The Truth about using AI

Introduction

Much has been written about AI art, how it is theft, how it is cheating, and most of all, how it is too easy and requires no skill. The biggest legal argument is one of permission. You can’t use content to condition your AI models without permission or payment.

The proponents for AI have their own point of view; AI is a powerful new tool for everyone and acts as a great leveler. There is a dirge of ‘AI-slop’ out there, but there is also content and productive data-analysis that would never have been created without AI. This content often removes gatekeepers and unfeasible costs that would otherwise prevent many projects starting. It allows the creator to more forward and shine on the things they are good at.

Both sides have good arguments and are strongly committed to them. This makes the discussion almost take on the mantra of religious fervor.

This article does not present a view. It simply takes the route the ancients took in religion; show, don’t tell. By experiencing real workflows, the reader can judge for themselves how much AI is about cheating and whether it can be a positive tool to augment clearly creative ideas.

This article does not present a view. It simply takes the route the ancients took in religion; show, don’t tell.

Torch-Tarot cards use AI to replace what would traditionally require either stock-photography or photo studio time. The assets were then brought together in Photoshop to create the final card (the author has used Photoshop commercially for 20 years, and been an avid photographer for far longer).

To show the difference between the raw AI images and the final cards, some examples of real workflows are shown below.

For those who are all about copyright, we will touch on that at the end.

Death

Although the initial AI render (below left) looks like an awful mess (a legless horse that looks more like a cow, and a rider that looks like a shapeless scarecrow), the author immediately knew this image set the tone for the required image (below right) in terms of color and atmosphere.

Death and his horse were almost completely digitally repainted using the color palette suggested by the original AI image. New photographic or AI generated sections were added as the piece was built up, then painted over.

Some of the assets used are real objects that were purchased as props and photographed. For example, Death’s scythe is a plastic toy that was added into the image and then overpainted.

The soldiers started life as plastic figures; the author often used the excellent DamToys Drawman-01 artist’s manikin, plus more specific historical plastic soldiers that were photographed, passed through AI, and then added to the scene with Photoshop.

Each soldier in the fore and middle ground had to be individually corrected. AI doesn’t do well generating images containing multiple characters (such as a battlefield) and you end up with one or two foreground characters looking okay but the rest quickly turns to a ‘human mush’.

The same problem occurred with other cards containing multiple characters such as the Ten of Pentacles. In this card we see many people sat at a feast. Just as in the Death card, each character was initially a riot of AI hallucinations (left image). Each final character had to be individually corrected via digital overpainting and comping new assets into the piece. Similarly, the art on the walls was initially nonsensical graphics that looked 'kind of right unless you actually looked at them closely’, which you absolutely don’t want on a Tarot card because you expect people to look at the image closely! Instead, photo references from ancient Greece and Pompeii were used to inform what the wall art should look like.

The completed piece for Death was 3000x3000 pixels. Final touches were added manually in Photoshop via a touch-sensitive stylus. This touch-up includes the subtle red glows in the bottom right corner, and highlights to Death’s armor (to make the image ‘pop’ despite the overall lack of color) The final printed card is the 3000x3000 pixel image reduced and cropped to Tarot print size (approx. 830x1420 pixels).

The Wheel of Fortune

The Wheel of Fortune card was problematic because there was so much symbology to add. Worse, the level of the initial AI images (example shown below, left) was almost comically bad…

The only way to get anywhere near was to

• Sketch out the required pose of the Goddess Tyche (or ‘Lady Luck’), using both Poser (a digital 3D figure system for artists and animators) and manual drawing, then
• Use the result as inputs to image-to-image AI renders, then
• Overpaint/correct it (much like already seen in the Death card), and finally
• Flesh out the symbology by adding the extra elements in Photoshop.

So why use AI at all? Simple. It is not initially faster for the same quality output (and sometimes it can't reach the same quality), but once you realize how it works and what to expect from it, it does become faster.

The process of knowing 'how it works' can take a long time, and includes 'not wasting time attempting anything AI cannot do' and instead doing those things manually or via traditional digital art techniques (such as 3D or photo references). This learning phase took about 6 months for the author via the creation of a 40 card test Oracle deck (unpublished, but a quarter of the Torch-Tarot Major Arcana was derived from the best ideas, so nothing was wasted in the end!).

The additional elements required for The Wheel of Fortune, such as the ship’s rudder and ship’s wheel were created using AI and/or by photographing toy/fantasy models. Note that the AI sections created for this purpose do not replace artwork; they replace the need to use stock photography for the mundane and secondary objects.

AI can give you exactly what you need in this use case. This is the real power of AI and where it absolutely shines in the creative workflow.

Some of the required elements were not available via stock-photography because they were so specialized. They were instead created via either AI or AI and 3D.

For example, there was no suitable statue of the Egyptian God Shai, and neither were there many simple images of the wheel alluded to in Ezekiel 1:4-28 (most were over-the-top esoteric, but the image notes required a basic wooden version).

Having seen some of the separately created assets, you can perhaps now begin to see how the card was put together. Every item was separately created using the best workflow for it, then the whole thing was composited and color corrected together in Photoshop so it looks like a continuous painting.

AI and Hands

Hands are a particular problem for AI, as are bare feet, but the Wheel of Fortune contains both. For the hands (which of course appear everywhere on nearly every card), a manikin hand model was used, specifically the Kotobukiya hand for artists.

This is a pure white plastic hand model that can be used as an image-to-image source so that AI can generate perfect hands, or at least, perfect enough that they can be easily corrected then merged in Photoshop. This is the exact same model the author uses as the go-to reference for traditionally drawn hands.

Finally, as already noted above for the Death card, everything was digitally overpainted and/or color corrected in a 3000x3000 pixel version of the image to make it all come together as a single consistent piece.

The Wheel of Fortune took about three weeks to create (given that it was restarted twice!). It took the longest time to create of all the cards in the deck, with the Empress and Judgement cards coming a close second (because there was so many items of symbology in the specification notes for these cards).

Creating a card from first principles; the Shadow

We have so far gone through the steps in creating the visuals, but what about the pre-planning stages that inform the visuals? What goes into the specification and notes for the images, and how different is the result from the standard 'single prompt' AI image? and most important, how is the intent of the card derived?

Many of the Major Arcana cards were created as a pair, along with their shadow archetype (and even though the shadow is not part of the physical deck). This was done to give a fuller understanding of each card, and allows the deck to be extended, or a companion deck to be created later. We will look at how the shadow for the Wheel of Fortune was fleshed out, and what the thinking was before any image was attempted.

The Wheel of Fortune in this deck is synonymous with the Seeker or Explorer psychological archetype used in personality tests and psychology. This archetype starts to manifest when the child breaks away from the mother and family. The wheel in the Wheel of Fortune card takes its cue from the Seeker archetype (rather than the more esoteric symbology in the standard Tarot deck) and is a ship’s wheel, implying ‘taking the chance to set out on your own and make your own fortune’.

After some research into mythology and how each candidate fits with the target Jungian shadow archetype, it was decided that the shadow seeker archetype was to be illustrated via the Goddess Nemesis.

She forms the balance of the good things Tyche gives us by making us pay if we get more than our fair share, or if we were unworthy to receive the chances we were given. Tyche and Nemesis were often seen together in ancient vases and other decorative art so this makes historical sense.

In mythology, Nemesis carries a balancing-scale in one hand to judge whether fortune has given us too much, and a whip/flail or dagger in the other to punish, and wings to quickly find the guilty.

Thus, Nemesis is our inner shadow that we use to punish ourselves when we consider ourselves unworthy through failure in the chances given to us, or guilt on been given more than we deserved. She is the part of us that makes us feel guilty for our fortune, but also when we blame ourselves for our misfortune. You just can’t win with Nemesis!

Rather than go through the construction of this image, we will end by looking at the common rules that were applied to all images created for Torch-Tarot for things such as intent and accuracy. They are;

• The image should be better than AI images created on the same subject. This means no AI hallucinations, no hiding parts of the body that are difficult for AI, and no blank AI expressions or wooden postures.
• The image should stand out against all non-AI images created in the 10 years before AI. It should have something to say rather than just following the herd.
• It should be true to the mythology of the character depicted. This is crucial for Torch-Tarot because we are using mythological characters as the closest visible version of the shared archetypes that exist in our subconscious.

These rules force any image to look significantly different from the usual AI images found on the web. A quick search on AI images of Nemesis gives this:

The screenshot above was the best section seen in the first five pages of Google results. We will leave the reader to judge how good these images are, but compared to the workflow for this deck, none would be used.

They don’t really say anything about the character of Nemesis. None of them have both a balance-scale and dagger or whip. Worst of all, the framing seems to be designed to hide the parts that AI is bad with, at the expense of good composition.

There’s other common AI issues (the wings are always too small, and look like a child’s fairy costume wings, the characters invariably have beautiful but blank faces, hands are badly rendered when they appear, etc.).

What about traditionally created images?

Looking at non-AI images (by searching for <subject> before:2022 after:2012, which gives us images in the ten years prior to AI), we get results like this:

Technically better, yes, but mythologically, not so much, and none of these images would be used by Torch Tarot.

All these images and photographs are closer to Themis (the Justice card) than Nemesis. Some get a bit closer via a younger character, but this is often the Goddess Dike. Perhaps okay for a creative work, but poison for a Tarot deck that relies on getting the archetypes and their symbology dead right.

This brings up an important point; traditional (non AI) and online creators tend to take inspiration from their peers and popular media such as film, video games and comic strips, so if the current dominant sources are wrong, there is a strong chance that substantial numbers of search results of traditional works from the same period will be wrong.

This is all good for Torch-Tarot, because we have the chance to make something that can stand out by being more correct through independent research of the root source material.

Here’s the final image.

Nemesis finally holds a balance scale and whip as per her mythology. Her home is Hades so she is underground. She has wings that look large enough (and they are not symmetrical, instead matching their opposite legs because that’s what birds do, and incidentally what 4-legged animals do with their fore-hind leg pairs - attention to detail is always crucial!).

She isn’t looking straight ahead. Instead, her pose matches her demeanor. Her head is down as per a fighter sizing an opponent up (she is noted for being worshiped by gladiators), and because of the weighting between the wings and legs, she doesn’t look stiff but instead perfectly balanced, confident and belligerent. She’s daring you to do something wrong in front of her...

The more observant may notice that Nemesis's legs look like they are somewhat based on the Temperance card, except that Nemesis has both feet on solid ground. She cares little about intuition and what your inner mind says in justification of your actions - she is all about moral balance and objective justice. She is firmly rooted on solid ground despite being surrounded by the subterranean lake of the subconscious.

Only when the basics are right, can we add a bit of something more current, which brings us to 'Why is she wearing an olive dress?'.

Because in the Matrix movies, the heroes wear subdued colors such as olives and earthy browns when they are not in the Matrix, and give themselves call-sign names from mythology. Some time was spent creating a new character informed by the world of the Matrix, complete with drab clothes and a modern urban haircut… and the call-sign Nemesis. You see this hidden design reference (or easter egg) better when Nemesis loses her wings:

The dress is also a modern interpretation of the dress worn on one of the few statues we have of Nemesis (Roman copy of an original Greek version as seen in the Kinský Palace, Prague).

Th final image follows the root creative idea of making Nemesis look like the antithesis of Tyche. The Tyche/Wheel of Fortune card takes its visual theme from a romantic renaissance painting, but Nemesis takes her cues from a harder and more cynical subdued cyberpunk look. The ‘adding something new’ part of this image is the ‘true to mythology yet contemporary’ intent of the Nemesis image. It also had to match the Wheel of Fortune card so the two look like they came from the same deck. A tall order, but again, you can judge how close we got!

Finally, what goes on behind the scenes in creating an image like Nemesis? Is it all from a single AI prompt? Hardly.

The background and every part of Nemesis was hand-edited separately then composited and color corrected to make a single whole. Here’s the Photoshop Layers.

Quite a lot of them (but Tyche had about three times more!).

But hey, what about copyright and ‘cheating’?

The rules on copyright are clear when it comes to AI. The moral argument against the use of AI is consistent with the legal one (which does not happen very often!).

Human authorship is essential for copyright. Any work that is created from prompts is not copyrightable because a prompt is seen as a specification for work or an invitation to do work (often called a tender), and not the work itself, irrespective of how much trial and error and research went into it. In normal conversation, the rule would be:

Asking for a house to be built does not make you a house builder.

Any technical fixes (such as correcting hands and fixing AI hallucinations) is not creative work either because a machine could do it eventually via brute force with a human merely accepting/rejecting the result.

Many recent failed AI creator copyright claims argued that the work involved in the accepting/rejecting part constitutes a human input (usually based on a ‘machine cannot do it alone’), but again, this is not a creator doing the work, but judging it for completeness. Sometimes the AI artist also claims the large amount of manual time it all takes as proof of ownership, but that only really proves if you had an infinite number of monkeys typing randomly for an infinite time, you’d eventually have something someone will get legal over.

What could be claimed is not in the output files from AI, but anything else that you actually did yourself to transform (rather than just fix) the assets, including

• The creative positioning and compositing of the individual items of media to make a final piece.
• The creative transformation of items of media to fit a coherent creative vision, irrespective of whether those items of media came from a stock library, your camera, open source images, or AI.
• Evidence of a wider, human created vision and intent that is not within any of the source assets.

This is how most pre-AI digital artwork, web design and graphic design has asserted copyright for years. Many such works include third party photography, fonts, open-source graphics and code, and other similar assets.

Bringing it back to the world of Tarot

Although there are many AI Tarot cards and many non-AI Tarot decks, all good Tarot decks show themselves in the same way. ‘Good’ is not just how good the artwork is, but:

1. How much symbology the creator added to each image in the deck, and how much the deck is unique to the creator rather than just ‘a reskin of Rider-Waite-Smith’ (or 'Marseilles' and the other common historic ones) .
2. The guidebook and whether or not it adds something new (and whether it exists at all, because a lack of one indicates ‘just following the rules of Rider-Waite-Smith’, and the best deck that follows Rider-Waite-Smith is the Rider-Waite-Smith deck so why bother?).
3. The overall consistent world being created by the deck.

And bad decks show themselves in their own way of not following the above.

‘All happy families resemble one another, each unhappy family is unhappy in its own way’. Anna Karenina, Leo Tolstoy

Flicking through the cards and guidebook should take you into a new and carefully crafted graphical world that is full of symbology. The deck should have its own personal slant on Tarot that makes the deck unique. This is often via additional symbology or new way of thinking about the rules of Tarot.

Conclusion

By reviewing the workflows behind Torch-Tarot, the reader is invited to make up their own mind on whether the deck is single prompt AI-swill or an overall workflow that uses all available media types to bring a well-defined creative vision per card to life.

The particular Goddess used above as an example of Torch-Tarot artwork vs totally non-AI artwork is absolutely not symbolic of 'AI-haters as the author’s nemesis'. On the off-chance it was both intentional and an enjoyable diversion, there's a whole deck plus guidebook full of similar 'show don't tell' moments.

Hopefully we can at least agree that this article gives an insight on how Torch-Tarot was put together!