Sign In

Tarot cards, part 3

Loading Images

Here are some more of the famous ones: Death, The Lovers, and The Tower. Plus a remake of The Magician.

See also part 2 and part 1.

Notes

WAI-NSFW-illustrious-SDXL v13.0, with inpainting.

Death should obviously be a goth girl with twintails, because why not. Her sickle needed some positioning work in GIMP, and sketch mode on top of that to fix the shaft. But I like the look the AI came up with for this character.

The Lovers, since I'm the one making these, must obviously be yuri. Here the quality of the txt2img was a shocker. I inpainted much of the image, but only to polish what was already great. Nothing in this shot needed a complete redraw. Maybe this kind of thing is well represented in Wai's training data, lol?

No Forge Couple this time, either, as I was happy with generic, matching characters - which is exactly what a yuri variant of this card needs. I like how the AI picked their hair and dress colors.

The Tower is the first exploration of the theme for this set. Many of us know the iconic look. Following that, I was going for a medieval castle tower struck by lightning, with a nice-looking anime character somewhere in the image, but the AI thought the image needed more towers - some from Istanbul, some from Prague, and perhaps a garden gazebo for good measure.

The aesthetic ended up more Final Fantasy Type-0 than AI anime tarot, but the dramatic lighting works, as does the way the lightning bolt and the surrounding sky highlight the character - given her dark clothing - as almost a negative space in the image. So I thought this shot was worth detailing. But as The Tower, this is not optimal, so I'll probably try again for an alternative variant later.

The Magician I decided to revisit almost before the GPU had cooled. While I like the image quality of Illustrious (in part 2), the design lacks oomph, for lack of a better word. In the original from 2023 (in part 1), AOM3 absolutely nailed both the composition and the use of colors.

So, how would a re-imagine of that look like? I loaded the "The Magician" from part 1 back into SD, and rerolled in img2img with Wai, at 0.75 denoise. The AI hit jackpot on the first attempt. The inpainted result is shown here.

The way Wai draws flames in lanterns and in lit campfires does not make much sense, but that's artistic stylization for you. And it fits the fantasy atmosphere well enough. So I'm letting the AI artist have its way there.

Also the stained glass could be better, but I kept rerolling for a while, trying different denoise levels and amounts of surrounding context, and it didn't improve. I suppose drawing stained glass better than this is beyond the AI's capabilities for now, at least for Wai. I'm declaring this version acceptable. If I ever end up making a deck of these, this shot is going in.

I'm thinking that in the major arcana, once a border or an overlay is added, it would be nice to have numbers only, no text. This both leaves open opportunities for some artistic liberties (e.g. making "The Emperor" actually a tomboy, enabling a full all-female cast), as well as makes for an interesting artistic challenge. Each image itself should stand on its own, making it obvious which card it is - so that a viewer with a passing familiarity with the tarot deck would not need to look it up. I expect some cards to be easier to make stand out like this than others. Very probably not all of them will be successful. We'll see.

Comments