I want to use as a model, this helps the AI create something that approximates it. It is still hunt and peck though sometimes, even with this feature. Here’s a recent example of something from Playground that is probably not going to show up here on the forum anytime soon.
(A Finnish woman captured by the Danes during a raid.)
However, as someone on Playground’s free plan, I have a limited amount of special effects edits I can make daily and during peak times it may not even allow me to make images. Of course, again, what do you want for free?
So off to find another platform and that is part of the fun of this moment as AI is an experimental trend and there are many new choices if you look. One of those backup choices is
Leonardo AI. Leonardo uses something called Leonardo Diffusion as an image generator. Currently, none of the images I’ve made on Leonardo have come to what I post on Paradox’ Forum but I have used them for other projects and they may show up here eventually. I haven’t found a way to manipulate Leonardo to get the same visual appeal I can get from Playground or Bing but I’m still lurching through it in the early stages of learning the platform. This all relates to the wording of the prompts and the combinations of filters and other visual offerings on each platform.
Here’s an example of one from Leonardo that will likely not be used in my AARs:
(King Fróði attempts to maneuver around a stag during his ill-fated hunting trip.)
This week, I added an image from Google’s experimental ImageFX platform to
a new AAR chapter just because I wanted to show off what it can do. Just like the other platforms, Leonardo has limits on its free plan for how many images I can create. So ImageFX became my third option. However, again, I’m not practiced enough to get the best out of ImageFX yet. It also has limitations on prompts. For instance, you can’t name an artist’s style you want to use, even a long deceased artist. Instead, I ask for the style of Danish artists and I usually get something close to what I want. Here’s the piece that came out of those ImageFX experiments.
By comparison, here’s a similar image from Playground using a similar prompt that I decided did not work.
I’ve also experimented with
RenderNet, because it promises it can lock in a character’s image and reuse it in different settings. With a graphic novel inspired AAR, you might imagine that would be useful. However, despite various experiments, so far I have yet to see RenderNet live up to that promise. Nothing produced there has been worth saving. That said, it is advertising new features and upgrades so I may go back to dabbling with it again.
There are a variety of other programs, and I’ve tried a few others, but these are the best of the bunch I’ve tried. As they say though, your mileage may vary depending upon conditions. Free versions will usually come with limits designed on getting you to purchase a monthly subscription.
You might also consider using the
free online service for MidJourney. That program comes recommended by others and there’s a lot of buzz about it. That will be the next one on my list to try.
There are a variety of image programs you can install on your computer, likely for a fee. Personally, until the bugs of AI are dealt with, I don’t trust it on my hard drive. I don’t want to go too far down the tangent of the
arguments you can read about as writers argue with AI. I’ve had them myself where the AI disputes known facts. That’s the scary part about using it.
Finally, a word about
LunaPic. This is the online image editor I use to add speech bubbles to characters. Currently, I need to train myself better on LunaPic because I know I’m unlocking less than 15 percent of what that image editing tool can do. But if you are looking for a free online image editor and one that has some special effects that are useful, this is the one I recommend. However, I will say the eraser tool doesn’t work as well as some of those connected to AI platforms like Playground. Although, that could be operator error. All these programs take time to learn.
Although far from completely comprehensive, this could be a good start to discussing how you use AI on your AAR or the ethics of AI use for images and other purposes. From experience, I know some folks are reluctant to share their AI image secrets. However, you will see I am fairly transparent about how the images I use are made. I hope others will share here too.
(
Postscript: There are various recent discussions about images which might also be useful to read. You can find a good discussion about image limits for AARs and why you should consider jpeg files rather than png files beginning in
this post of
Jak’s Game Guide: An EU4 Tutorial with comments starting with
@Bullfilter . There is also a brief conversation with
@Rensslaer about
the images in Lost Seasons of the Danes that relate to image continuity. My answers about those can be found
here. You can also see some of the oddities of AI art (a two-headed horse) in
this post from
@Nikolai in his
Dreams of Elysium where he mentions some of the limitations of using free AI image generators. Hoping you agree that the discussion of visual creativity and the process that supports it can be interesting.)