Cool Pictures from the Mind of a Machine: AI Generated Pictures

I guess it's an acquired taste. The more noisy nervous bad paintings you look at, the more you appreciate the occasional good painting that attracts your attention and allows your eye to linger on familiar and coherent objects instead of getting jagged at every end.

I've spent a significant amount of today looking at abstract or abstract-adjacent artworks (on my phone, not at a museum, I still believe Sadequain a terrible painter, though now I admit I might revise my opinion were I to be subjected to his works next time) and I've somehow got myself to actually like some abstract paintings, though still maintaining a distant attitude to the rest of the artist's oeuvre. It was a long time coming anyway. Some time ago I saw a Hilda af Klint painting that stuck with me, in spite of my almost violent feelings towards abstract art; I am slowly succumbing to the fatal allure.
It is a question whether ai really can mimic the spirit of abstract art, when it clearly would have no idea of how it differs on a deeper level from non-abstract.
I had asked some programs to produce stuff in the style of Paul Klee, but I could always tell the style is very different - they'd not pass as Klee originals.
 
Can you use this art n commercial product?

I understand there's copyright issues but if it's an original AI creation....
 
This afternoon was reading Dante´s Hell illustrated by Gustavo Dore, so i tried to make something similar with Stable Diffusion, sensual and biblical simultaneously, of course it finished looking more like a huge brothel, but IMO it is pretty amazing anyway:

00096-2457881375.jpg


Original PNG is a huge 8000x6000 pic or something like that. Some detail:

det.jpg


It is pointless to write any prompt here, since making images like this is more complex than that, only remarkable thing to say i included "art by Pieter Bruegel the Elder and Hieronymus Bosch", used several different models, some LORAs and a lot of negative embeddings.
 
All the women took very similar. Their eyes and mouths often are very close. Otherwise, very nice.
 
All the women took very similar. Their eyes and mouths often are very close. Otherwise, very nice.
Yes. AI tends the clone the characters in an image. Add that every model seems to have a preferred kind of woman. Obviously every author overtrained his model with his own female preferences.
(It is a very 'human' behavior anyway, if you look at any big composition by Rubens, Tiziano or any other old master you will notice every woman looks similar )

A way to avoid that is inpainting every face using a different model. I did it with some faces that were too similar and to add some variety but didn't have the patience to do it with every face. Also as soon as you run the final general pass to make the whole picture more cohesive, the likeness appears again.
 
Last edited:
This afternoon was reading Dante´s Hell illustrated by Gustavo Dore, so i tried to make something similar with Stable Diffusion, sensual and biblical simultaneously, of course it finished looking more like a huge brothel, but IMO it is pretty amazing anyway:

View attachment 664788

Original PNG is a huge 8000x6000 pic or something like that. Some detail:

View attachment 664790

It is pointless to write any prompt here, since making images like this is more complex than that, only remarkable thing to say i included "art by Pieter Bruegel the Elder and Hieronymus Bosch", used several different models, some LORAs and a lot of negative embeddings.
Impressive :)
Not really artistic, though; has more of a 3d game feel than a classic painting.
Which model was used for the main female?
 
Impressive :)
Not really artistic, though; has more of a 3d game feel than a classic painting.
Which model was used for the main female?
Couldnt say exactly. Used a bunch of models successively. Main ones were reliberate_v10, dreamshaper_631, rmadartSD15, aZovyaPhotoreal_v1 and some others, all downloaded from civitai.com. The one used for the final pass was rmadartSD15 becuase it seems to give a more "imperfect" less 3Dish feeling precisely.

Anyway I first rendered other more ´classical´versions like this one but it wasnt that i was looking for:

00071-3013536342.jpg




This one was not so heavily inpainted and retouched as the final one, in fact it was pretty inmediate. The prompt and other info was:
an old painting of {a seminaked winged angelical woman} standing on top of a pedestal in hell, reigning over a multitude of doomed (((men))) and women, metatron, standing triumphant and proud with her right hand up and fist closed, covered with some transparent clothes, angered, conquest, apocalyptic hellish world, (((rocky mountains at the background ))) Romanticism <lora:detailmaker:1>
Negative prompt: easynegative, 16-token-negative-deliberate-neg, 7dirtywords, an5, anime-style-negative-embedding, bad-picture-chill-75v, By bad artist -neg, DreamArtistBADHAND, DreamArtistBADHAND-neg, neg_grapefruit, NaturalHand2-3500-neg, negative_hand-neg, negativeembed, NG_DeepNegative_V1_75T, nncursedV0-neg, realisticvision-negative-embedding, rmadanegative402_sd15-neg, rmadanegative4_sd15-neg, sd1-failed-oekaki, Style-Empire-neg Unspeakable-Horrors-Composition-4v, verybadimagenegative_v1.3, ((wings at background))
Steps: 20, Sampler: Euler a, CFG scale: 5.5, Seed: 3660902329, Size: 1200x1800, Model hash: 980cb713af, Model: reliberate_v10, Denoising strength: 0.21, Clip skip: 2, ENSD: 31337, Lora hashes: "detailmaker: e1b1a08b43b5", Noise multiplier: 1.1, Version: v1.3.2

And the starting point for all the process was this quick sketch i made in Rebelle (kind of photoshop but more painting oriented):

ghhhj.jpg
 
Impressive :)
Not really artistic, though; has more of a 3d game feel than a classic painting.
Which model was used for the main female?
hard disagree
 
This afternoon was reading Dante´s Hell illustrated by Gustavo Dore, so i tried to make something similar with Stable Diffusion, sensual and biblical simultaneously, of course it finished looking more like a huge brothel, but IMO it is pretty amazing anyway:

View attachment 664788

Original PNG is a huge 8000x6000 pic or something like that. Some detail:

View attachment 664790

It is pointless to write any prompt here, since making images like this is more complex than that, only remarkable thing to say i included "art by Pieter Bruegel the Elder and Hieronymus Bosch", used several different models, some LORAs and a lot of negative embeddings.
It's impressive considering the work that went into it, and as an example of what AI image-generating techniques can do, but the end result in of itself really isn't. Not helped by the fact that almost every woman has the tiresome 'parted-lips' expression.
 
That wing has a hand growing out of it ^^ And various other unintended distortions.
Obviously. AI needs help to get things exactly as we want, specially when there are complicated things as human's anatomy involved.
 
It's impressive considering the work that went into it, and as an example of what AI image-generating techniques can do, but the end result in of itself really isn't. Not helped by the fact that almost every woman has the tiresome 'parted-lips' expression.
In fact it was not much work. Couple of hours at most, and most of the time rendering again the parts that were wrong. Feet for instance were unfixable. SD1.5 is hopeless with feet. Finally had to do it myself.
 
This afternoon was reading Dante´s Hell illustrated by Gustavo Dore, so i tried to make something similar with Stable Diffusion, sensual and biblical simultaneously, of course it finished looking more like a huge brothel, but IMO it is pretty amazing anyway:

View attachment 664788

Original PNG is a huge 8000x6000 pic or something like that. Some detail:

View attachment 664790

It is pointless to write any prompt here, since making images like this is more complex than that, only remarkable thing to say i included "art by Pieter Bruegel the Elder and Hieronymus Bosch", used several different models, some LORAs and a lot of negative embeddings.

that's truly a lots of work, working at that resolution, and so many people without many anatomical horrors...

I'll try to generate a few pictures based on your sketch with different styles :D
 
Last edited:
Finger toes!
 
Serie based on @Thorgalaeg sketch (thanks) , with different models for quite different styles, using ControlNet and Regional Prompter.

As I didn't plan to edit them, I've directly generated them at a high resolution so that the faces were visible without fix/inpainting (and even then, some model struggle on faces when it's not a portrait)

Those are the few best of many generated pictures, there was a lot of fails.

(click for full resolution @1600x2400)

DisneyPixarCartoon v1.0



MajicMixHorror v1.0




AbsoluteReality v1.0

 
Serie based on @Thorgalaeg sketch (thanks) , with different models for quite different styles, using ControlNet and Regional Prompter.

As I didn't plan to edit them, I've directly generated them at a high resolution so that the faces were visible without fix/inpainting (and even then, some model struggle on faces when it's not a portrait)

Those are the few best of many generated pictures, there was a lot of fails.

(click for full resolution @1600x2400)

DisneyPixarCartoon v1.0



MajicMixHorror v1.0




AbsoluteReality v1.0

Fantastic! I like the second one the most, with all that grinded meat looks much more hellish than mine. How did you get the feet so perfect? Did you use controlnet in pose mode maybe?
 
Fantastic! I like the second one the most, with all that grinded meat looks much more hellish than mine. How did you get the feet so perfect? Did you use controlnet in pose mode maybe?

brute force (aka multiple generations)

could not use ControlNet in pose mode with your sketch, used Depth and Canny, but Depth was particularly insistant on making hands gripping the pedestal instead of feet while Canny was struggling with the group of people.

(edit: and stylized models were better for the feet, for the more realistic model, I gave up and asked for boots)
 
For some reason I'm finding the Pixar one quite disturbing. :lol: I like her minions the most in the one where she looks the most Angelic, in all-white.
 
Top Bottom