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

[Kyriakos]

Regarding style-copying, it also depends on how believable it is, even as a digital image. Some styles may be easier for AI to copy - iirc even the free Dall-E can decently emulate Paul Klee, for example, though often it is glaringly not something by Paul Klee. It has less success with Munch, in my experience. And I doubt it would be consistently believable with something like Caravaggio.
Then again, at times the AI painter will just have legs/feet/other appearing in random places, obviously because it doesn't actually have an understanding of what those are and tries to account for them mathematically. Which is not helped if part of its input was, say, Picasso

 

[Caesar of Bread]

Adding the images I was referring to, because I can't find the original post:

View attachment 658951
View attachment 658952View attachment 658953View attachment 658954View attachment 658955View attachment 658956View attachment 658957View attachment 658958

Just wondering, on the original post (Page 3) and this new one, who are these leaders?

 

[Bonyduck Campersang]

Just wondering, on the original post (Page 3) and this new one, who are these leaders?

Page 3: Genghis Khan, Catherine of Russia, Cleopatra, Catherine of Russia (again)

This page: Ivan the Terrible, Elizabeth I, Selim I, Nero, Bismarck, Julius Caesar, Napoleon

 

[Thorgalaeg]

I think it would be interesting to post the method used to obtain the images, If it was Midjourney, Dalle-E, Stable diffusion in some of his many flavours etc, the prompt, and wathever other details.

This one was generated using A1111 webui, with Deliberate 2.0 model (Stable Diffusion 1.5), starting from a quick sketch made with my mouse (a few lines), and controlnet in guess mode, plus some img2img work and clip interrogation (it is long to explain). Initial prompt: "tree woman"

 

[tjs282]

Might also be interesting to see your original sketch, if you wouldn't mind uoloading it?

 

[Thorgalaeg]

It was not saved as i drew it in the tiny controlnet window included in the A1111 interface. But it was something like this, only worse and at even lower res:

 

[Gedemon]

Cyber Easter Bunnies

Spoiler Prompt / Parameters / Model :

Raw photo, best quality, masterpiece, (cyberpunk:1.2) very cute Easter bunny, cybernetic implant, slim, bokeh, (bladerunner:1.1), 8k uhd, soft lighting, high quality, film grain, Irish Russian, (annoyed, frown:1.2), (high detailed skin:1.2), ultrarealistic uhd faces, detailed skin texture

Negative prompt: (bad art), (deformed),(mutated), lowres, bad anatomy, text, error, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, artist name, blurred, blur, monochrome, blurry, artist name, blurred, blur, logo, text, letterbox, (bad anatomy), (disfigured), (long neck), (poo art), (tiling), (bad art), (deformed),(mutated), chubby, (yellow teeth), chubby, fat

Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 881386312, Size: 960x640, Model hash: c0d1994c73, Model: RealisticVision20, Denoising strength: 0.5, Hires resize: 0x1440, Hires steps: 16, Hires upscaler: R-ESRGAN 4x

 

[Thorgalaeg]

An experiment. Variations about something i painted many years ago. Original oil on canvas:

first one is an interesting one with grapes instead of handles (the AI had a hard time understanding the heavily ornamented silver handlings), second one is kind of Van Gogh style, third one
Picasso (or Pixcasso because it looks like pixelated), the fourth one is a more photographic style, curiously here the machine decided to add sliced lemons and some grapes (AI loves grapes apparently),
which btw is a great idea i should have included it in the original painting.

Spoiler :

It was an img2img thing of course, using RelisticVisionV20 (SD 1.5) model in A1111, without using controlnet. Added the prompt:
"a picture of a silver bowl with lemons, oranges and bananas, on a white tablecloth, glossy shiny reflective, white silver, color bleeding, strong sunlight, specular radiance" for more clarity, plus"art by Pablo Picasso", etc as desired.

 

[Kyriakos]

Cyber Easter Bunnies

Spoiler Prompt / Parameters / Model :

Raw photo, best quality, masterpiece, (cyberpunk:1.2) very cute Easter bunny, cybernetic implant, slim, bokeh, (bladerunner:1.1), 8k uhd, soft lighting, high quality, film grain, Irish Russian, (annoyed, frown:1.2), (high detailed skin:1.2), ultrarealistic uhd faces, detailed skin texture

Negative prompt: (bad art), (deformed),(mutated), lowres, bad anatomy, text, error, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, artist name, blurred, blur, monochrome, blurry, artist name, blurred, blur, logo, text, letterbox, (bad anatomy), (disfigured), (long neck), (poo art), (tiling), (bad art), (deformed),(mutated), chubby, (yellow teeth), chubby, fat

Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 881386312, Size: 960x640, Model hash: c0d1994c73, Model: RealisticVision20, Denoising strength: 0.5, Hires resize: 0x1440, Hires steps: 16, Hires upscaler: R-ESRGAN 4x

Now that is seriously cool.

 

[Samson]

Sony World Photography Award 2023: Winner refuses award after revealing AI creation

 

[Thorgalaeg]

Sony World Photography Award 2023: Winner refuses award after revealing AI creation

For anyone familiar with AI image generation it is obviously AI generated. Mostly because the misplaced and deformed hands. As i understand it, judges knew AI was involved but decided to give him the prize anyway which the photographer refused.

 

[Gedemon]

A few wallpapers I've generated recently (2560x1440)

Prompt should be attached to the full resolution image.

 

[Bonyduck Campersang]

A few wallpapers I've generated recently (2560x1440)

Prompt should be attached to the full resolution image.

Were these made with Stable Diffusion?

 

[EgonSpengler]

This one makes me think of an old Dungeons & Dragons adventure called Ravenloft. People who run tabletop RPGs must be dancing a jig about these new tools available. I was okay with pencil sketches back in the day, but I never would have taken the time to create an illustration of this quality, even if I had the skill.

 

[Gedemon]

Were these made with Stable Diffusion?

Yes, I've been testing a few of the available Stable Diffusion models that are kind of specialized in fantasy image, my favorite so far being "A-Zovya RPG Artist Tools (v2 offset)"

 

[Thorgalaeg]

@Gedemon, the one with the burning ship is awesome. It is interesting as AI turned fire into red leafs in the last one.


BTW, drawn this in Rebelle to test my brand new Huion Canvas 4K tablet (which is fantastic btw, almost as good as any Wacom):

As you can see it is a bedouin woman or whatever, something medieval, kind of realistic thing.

Then sent It to A1111 to see what AI could make of it:

Of course the AI turned it into a very evocative sci-fi dune-esque scene. Most SD models seem to love Fantasy and Sci-fi more than anything else. Clothes, face and everything look gorgeous now (even if there was a slight ethnic change), but had to retouch feet and legs myself as AI seems totally unable to get it right even having them just in front of its figurative face, specially when it tries more complicated poses as these crossed legs under the skirt, which apparently blow AI's mind. (May be a bit my fault too as the original skirt is kind of messed up)
Finally added the horrible alien steampunk pet to balance the composition and becuase she looked so lonely.

Spoiler :

models used: mostly RealisticVision2.0 and a bit of A-ZovyaRPG2 (both based on SD 1.5)
Prompt to help the img2img process: "a woman sitting in the desert looking at the horizon, with a weird white alien pet next to her, desert citadel at the background, dunes, intricate, sharp focus, highly detailed, concept art, illustration, volumetric lighting, intricate, hyper detailed, artstation, cgsociety, desolation, windy, two legs, barefooted"
neg prompt:"(deformed, distorted, disfigured:1.3), poorly drawn, bad anatomy, wrong anatomy, extra limb, missing limb, floating limbs, (mutated hands and fingers:1.4), disconnected limbs, mutation, mutated, ugly, disgusting, blurry, amputation, ships"

 

[Samson]

Slumdog billionaires

Spoiler More :

 

[Thorgalaeg]

Spoiler :

A woman walking on the beach on an extraterrestrial planet, the sky of violet color, You see several planets in the sky, High quality image, High definition, Spectacular, Desolation, Tranquility,Immensity of the sea,Wind weather.Cinematic, dramatic, dramatic lighting, cinematic, artstation, digital painting, concept art, smooth, sharp focus, illustration, art by Krenz Cushart and Artem Demura and alphonse mucha, UHD, trending on artstation, octane render, HDR, HDR, 8k, 16k, intricate details. Realistic, Hyperrealistic, Refined, Refined, Highly Detailed, Highly Detailed, Highly Detailed, Highly Detailed, Cinematic Lighting, 8k, HD, beautiful lighting, HD, beautiful scenery, masterpiece, artstation, award-winning-winning. Realistic, Hyperdetailed, Hyperrealistic, Refined, Highly Detailed, Highly Detailed, HDR, Refined, Highly Detailed, 4k, HD, beautiful lighting, Beautiful, 8k, HD, beautiful scenery, detailed, epic, artstation, masterpiece, art by artstation, award-deco, trending on deviantart, octane, masterpiece, masterpiece, digital painting, trending on artstation, masterpiece, octane, artstation HD, vibrant color
Negative prompt: (deformed, distorted, disfigured:1.3), poorly drawn, bad anatomy, wrong anatomy, extra limb, missing limb, floating limbs, (mutated hands and fingers:1.4), disconnected limbs, mutation, mutated, ugly, disgusting, blurry, amputation
Steps: 20, Sampler: Euler a, CFG scale: 7, Seed: 218967473, Size: 768x840, Model hash: 4d91c4c217, Model: +used_lyriel_v15, Denoising strength: 0.5, Noise multiplier: 1.15

 

[Birdjaguar]

Where, how do the negative prompts show up?

 

[Thorgalaeg]

Where, how do the negative prompts show up?

I don't understand the question.