|
But surely there will be less people willing to pay for commissions of Charizard with fat titties when you can just type "Charizard with fat titties" into stable diffusion and get all the charizards you could ever want.
|
#
?
Sep 6, 2022 22:27
|
|
|
|
| # ? May 29, 2024 20:09 |
|
|
my understanding of the furry market specifically is that it's such a rich vein of commission money because nobody ever just wants charizard with fat titties, there are like fifty extremely specific things fat titty charizard has to be doing or it's not sexy anymore just repulsive, a mockery
A Wizard of Goatse fucked around with this message at 22:36 on Sep 6, 2022 |
|
#
?
Sep 6, 2022 22:32
|
|
|
Rahu posted:But surely there will be less people willing to pay for commissions of Charizard with fat titties when you can just type "Charizard with fat titties" into stable diffusion and get all the charizards you could ever want. try it. run a batch of 25 with just that spec. do any of the individual pictures, in your scholarly opinionn, "work" as a finished product for the imagined target audience? I think at least for the near future that producing "art" worth anything much with this tool involves a multi-step iteration and curation process with many, many generations of trying out prompts, tweaking weights, finding viable candidates. Making a batch of Img2img alternates with variable CFG scale. Re-gurgitating them several times through masking, copying and pasting, inpainting, clone stamping, touch up, color correction, upsizing. fixing the upsizing. This can easily take hours of work even with a fat 3090.
|
|
#
?
Sep 6, 2022 22:36
|
|
|
Rutibex posted:why? who cares about humans if it looks good I guess this is a step up from you directly plagiarizing entire games and trying to pass the design off as your own or sharing the files.
|
|
#
?
Sep 6, 2022 22:45
|
|
|
I've commissioned some fan art and Stable Diffusion is not even close to filling that role, mostly because of A Wizard of Goatse posted:I don't think I've ever seen a commissioned work over like $20 where you just get a finished product, no input. Most of the appeal now is in getting to personalize it and fuss over whatever dumb details nobody but you cares about. And if SD doesn't know about the character you want, you're hosed. You can try to to get close by listing characteristics (like, red hair, green dress, etc) but chances are the result won't be instantly recognizable which defeats the whole purpose of fan art.
|
|
#
?
Sep 6, 2022 22:52
|
|
|
Bottom Liner posted:I guess this is a step up from you directly plagiarizing entire games and trying to pass the design off as your own or sharing the files. this is thing you have about me is really weird bottem liner, please keep your opinions about me to yourself
|
|
#
?
Sep 6, 2022 22:55
|
|
|
I mean at this moment it's also practically impossible to teach the AI that <this> character is called Gary so that it can, for instance, reproduce the same Gary over several panels.
|
|
#
?
Sep 6, 2022 23:01
|
|
|
Lord Stimperor posted:I mean at this moment it's also practically impossible to teach the AI that <this> character is called Gary so that it can, for instance, reproduce the same Gary over several panels. Textual Inversion is supposed to help with that but I haven't seen a single good example. 
|
|
#
?
Sep 6, 2022 23:03
|
|
|
Boba Pearl posted:A.) That's not how that works, and B.) they'll never be able to enforce that policy other than keeping the bad AI art off, which like, Ok. Well, they did it in Dune.
|
|
#
?
Sep 6, 2022 23:10
|
|
|
Good Sphere posted:Well, they did it in Dune.  Warhammer 40K also, but that aspect was just a ripoff of Dune.
|
|
#
?
Sep 6, 2022 23:13
|
|
|
frumpykvetchbot posted:try it. run a batch of 25 with just that spec. Boba Pearl fucked around with this message at 23:24 on Sep 6, 2022 |
|
#
?
Sep 6, 2022 23:18
|
|
|
lunar detritus posted:Textual Inversion is supposed to help with that but I haven't seen a single good example. Yeah I presume at some point it'll be possible to encode some kind of "DNA" of something the AI generates. But I also guess the companies would charge serious money to use such a feature.
|
|
#
?
Sep 6, 2022 23:26
|
|
|
Bottom Liner posted:I guess this is a step up from you directly plagiarizing entire games and trying to pass the design off as your own or sharing the files. it's a really obnoxious statement being put forth by the most loathsome tech drones on the planet right now but he's not totally wrong. the people paying for art mostly don't give a poo poo about the humanity of the person creating it, and if that's the product you're expecting artists to sell you're really asking them to be influencers, not artists. If AI art generators are allowed to progress to the point where the end results are really comparable, a lot of artists really are going to be just straight up replaced Right now nobody really has to worry too much about AI art 'looking good' except when graded on a really strong curve, though, much like GPT3 gibberish might be good enough to replace clickbait listicles in search results but nobody's reading a novel of it. The main problem for artists right now, same as for anyone trying to use fuckin Google anymore, is that every means of finding anything good online is being choked by high-volume algorithmic garbage generators as a few motivated crap salesmen try to brute-force their way into relevance. Boba Pearl posted:
the ones with the neck boobs might have potential but beyond that I'm afraid no systems analysts with extremely niche perversions are going to be making you rich today A Wizard of Goatse fucked around with this message at 23:34 on Sep 6, 2022 |
|
#
?
Sep 6, 2022 23:30
|
|
|
I don't want to win an argument this badly.
|
|
#
?
Sep 6, 2022 23:34
|
|
|
Lord Stimperor posted:Yeah I presume at some point it'll be possible to encode some kind of "DNA" of something the AI generates. But I also guess the companies would charge serious money to use such a feature. When it's open source it's free
|
|
#
?
Sep 6, 2022 23:34
|
|
|
I mean Stable Diffusion and others already do invisible watermarking on the images that you can run detectors on to tell they were AI generated? First thing I ripped out of my local copy when they released the weights, but still.
|
|
#
?
Sep 6, 2022 23:42
|
|
|
Dall-E and Midjourney have really different abilities. here is a comparison using wizards, same prompt each time. Dall-E is on the left, Midjourney in on the right A Wizard By Wassily Kandinsky   A Wizard By Caravaggio   A Wizard By Erol Otus   A Wizard By Jack Kirby   A Wizard By Jackson Pollock   A Wizard By Edvard Munch   A Wizard By Andy Warhol  
|
|
#
?
Sep 6, 2022 23:50
|
|
|
the only way we're gonna be able to tell AI images in the future is with more advanced AIs that reverse engineer the images. The AI fake furry drawing wars will spill over into real wars.
|
|
#
?
Sep 6, 2022 23:53
|
|
|
 close up, a beautiful powerful sorceress wearing a feather cloak and casting magical prismatic glowing geometric runes, inky swirly mist, prismatic color fireballs, irridecent phoenix, art by Magali Villeneuve and artgerm and greg rutkowski and Scott Campbell tried to hit as many power words for high fantasy art as possible
|
|
#
?
Sep 6, 2022 23:53
|
|
|
Stable diffusion versions.Rutibex posted:Dall-E and Midjourney have really different abilities. here is a comparison using wizards, same prompt each time.  Rutibex posted:A Wizard By Caravaggio  Rutibex posted:A Wizard By Erol Otus  Rutibex posted:A Wizard By Jack Kirby  Rutibex posted:A Wizard By Jackson Pollock  Rutibex posted:A Wizard By Edvard Munch  Rutibex posted:A Wizard By Andy Warhol  Boba Pearl fucked around with this message at 00:02 on Sep 7, 2022 |
|
#
?
Sep 6, 2022 23:59
|
|
|
lol this was Dall-e 2s best A Wizard by Junji Ito  
Rutibex fucked around with this message at 00:02 on Sep 7, 2022 |
|
#
?
Sep 7, 2022 00:00
|
|
|
Rutibex posted:lol this was Dall-e 2s best I swear I'm not nitpicking but did it know you meant Junji Ito or was that a typo in your post? Those are really accurate
|
|
#
?
Sep 7, 2022 00:01
|
|
|
Boba Pearl posted:Stable diffusion versions.  ohhh thats the best jack kirby I've seen. I guess I need a third art AI lol
|
|
#
?
Sep 7, 2022 00:01
|
|
|
TheWorldsaStage posted:I swear I'm not nitpicking but did it know you meant Junji Ito or was that a typo in your post? Those are really accurate  they typo was in my post not the prompts
|
|
#
?
Sep 7, 2022 00:03
|
|
|
Rutibex posted:lol this was Dall-e 2s best 
|
|
#
?
Sep 7, 2022 00:04
|
|
|
how do you tell stable diffusion to only edit a certain part of an image?
|
|
#
?
Sep 7, 2022 00:05
|
|
|
e: I was mistaken.
frumpykvetchbot fucked around with this message at 01:20 on Sep 7, 2022 |
|
#
?
Sep 7, 2022 00:09
|
|
|
Mozi posted:how do you tell stable diffusion to only edit a certain part of an image? That's inpainting, you need to use the inpainting script and make a mask, or use the krita plugin.
|
|
#
?
Sep 7, 2022 00:14
|
|
|
|
|
#
?
Sep 7, 2022 00:19
|
|
|
A Wizard by Juni Ito, colored by lisa frank Steps: 75, Sampler: Heun, CFG scale: 12, Seed: 2273524854 
|
|
#
?
Sep 7, 2022 00:32
|
|
|
Boba Pearl posted:That's inpainting, you need to use the inpainting script and make a mask, or use the krita plugin.  ... well, i'm sure i'll figure it out
|
|
#
?
Sep 7, 2022 00:56
|
|
|
Mozi posted:how do you tell stable diffusion to only edit a certain part of an image? You can do it with some work in img2img, whatever you want to keep crop out in photoshop or GIMP. After you are done place it back in the image and set the denoise to like 0.1 extremely low. This will basically just do nothing but try and combine the two keep adding steps until it looks good. Other things may shift a bit depending on home many steps you need. It helps if you don't MS paint do it and at least half rear end blend it with blur or something to reduce steps. Oh you want to work with Mona Lisa, have fun that's the hardest thing you can work with.
|
|
#
?
Sep 7, 2022 01:14
|
|
|
SD - comic love across the universe, colorful, creative yea..I went with a hippie dippity doo theme other variation turned into a 70's album cover  and a 1980's art for a girl's Trapper Keeper 
|
|
#
?
Sep 7, 2022 01:35
|
|
|
Sorry, I tried the search function with the thread, but it's not bringing anything. I just wanted to ask: do we have any exemples of AI art in motion? Either of a succession of image an artist compiled, or even full video capacities of having the AI build a "scene" in its code to then navigate? I seem to recall some noise in that regard, but I can't remember nor find for the life of me. I'll search more myself tomorrow, but until then if someone has a link/reference handy it would be welcome. Basically I've been debating the technology with cinematographers, and the ones I've talked too would like to see more how this could be used in motion, instead of just concept art. (The fashion tweet up top was interesting, but nothing that can't be done in a day now with basic CGI. The powers that be seem to have their pipeline entrenched and unable to see how this could disrupt the current field of cinema).
|
|
#
?
Sep 7, 2022 03:17
|
|
|
"a wizard by zdzisław beksiński" Stable diffusion, upscaled.   
|
|
#
?
Sep 7, 2022 03:17
|
|
|
"Isometric full frame game card with decorative border, Hyper detailed game card, official art"  so midjourney will just design a card boarder for me! combine this with MtG Set Editor https://magicseteditor.boards.net/ As well as Toothycat magic card generator https://www.toothycat.net/wiki/bnf.pl?page=AlexChurchill/MagicCardGenerator and you can generate an infinite number of new magic cards. lol goodbye wizards of the coast i'm coming for you
|
|
#
?
Sep 7, 2022 03:43
|
|
|
Popoto posted:Sorry, I tried the search function with the thread, but it's not bringing anything. I just wanted to ask: do we have any exemples of AI art in motion? Either of a succession of image an artist compiled, or even full video capacities of having the AI build a "scene" in its code to then navigate? I seem to recall some noise in that regard, but I can't remember nor find for the life of me. I'll search more myself tomorrow, but until then if someone has a link/reference handy it would be welcome. https://i.imgur.com/d9DTLUs.mp4 From Reddit
|
|
#
?
Sep 7, 2022 03:54
|
|
|
^ Thats incredible. I've been running simple props through a bunch of img2img iterations and then running that through midjourney in the style of Alan Lee. This is after running through the creative filter which kind of removed all the Alan Lee-ness of it.  Still kickin' rad Cannot stop using this thing        
ddiddles fucked around with this message at 07:41 on Sep 7, 2022 |
|
#
?
Sep 7, 2022 04:41
|
|
|
https://www.youtube.com/watch?v=nVhmFski3vg This Two Minute Papers video shows some cool demos. Is there a webUI front end overlay for the local install to put sliders on the screen for tweaking the various expressed weights? Around the 1:40 mark there is a really neat sequence with smooth dialed weights morphing outputs seamlessly across latent space. If there isn't one, I'll have to try and make one. I've been meaning to learn to like tinkering with python. I can do it provisionally and I love many of the python libraries but am not really sold on significant whitespace (30 years of programming languages with C-like syntax to unlearn)
|
|
#
?
Sep 7, 2022 07:34
|
|
|
|
| # ? May 29, 2024 20:09 |
|
|
https://twitter.com/supercomposite/status/1567162288087470081?s=21&t=PKbvh-83UEfWpJi322Hnkg
|
|
#
?
Sep 7, 2022 11:26
|
|
