Fundamentals of AI Art Generation

[Chac1]

Some may have noticed a deep dive into AI art posted in the SolAARium recently.

After some debate, the thought was some of this discussion should move out of the SolAARium and into the general forum. But as @Lord Durham noted in that discussion, this thread shouldn’t be just for posting pretty pictures. The concept here is to post some of your art to help instruct others about prompting and the use of different AI image generators. Perhaps there should be an interchange of ideas here.

So let me kick this off by further discussing some of the art in that SolAARium post and expand a bit on what was discussed there. (The ethical part of that discussion is still going on there, and it has expanded to cover AI not only for making images, but also music and writing AARs too.)

Getting AI to work for you in your AAR boils down to using words as prompts to get the AI to make the images. In that post, I glossed over how I had created one piece of art for the latest chapter in my AAR.

For those who are interested, here’s more of a general step-by-step approach. My platform of choice is Playground. So I go there and pick the Stable Diffusion XL image generator. I can choose several different image generators but this is the one I like best. Then I punch in this for the prompt: “A large African man, six and a half feet tall, who is a guard armed with a scimitar dressed in a long flowing white robe painted in the Romantic style of Danish artist J. L. Lund.”

I use the filter style: “Masterpiece” which adds a variety of additional items to the prompt.

I also use a negative prompt. For those who don’t know, a negative prompt supposedly stops items from appearing. But it doesn’t always work. Here’s the negative prompt I used: “ugly, deformed, noisy, blurry, distorted, out of focus, bad anatomy, extra limbs, poorly drawn face, poorly drawn hands, missing fingers, nudity, nude.”

Here’s one of the better results of the four images produced:

Here’s a link to the work so you can see all the settings.

Not exactly what I asked for in the prompt. What’s that giant spear all about? Not to mention the disk hanging from his belt.

Five rounds later, I still don’t have what I want.

You can see from this link to my prompts, I have adjusted the main prompt and changed the filter to “royalistic” but it isn’t improving. What you see below is one of the best of the more than 20 images produced.

I decide to move onward to a different platform: Leonardo AI.

On Leonardo, I pick the Leonardo Diffusion image generator (there are various you may choose) and I pick the Digital Painting filter. However the links on Leonardo only take you to my finished image, not the prompts, so I will include a screenshot that shows you the information after the finished image. You can see I have expanded the prompt I’m using beyond what I used on Playground and greatly expanded the negative prompt too. Here is one of the first results:

Still not what I want. You’ll see after another round with Leonardo, I adjust the main prompt and the negative prompt.

The issue is the curved sword. Neither of these platforms wants to produce the sword.

Here’s what I tried on the last and final round, with the instructions following.

This is when I turned to Google’s experimental ImageFX. The problem with ImageFX is that if you don’t store the URLs of your images you can’t see the prompts. There’s no account with a bank of images like the other services offer. Sorry to say, I did not save the URLs in this experiment so you can’t see this with exactitude. However, I used a prompt similar to this:

“A large African man, six and a half feet tall, who is a guard armed with a curved half-moon-style sword, a scimitar, dressed in a long flowing white robe inside a beautiful Arab palace painted in the Romantic style of a Danish artist.”

This time, I get three possible images to use:

I decide I like the last one best and use it for the post.

However, in replicating this experiment in the preparation for this discussion, I find that ImageFX has improved its quality of what this prompt produces and two of the images are similar to this first round. These I can share with you along with their URLs so you can see how this works a bit more. Here are the two best, starting with this one:

Now, the second and likely the best of the lot:

For those who are new to AI art, you can see, this relates to playing with the words to get the right combination, along with finding filters and platforms that might work best for you.

Perhaps others will post some of their art and share their process so folks here in AARland can learn from our experiments.

 

[CBR JGWRR]

My Experience With AI-Generated Content​

My exploration of AI artwork began with this post in my Life2.0 - A Broken Shackles Narrative, seeking to get artwork for Naomi and Buri, an interspecies couple who, along with their daughter Rivkah, are the three main protagonists of the narrative.

Naomi is an Ndebele (Zulu offshoot who headed north from KwaZulu Natal into what is now southern Zimbabwe; my wife's home country) woman who undergoes a transformation into wolf-like body during the events of the narrative. Buri, her husband, is a Xenaya; a race of desert-dwelling ambush predators who are best visualised by starting with thinking of an upright gorilla, with the head of a smilodon and auroch horns. Rivkah is also a Xenaya, but being adopted from emergency caesarian section she looks nothing like her Dad.

In thumbnail form: (click to expand)

The first is Naomi and Buri on their wedding day, the second is Naomi and Rivkah. Both have been modified to make them closer to my vision; Rivkah's arms and Buri's suit and teeth are obvious alterations, but Naomi's veil is also added in. Rivkah here is also Naomi's perception of her daughter - Rivkah has many interesting dynamics in how she is percieved, and here her Mum sees her cuddly not-so-little any more daughter.

This is perhaps how Rivkah sees herself:

Feminine, yet still a certain intimidating potential.

By contrast, this is how others see Rivkah:

You know, as the carnivorous beast that she physically is.

And this is how Scipio Of Unity - Her newly wedded Olinbari husband - sees her:

Scipio, being a sixteen year old male, is thinking that A) he does still recognise he's dealing with an apex predator, but also B) that she is hot to him, which shows in the loin cloth that draws his eyes to the sides of the cloth and seduces him to remember what she has beneath, large breasts (relative to the Humanoid Olinbari) and wide hips.

I'll also add a few more to explore how the AI interpreted the promt I gave to get Naomi's wolf-like transformation:


This one is one of the first images the software generated that matched the appearance I had in mind for Naomi's new body; the braided hair, the sense of agility over the brawn that typifies her husband.


This one captures the feeling of how she appears to others best, as well as having the most suitable colouration of armour.

These three were other concepts towards refining her appearance. The first conveys the agility, the second conveys the intimidation factor and almost has a proportional sword, while the third has aspects of the first picture, but crucially, has the cape in the right place, and toes.

The prompt for these?

Variations on:

"A humanoid athletic mammalian. She has sabre teeth, horns on her head, and thick black braided fur, standing on two legs."

Different settings and filters get the variable outcomes, plus adding bits like black armour and red cape.

To finish up with these took hundreds and hundreds of images - the overwhelming majority were white Human women with stereotypical devil style horns. It was very much an exercise in patiently generating images until I got the rare matches.

Now, let's look at an example of getting loads of great results: "Ruqayya bint Rahil Amah-Masih"

That is the prompt; I literally put her name in. No filters, no changing settings, just pressing generate. I'm narrowing this down to just the best four that the software gave me:




The first captures the sense of purity of purpose that Ruqayya is intended to have. The second looks a little too young, but captures her style. The third inspired Ruqayya's introduction in the narrative by inspiring a connection to the natural world, while the last was generated the last of these four, but captures her best.

The other two Human characters I've put the most effort into are Alexandra and Thando HaMaadimi. Alexandra enters the narrative as a freed Indentured Asset with Valerius Albanius, an Olinbar defector to Life2.0 who sells off his family's MSI subsidiary to buy a ship and as many Indentured Assets as he could; Alexandra believes herself the heiress of the Autokrator, and took the tactical officer position, and convinced Naomi to give her a command role. This ended badly as Alexandra encounters MSI's elite special forces head-on and suffers severe injuries in the resulting confrontation. Soon after, Thando HaMaadimi - Naomi's twin brother from an IVF treatment - joins Life2.0.

Alexandra carries a dual heritage of Celtic and Greek influences in her genetics, although the Celtic coming through her father is something she hides. She wears her idea of ancient Macedonian bronze armour with a red tunic around, something the AI doesn't pick up on.



But what the AI does capture very well is her appearance.

With Thando, he is very much inspired by Assassin's Creed in terms of his outfit, but adapted to the setting he is in. Generating him therefore, wasn't easy - it doesn't have the context for a guy who hides a shops worth of combat and throwing knives and pistols under a trenchcoat, and well... The closest is:


Which I'm still unhappy with.

Getting it to do Olinbari... Not so easy. Over the course of Life2.0 and thinking about it's unwritten prequel the Olinbariad, I have a rough idea of the two senses of Olinbari architecture, with imposing and grand opulence that flaunts their sophistication for public perception facing architecture, and functionalist and brutalist architecture everywhere else.

Starting with the aesthetics:


These are my two favourites for the rather generalist prompt "Aliens with a Roman aesthetic, Stellaris Style." with the first representing a Shareholder's urban residence, and the second the kind of hall the sub-plot of "The Midnight Gala" takes place in.

The Olinbari themselves, I am yet to get a good enough image to be postable. I began with using the image-to-image feature to duplicate the base race the Stellaris devs tweaked to get the Olinbari, and made the software iterate changes, but I wasn't happy with the results.

If my experience with Humans, Xenaya and Olinbari can be described as hit and miss, the results for trying other alien races from the narrative are such I opted to stick with the base Stellaris images after working on Daas and Vorosh - a molluscoid couple; Daas I'm not sure what the Stellaris image depicts, while Vorosh is a multi-armed snail. Even after hundreds of images for both of them, I am yet to find one for either I think is suitable.

Regrettably, I have to go to work, so I'll write a conclusion:

It shows a great deal of promise, but, it does have it's limits, and the vast majority of the time you end up generating ten to a hundred images to find one you like, tweaking your prompt and the settings until eventually you get one you like.

 

[Chac1]

This is intriguing and makes me want to check out your AAR all the more.

Also, I should have followed your example and used thumbnails that folks could expand. I wasn't thinking about how these posts might look on mobile (as I don't come to the forums that way).

You are getting great results with Playground, but as you point out, patience is needed and you may wade through a lot of images you don't use. I like to keep many discards because I find ways to repurpose.

I wonder if you use your images that you like as reference images to get more defined outputs? As many of the images I use are portraits, they don't work so well for wider shots of action or other types of images.

Looking forward to your second part.

 

[CBR JGWRR]

This is intriguing and makes me want to check out your AAR all the more.

Thank you. It is 67 pages and counting though, so it may take a while.

Also, I should have followed your example and used thumbnails that folks could expand. I wasn't thinking about how these posts might look on mobile (as I don't come to the forums that way).

On mobile, they fill the available space the same way they do on desktop/laptop/tablet.

You are getting great results with Playground, but as you point out, patience is needed and you may wade through a lot of images you don't use. I like to keep many discards because I find ways to repurpose.

Indeed; in hindsight I should have retained more, but equally, I got through so many it was lagging the software severely having so many displayed.

I wonder if you use your images that you like as reference images to get more defined outputs? As many of the images I use are portraits, they don't work so well for wider shots of action or other types of images.

I've found that it doesn't work as well...

Take "Wolf Naomi 1.PNG" for example. I used that for image-to-image, and what I found was it began treating that little bit of fabric that covers the crotch area as a part of the body, and it began swapping out the golden claws and making them into purely fur, and both obviously don't work.

Likewise, as briefly touched on with the Stellaris Olinbari portrait section, iterating that started turning the head ridges into hair, and replacing the yellow eyes with normal Human-like ones.

Looking forward to your second part.

Well, the second part continues with wider shots and action images.

One possible use considered - until I realised I'd have to do hundreds of final selection images - was header images for individual posts and for the main header image for Life2.0.

Exploring what the software could do with asking it to mimic a Stellaris style, I told it to generate "A Stellaris-style banner". It came up with this:



Which... You can see the resemblance.

This led to thinking of candidates for the main header, intended to depict "A Stellaris-style banner depicting a crashed starship, its passengers of many species examining the uninhabited world they've landed on."

This is eventually what I selected from what it gave me:



Another one that was considered was this:


Ultimately, I decided against it, because the ships didn't feel very Olinbaric, and I ended up doing very little with it once I realised the amount of time it would take to give each post it's own image, plus the unsuitability of some posts for depiction.

Another area that was considered was making example emblems for Leonardo Veres' take on the uniform for Life2.0:


In order, we have the emblems for Daas' small modular reactor program that aimed to deliver 100kW, 1MW, 10MW and 1GW modular nuclear reactors suitable for various applications, the SODS Rocket program emblem, and a medal for participating in a boarding action against MSI, with the red paint displaying that the wearer was in a leadership position for the respective achievement.

They are perhaps a little too detailed.

On the subject of Leonardo Veres, his wife Gillian Veres, a Companionship Asset sold to the docks who took Naomi under her care when Naomi was sold from the high society brothel she had been used in, was bisected in a boarding action by MSI's elite marines attempting to hunt down Naomi, losing both legs above the knees.

Just getting the software to depict a woman in a wheelchair with no legs was a challenge, and adding to that her husband - an admiral in the People's Democratic Republic Of Terra And Colonies - left me with only one candidate image:




The prompt being "An Italian man, early 30's, admiral of a star fleet, in a navy blue uniform.

Double above-knee amputee woman, in a wheelchair.

Futuristic style."

So many images put him in the wheelchair instead...


Overall, there is a lot of potential, but there's also a lot of wasted time.

 

[Chac1]

Overall, there is a lot of potential, but there's also a lot of wasted time.

Yes, I think that sums it up perfectly, @CBR JGWRR . One of the reasons I've cut back on AAR posting (among the IRL issues) is the time spent on creating new art. More time consuming than it may appear. Isn't that ironic as we think about the intended function of AI?

Another area that was considered was making example emblems for Leonardo Veres' take on the uniform for Life2.0:

Love these emblems. Perhaps too detailed for uniform patches but it shows off what is possible.

Just getting the software to depict a woman in a wheelchair with no legs was a challenge, and adding to that her husband - an admiral in the People's Democratic Republic Of Terra And Colonies - left me with only one candidate image:

Great example. Yes, the more detailed the idea, the harder it is to make. Right now, basic ideas are best. One idea per image turns out great. Any more than that and luck enters into the picture.

 

[CBR JGWRR]

Yes, I think that sums it up perfectly, @CBR JGWRR . One of the reasons I've cut back on AAR posting (among the IRL issues) is the time spent on creating new art. More time consuming than it may appear. Isn't that ironic as we think about the intended function of AI?

It's actually a primary subject for one of my narrative ideas I'm considering going for after Life2.0; "Rogue Trophy - A Stellaris Story" looking at an alternative future of Humanity having spent since the 2030s under a Rogue Servitor, when a Human in the 24th century realises Humans haven't done anything for themselves in centuries. One of the critical dynamics is how he or she relates to the AI that does everything for them.

Although, it might actually be a gameplay AAR too thanks to the Machine Age that's coming.

Love these emblems. Perhaps too detailed for uniform patches but it shows off what is possible.

Yes. Obviously the software can't do writing very easily, but it can do these.

Great example. Yes, the more detailed the idea, the harder it is to make. Right now, basic ideas are best. One idea per image turns out great. Any more than that and luck enters into the picture.

Quite. I expect as the technology improves this will get better.

 

[Captured Joe]

One of the reasons I've cut back on AAR posting (among the IRL issues) is the time spent on creating new art. More time consuming than it may appear.

This was also a problem with my first AAR, since I wanted to make a drawing for each chapter. Made it much harder to retain a regular update schedule and, as far as I can tell, did not add significantly to the whole.