Artwork Messing around with AI, A Art thread.

Scooby Doo

Well-known member
eretzyegern_Brazilian_citizens_combating_against_foreign_invade_de0d6d19-2dd9-42e5-aac8-9905f18a3181.png

eretzyegern_Brazilian_citizens_combating_against_foreign_invade_09e264cb-9d15-4d67-9040-60b0effa8b82.png

addisonclarke333_Brazilian_advanced_guerrilla_soldiers_in_the_j_5aec35fa-ca92-49cd-9107-da31ec50648d.png

addisonclarke333_Brazilian_advanced_guerrilla_soldiers_in_the_j_31b30b66-74f7-4eaa-ab69-8e9bcd97e45a.png

addisonclarke333_Brazilian_advanced_guerrilla_soldiers_in_the_j_d3032d73-4e10-49d3-ab5c-46bf50987626.png

pecaha9638_3d_render_of_ultrarealistic_combat_between_Brazilian_e3f17be7-9366-4bad-8850-83c529fa336e.png

pecaha9638_3d_render_of_ultrarealistic_combat_between_Brazilian_988853c7-dcd3-4fc3-a53d-73e89e63e416.png

pecaha9638_3d_render_of_ultrarealistic_combat_between_Brazilian_456eaf4c-146b-498d-8a59-cb9f9177e031.png

pecaha9638_lovecraftian_soldier_full_body_14d9ce73-7ac0-420f-b43a-b5fa3fb29990.png
Bruh these hands 😂
 

Bear Ribs

Well-known member
So I did this as a complete lark but I found I gained some interesting insights into how the AI worked from deciding to have it draw a gag image and decided to share.

The prompt was: "sailor moon destroying majin buu after some idiot posted that in vs. debates" using the Midjourney checkpoint.

There were no img2img cleanups, this was its first draft, bad hands and all.

f1cnsNu.png


This actually provides some cool windows into how the AI thinks. It completely ignored the vs. debates dig as it clearly has no idea what they are (I actually expected it to do something like put a browser post in the background). So it will seemingly ignore details if it doesn't understand them at all. Apparently didn't know how to draw Majin Buu (This is unsurprising as he's likely to have relatively little fanart so it wasn't trained well on him) but did know that Buu is a Dragonball Z character, as the characters are posing in typical DBZ fashion, appear to be gathering energy in their fists, and the hair is closer to DBZ's style than Sailor Moon's. In the first picture Usagi's ear has also turned into one of the very typical Akira Toriyama alien ears and gave her his typical O-ring mouth. Rather than separate them by character it appears to have decided to blend the typical Sailor Moon clothes with the styles used in DBZ. As an aside, DBZ's wide-legged horse stances with the crotch thrust forward are creepy when applied to a little girl.

I used the extremely complex standard negative prompt for Hentai Diffusion even though I don't use that Checkpoint, as it's much more detailed than any others I've found.

The AI understood that I wanted two characters but didn't grok that I wanted them to fight. This is likely my choice of the word "destroying" and it maybe would have worked better if I'd used battling, fighting, or another keyword there.

Even though there shouldn't be any real connection, it decided to put the moon in the background in 3 of the 4 images. I'm not sure if it's simply common for people to draw the moon in the background of Sailor Moon fanart, so it learned it there, or if it decided that the word "moon" in Sailor Moon was significant enough to use it for more than the character.

Overall I feel like this spells out one of the classic issues AI is having right now, it's bad at putting multiple people in the same picture. These are actually fairly decent for first drafts and could probably be cleaned up to something useful in img2img and inpaint with a few more renditions, and a better keyword selection might solve the problem of them posing instead of fighting as asked for. However, the fact that it blended two disparate clothing/art/anatomy styles rather than using each one on the appropriate character is a major hurdle and I have no idea how you'd word a prompt to get it to knock that off and understand that the two are to be kept separate.
 

Scooby Doo

Well-known member
I know art ai thread but I tried it and can't quite get something specific I want.


Does anyone have any references to an artist I can commission?
 

Bear Ribs

Well-known member
I know art ai thread but I tried it and can't quite get something specific I want.


Does anyone have any references to an artist I can commission?
Depends on what you're looking for. I know @Culsu takes commissions. I do pieces pro bono for people sometimes but only if they're writing a story I'm enjoying reading, or otherwise producing content, as a way of saying thanks.
 

Scooby Doo

Well-known member
Depends on what you're looking for. I know @Culsu takes commissions. I do pieces pro bono for people sometimes but only if they're writing a story I'm enjoying reading, or otherwise producing content, as a way of saying thanks.
I'm looking for someone to 2-D cover art for a original series fanfic I'm working on.

I know exactly the way I want it.
download.jpg



Sort of in this style but different content
 

Yinko

Well-known member
FWCarto_a_dramatic_shot_of_battalion_of_Republic_clone_troopers_dc2bd64b-18e9-44d6-b15e-f12311accb0e.png


FWCarto_a_dramatic_shot_of_battalion_of_Republic_clone_troopers_b94ad9b8-00ac-4d40-8701-e799619e56d4.png


FWCarto_A_dramatic_shot_of_a_cyborg_female_Sith_her_lightsaber__68d3861d-d4cf-4b71-b976-b4189516478f.png


A couple SW themed generations.
So, with Chat GPT I know that you can instruct it to change certain parts of a text, can you easily do that with the image AI? Most of those restrict generations per account last I checked so I haven't gotten too into it. The lady at the bottom image has her forearm the wrong way around, so that would be a situation where situational editing would be ideal.
 

Culsu

Agent of the Central Plasma
Founder
So, with Chat GPT I know that you can instruct it to change certain parts of a text, can you easily do that with the image AI? Most of those restrict generations per account last I checked so I haven't gotten too into it. The lady at the bottom image has her forearm the wrong way around, so that would be a situation where situational editing would be ideal.
No, you can't do that. MJ and other generators have a hard time grasping the *general* idea of what your prompt means, and leave little to no possibilty to usefully tweak specifics. And in general limbs, and hands in particular, are a complete mystery to AI.
 

Bear Ribs

Well-known member
So, with Chat GPT I know that you can instruct it to change certain parts of a text, can you easily do that with the image AI? Most of those restrict generations per account last I checked so I haven't gotten too into it. The lady at the bottom image has her forearm the wrong way around, so that would be a situation where situational editing would be ideal.
Yes, that's called InPaint. In Stable Diffusion, you can move an image to InPaint, put a mask over her forearm, and then tell it to redraw only the masked section and leave the rest alone. This is commonly done when the image is mostly good except for its eldritch horror hands or screwed-up face since those are what it fails at the most. You don't usually need to change the prompt so it will keep the same art style and usually make it seamless.

It won't necessarily figure out how to shape her arm properly the first time and will probably need several redrawings before it randomly aims her arm the way you want it, and in that specific case also figures out to give her five fingers instead of what... I count about eight fingers on that hand?
 

Bear Ribs

Well-known member
Okay, this is kinda interesting and a bit glitchy.

My prompt was
comic book style, lineart, black and white, no shading, snowy steppes

I've been using this prompt with minor variations (dark forest, wetland, badlands canyon, sandy dunes, ancient ruins, etc.) for quite some time trying to reduce how much Midjourney shades stuff and trying to get it to churn out lineart similar to my own to use as backgrounds for my drawings.

For whatever reason, it seems to always draw characters when I ask for snowy steppes but has never drawn figures in any other landscape I've tried to produce.

8hFDNFK.png

DhchyqE.png

Deh5vrE.png

t8qibc5.png


This bugs me because it's inconsistent and I don't understand why. I've been doing AI art for a big two weeks and I'm mostly feeling out how prompts work and trying to build a general idea of its logic and thinking, rather than expecting any good art at this stage. And mostly it's going well, but this is anomalous compared to the other results I've gotten. I don't understand why it would think power armor is specific to snowy steppes instead of swamps, forests, or the like. Something here is weird and I don't know what. Why are snowy steppes different?

Also, I can't get it to produce ruins, even with that prompt they're almost always in color and always, always shaded instead of lineart. I've made at least 200 ruins images and they're all like that.
 

Yinko

Well-known member
For whatever reason, it seems to always draw characters when I ask for snowy steppes but has never drawn figures in any other landscape I've tried to produce.
Have you tried using terms like "landscape" or "windswept"? Keywords that imply a lack of people or buildings.
 

TheRejectionist

TheRejectionist
@Culsu @Bear Ribs @Scooby Doo this is what I recommend to get MUCH MORE highly detailed stuff

prompt /imagine (insert what you to imagine here) followed by Photography, hyper realistic, photorealistic, Studio Lighting, reflections, Dynamic Pose, Cinematic, Color Grading, Shot on 50mm lens, extremely detailed, beautifully color-coded, intricate details, beautifully color graded, Unreal Engine, Color Grading, Tilt Blur, Shutter Speed 1/1000, White Balance, 32k, Super-Resolution, Megapixel, ProPhoto RGB, VR, Massive, Halfrear Lighting, Backlight, Natural Lighting, Optical Fiber, Cinematic Lighting, Volumetric, Contre-Jour, Beautiful Lighting, Accent Lighting, Global Illumination, Screen Space Global Illumination, Ray Tracing Global Illumination, Optics, Scattering, Glowing, Shadows, Rough, Shimmering, Ray Tracing Reflections, Lumen Reflections, Screen Space Reflections, Diffraction Grading, Chromatic Aberration, GB Displacement, Scan Lines, Ray Traced, Ray Tracing Ambient Occlusion, Anti-Aliasing, FKAA, TXAA, RTX, SSAO, Shaders, OpenGL-Shaders, GLSL-Shaders, Post Processing, Post-Production, Cel Shading, Tone Mapping, CGI, VFX, SFX, insanely detailed and intricate, hypermaximalist, elegant, hyper realistic, super detailed, dynamic pose, photography, 8k --v 4

Could I ask you guys to maybe use your AI programs for a few things of mine ? I have exhausted and can't find any other Gmail generator that can bypass Discord controls to do on my on as of now.
 

Yinko

Well-known member
The faces are pretty much all fantastic. The fingers are terrible, but I also noticed that clothing seems don't line up and machinery like guns don't make any sense.
 

Users who are viewing this thread

Top