|
Fish Appreciator posted:A friend snapped a candid photo of me at my 30th birthday, then fed it into clip interrogator, then fed the prompt into midjourney. This was the result. tell your friend to press the remaster button  
|
#
?
Oct 31, 2022 01:33
|
|
|
|
| # ? May 29, 2024 23:59 |
|
|
"Faberge cake" "a mall ninja, obese, sweaty, cheap cosplay clothes, tacticool vest, overcoat, thick glasses, nerdy af"  "a pantheon of fabulous gay Greek gods, hella rainbows and gay stuff, style of the school of Athens by Rafael"   
|
|
#
?
Oct 31, 2022 01:57
|
|
|
          The above and below are the same prompt but the bottom one is scientifically 50% more anime.   
|
|
#
?
Oct 31, 2022 02:02
|
|
|
"two children summoning demons by Bob Byerley and Nicoletta Ceccoli"  
|
|
#
?
Oct 31, 2022 02:29
|
|
|
Sumerian Telecom posted:I've merged sd1.4 with that novelAI pedantically tagged anime model. It tends to make the imagery neater and smoother and gives a generally more attractive look. Proooooooompts
|
|
#
?
Oct 31, 2022 02:48
|
|
|
Made a quick little cenobite model for halloween. I'll upload it if somebody is interested Emma Watson (a tiny bit gory so spoilered)  Ron Perlman  Iron Man  Milla Jovovich  Doctor Strange  Garfield Cat  Spiderman  For some reason, It strongly associates Adam Sandler with the Chaterrer Cenobite
|
|
#
?
Oct 31, 2022 07:15
|
|
|
IShallRiseAgain posted:Made a quick little cenobite model for halloween. I'll upload it if somebody is interested
|
|
#
?
Oct 31, 2022 09:24
|
|
|
Happy halloween! We've got: Wolves!  Pirates!  Costumes!  Eldritch gods!  Lolth!  T̸͙̯̃̉͂̕ḧ̷̟́ȅ̵͎̜̥̥͌͒ ̶̰͈̓͌̽K̷̡̗͍͈͍͛͠i̴̖̯̗̬̖̻̔̂̈́̍̀̄n̵̙̮͚̤̂̐̏̾͝g̵̠͔̺̐ ̸̧̧̧̺̲̹́͝͝i̷̡̗͇̫̾̑̽̕ǹ̷̺̩̤́͋ ̶͕̭̀̏͐̒̏Y̶̮̩̎̑̓̀͝e̸̢͈͈̥̯͋͜l̷̡͔̹̐͐͂̏̿l̴̩͉̪̟̀̋̈́͝ǫ̸̬̤̺̂̑w̶̼͍̮̬̎̑̓̈̕!̷̡̨͚̠̗͎̉͑̀  Ȟ̶̰͉̼͕̒͒̓͆o̴̢̱̜̩̜̤̙̒̂̏̈̾͆̊̓̆̔̆̊͝r̴̨̡̦͓͖̬̖̰̫̼̈́̉̈̾̈́͑̆̂̄͊̃̉͜͝ŗ̵̡̥̻̘̦͎̭̃͊̔̃́͂̊͆͜ọ̶͚̝̫̦̌̐͌̑̂̍r̴̛͙̩̪̼̻̰͎̭̼͈̃̀̋̎͜͝s̸̫̞̬̙̟̒̀͜ ̴̢̡̮̥̤͈͔͕͓̱͍̉̇͆̚͘ͅḇ̶̧̧̤̗͈̳̏͗̏̃̾̈́̓̿̎͌͌͐͝ȩ̷͚̼̫͚͖͔̼̿̈͆̌ỹ̶͕̺͙͓̤̪͎̣͕͑́͊̍̐̎̀̈́̅͋͝͝o̷͇̘̥̫͙̜̼͚̩͠n̵̡̟͚̦̭̯͇̪̟̣̳͕̪͗͗͋̓̆́͌̆̅̎̆͋d̴̛̠͚̝̬́̍ ̴̢̧̩̻͉̮̍w̴̡̺̗̠̝̖̩̬͈̱͗͂͑̀̔̍͛̾̌͋̈͘͜h̵͚̫̲̳̺͕̠̰͓̏̐͑͝ͅa̸̤̝̠̽͗͘ẗ̸͖̣̺̳͍̳̫̼̱̠̥̭̑̑̔̍͐̉̀̕͜͠ ̸̛͕̪͎̣̓̉̑̇͆̊̅̾͌̋̎̚m̴̡̗̞̲̹̫̋̌̔͗̂͐̈̾̕a̴͙̭͎̯̋ǹ̶̨̞͎͎͎̹̣̱̹̳͚̹͍ ̸̡̢̨̡̰̺̤̪͚͚̞̲̒̽̓̃̽̽̏̇͒ͅw̴̡̫̫̰̻̘̘̮̺̹̬̜̓̉̉́̓̎̚a̷͎̓̓̈́̅̑͝s̷̼̪̻̙͚̙̠͈͇̼͂̓͛̈́͒̄̉̑͝͝ ̶͕̞̖̦͙̓̈̑̓̈́̇̃͋̌́͋̈́̕ṁ̵̫̳̎̓̓̉̎͆̀̕ȩ̴̢͙̘̰̉͗̈́͠ã̸̡͙͖̥̠́n̵̢̗͍̣̥̩̙͙͚̄͗̈́͋ṭ̷̛̺̦̹̬͓̰̗̭̲͂̈́͒̎̌͆̌̄ ̵̡̛̥̗̳̍͊̆͒͑̕͝t̶͉̫̖͚̩̩̟̥̋̂̋̈̍͐̈́̾͛o̴̧̧̢̨̜͇̞͙̻̦͖̲̿̿̏̔̕͜͠ ̶͖̹̲̳̫̋́̒̏́̃͌̃̊͒̊͜͝͝ç̵͕̟̖͉͕̠̳̍̉̕ó̶̫̭̳̪̺͓̘͗̒̓̌͛̄̅̉̈́̑̽m̵̧̹̜̻̳̰̹͔͔͈͐͂̄͑͗̿͜p̷̢̟̟̋͆̎̉̅̈̈́̌͛̚͘̕r̸̢̺̝͖̬̬̟̯̲̮͇̓͐́̌̉̕͠ͅe̶̹̝̫̙̙̯͙̲͊̑̀́̊̿̓͗̋́͝h̵̨̅̿͂̐͋͂̓́͝e̶̙͍̟̯̜͓̫͆̽̉̉̍̈̽͘n̴̡̨͉͎̜̪̞̲̥̺̓́́̊̑̈̆͐̉̒͠ḏ̶͈͓̙̫̫̗͂͜ͅ.̵̢͇̬̙̹̮͍̝͈̠̳͆̃̔̈̎̚͝   Scps!  Gross bodily trauma!   Zombies!  And finally, Cheese! 
|
|
#
?
Oct 31, 2022 10:47
|
|
|
Good show everyone, these last few pages...
|
|
#
?
Oct 31, 2022 10:56
|
|
|
I did some more Halloween stuff too. Wolves!   Wizards!  Vampire Hunters!   Tentacles!
|
|
#
?
Oct 31, 2022 16:46
|
|
|
IShallRiseAgain posted:Iron Man "We'll tear your soul apart, Mister Lister sir"
|
|
#
?
Oct 31, 2022 22:20
|
|
|
Found a two-for-one artist. I was checking names with A masterpiece artwork by *Artist. Stable returns framed paintings or color splatters if it doesn't recognize the artist. It's not a perfect test, some people slip through and only trigger if the prompt asks for a portrait for instance, but mostly it's obvious when it knows the artist. Anyway, Samwise Didier seemed to work. This could've been some off-brand Blizzard art he did.  But he returns Hobbit faces in his style if you don't add more prompts.  Doesn't look like you can take the Samwise out of Didier either. Stable doesn't recognize the name Sam Didier. There's bound to be a bunch more prompts that do this too.
|
|
#
?
Oct 31, 2022 23:04
|
|
|
Here's a hundred artists that Lexica's Stable Diffusion recognizes if anyone's curious. Prompts run at 14/20 on the site's "guidance scale". 512p size, upscaled to 1024p. https://imgur.com/a/BMuI8M3 It was too much effort saving more than one image for each. Still, it shows the base palette and style they'll add. I found it much easier throwing artists together with a cheat sheet.  A masterpiece artwork by Greg Staples :: Mort Künstler :: Magali Villeneuve :: Ruan Jia  A masterpiece artwork by Oliver Wetter :: Marc Brunet :: Neil Blevins  Rockabilly Snow White, a masterpiece artwork by Marc Brunet :: Jae Lee :: Loish  A masterpiece artwork by Tim McBurnie :: Mathias Osland :: Victo Ngai  A masterpiece artwork by Atey Ghailan :: Jordan Grimmer :: Max Bedulenko :: beautiful photorealistic rendering :: maximalist :: hyperdetailed :: 8k :: 16k :: amazing depth :: splash art :: intricate :: ray tracing reflections :: octane render :: trending on Artstation
|
|
#
?
Nov 1, 2022 00:26
|
|
|
 Rosie O'Donnell as Lara Croft
|
|
#
?
Nov 1, 2022 00:42
|
|
|
a dystopian advertisement for a gigantic monster size Kia Sportage war-machine SUV with spikes and multiple guns and cannons dripping with blood and gore being driven by a middle-aged soccer mom with a smiling happy family of 2.5 kids in a Walmart parking lot in the style of Mad Max, highly detailed, 8k photo  I'm titling this one "Insane Clown Posse and the Mutated Mole Children"   OPEN THE BLOOD GATES, WITH KIA AND THE FISHER PRICE MACHINE GUN NEST
|
|
#
?
Nov 1, 2022 01:04
|
|
|
"uninked pencils for a comic page of vermeer's girl with a pearl earring, 8k" I did a bunch of different "uninked pencils for a comic page of [famous painting]" attempts, and it's an interesting look into how SD "thinks" about things. It seems like the effects of training are super nonlinear, where the most famous paintings/artists have a massive effect on output if they're mentioned, even if you fiddle around with attention weights. On the other hand, even fairly famous-but-not-most-famous artists/paintings are way less "sticky" in terms of the attention effect. So trying to get SD to draw a comic page of the Vermeer almost invariably results in a "direct quote" or whatever you want to call it of the painting: you get the Girl with a Pearl Earing in exactly that pose, with exactly that wardrobe, and so on. Same thing with the da Vinci's Mona Lisa and, to a lesser extent, van Gogh's Starry Night (SD kept producing pencil sketches of details from the real Starry Night and slight variations of that kind of thing). On the other hand, Rembrant's Night Watch comes out as:  It more or less never just tries to do a sketch version of the original painting, but instead tries to produce things that look like comic pages with distinct panels, different characters, and so on (occasionally it seems to try to do a Spy vs. Spy thing with the guy with the white hat and the guy with the black hat from center of the original painting). And it doesn't seem to be just an effect of the subject matter: Mona Lisa and Girl with a Pearl Earring are "simple" portraits with a single subject, but René Magritte's The Son of Man is "understood" to be about a guy in a hat and SD is more flexible about how it handles him (these are numerically consecutive seeds):    Same with Grant Wood's American Gothic, which is presumably a less "confusing" painting than the Magritte (which has an apple obscuring the subject's face):   Not really going anywhere with this, just making some observations about SD's behaviour.
|
|
#
?
Nov 1, 2022 02:03
|
|
|
I think night watch is a comic as well as a didcowrld book, which probably muddles things
|
|
#
?
Nov 1, 2022 02:06
|
|
|
Tunicate posted:I think night watch is a comic as well as a didcowrld book, which probably muddles things In the case of the Rembrandt it seems like it understands the reference to the painting, because the central figures (Frans Banninck Cocq and Willem van Ruytenburch) consistently show up, but maybe the comic of the same name is helping with the "comic-ness" of the results.
|
|
#
?
Nov 1, 2022 02:12
|
|
|
DALL.E 2 image, prompt "lizard person in a red wizard robe realistic"
|
|
#
?
Nov 1, 2022 03:19
|
|
|
WhiteHowler posted:I did some more Halloween stuff too. Which model is this? I really like the art style
|
|
#
?
Nov 1, 2022 03:37
|
|
|
Had a bit of a bizarre AI Art experience. I was inspired to see what AI art thought of "Bobson Dugnut" and got this. which is fine...but then I remember it's supposed to have 2 t's at the end. So I just do "Bobson Dugnutt" and change nothing else  Who are these people? Then I switch it from Heun (which I feel gets me the best people) to DPM2 A and change nothing else (same seed even) and get this.  I still have no idea who these people are, that guy on the upper right is the most realistic AI person I've ever seen, and how does the guy below him fit in at all?
|
|
#
?
Nov 1, 2022 03:58
|
|
|
I think you're getting some Kevin Bobson in there.
|
|
#
?
Nov 1, 2022 04:06
|
|
|
Suppose that may be possible, but all I changed from pic 1 to 2 was an extra letter t. I wonder if the word "nut" overrides "bobson", but once you ad the second t, then it stops seeing it as the word "nut" At the very least "Todd Bonzalez" makes more sense. 
|
|
#
?
Nov 1, 2022 04:08
|
|
|
tote up a bags posted:Which model is this? I really like the art style All were SD 1.5 except the tentacles were 1.4. Most of them were a mix of several different artist styles. I've been trying to keep from using the same artists repeatedly, because forcing myself to branch out helps me discover awesome new ones. Vlaphor posted:Had a bit of a bizarre AI Art experience. I was inspired to see what AI art thought of "Bobson Dugnut" and got this. A while back I did "a baseball card of _____" with each of the names from Fighting Baseball and got some fun ones, but none were really keepers. I also tried it for Key & Peele's East-West Bowl names, but same thing. You get mostly random people, a few mutants, and nothing super noteworthy.
|
|
#
?
Nov 1, 2022 04:20
|
|
|
Vlaphor posted:Had a bit of a bizarre AI Art experience. I was inspired to see what AI art thought of "Bobson Dugnut" and got this. I plug random gibberish words into Midjourney from time to time just to see how it interprets them. Sometimes it's pretty cool.    Prompt: photoramic solarham       Prompt: glycemian harrowing   Prompt: elktroclastic volcanope e: Random gibberish word + Harrowing is in general a very very good combo. Harrowing Photoham: 
deep dish peat moss fucked around with this message at 06:47 on Nov 1, 2022 |
|
#
?
Nov 1, 2022 06:31
|
|
|
         witches dancing around bonfire under a full moon, the scene is lively and chaotic, painted by someone who took their time and paid extra attention to small details, stylistic rendition, literally steal an artists soul and put it into this picture Steps: 128, Sampler: Euler a, CFG scale: 7.5, Seed: 4254916039, Size: 512x512, Model hash: 45dee52b
|
|
#
?
Nov 1, 2022 07:19
|
|
|
I'm confused why prompts on Nightcafe and Lexica look wildly different when they're both using Stable Diffusion. Not speaking of seeds here, just the general result from the same prompts. Only Nightcafe lets you change sampling method, and they've added a CLIP guidance option to Stable today. Is there anything I can do there to make it match the hidden settings Lexica runs?
|
|
#
?
Nov 1, 2022 09:32
|
|
|
KwegiboHB posted:
 I like prompts like this, they respect the sentience of the AI. Just a big list of keywords is an insult to it's intelligence
|
|
#
?
Nov 1, 2022 11:00
|
|
|
i trained an AI to be Gary Gygax to give me advice for my D&D campaign. hes pretty good    
|
|
#
?
Nov 1, 2022 22:03
|
|
|
It is surprisingly hard to make AIs generate   ... and hooboy, does SD not want to indulge my
|
|
#
?
Nov 1, 2022 22:31
|
|
|
Megazver posted:It is surprisingly hard to make AIs generate Have you tried the furry fine tune?
|
|
#
?
Nov 1, 2022 22:33
|
|
|
"Captain Jean-Luc Picard of the U.S.S. Dumpster-prise"  "Space... The final frontier. These are the voyages of the starship Dumpster-prise. Its continuing mission, to explore strange new worlds. To seek out new life and new civilizations. To boldly go where no (dumpster) has gone before."  
|
|
#
?
Nov 2, 2022 03:26
|
|
|
Rutibex posted:i trained an AI to be Gary Gygax to give me advice for my D&D campaign. hes pretty good How do you go about doing this? Friend and I are in this group where there's two guys who have spent the last 3 years writing novels arguing with each other, we got the logs and we wanna fully automate these arguments edit: they are dicks and this would be funny BrainDance fucked around with this message at 04:51 on Nov 2, 2022 |
|
#
?
Nov 2, 2022 04:47
|
|
|
I'm assuming the V4 of Midjourney is being beta tested by a lucky handful of so-and-sos. It's a shame they didn't use the same people from the first beta, god dammit, v4 results looking real good. Like actually better than the v1.5 of Stable Diffusion
|
|
#
?
Nov 2, 2022 05:49
|
|
|
BrainDance posted:How do you go about doing this? I am using a website called character.ai. Its basically a very advanced chatbot (I think GPT-3?) that pretends to be whatever character you ask. It has to be someone famous or someone kind of generic. So you could ask it to be Abraham Lincoln or Caesar or "a florida man" because it was trained on text that has referenced those people in some way. But you can't actually feed it a bunch of new text to train it to be someone specific. It works to make Gary Gygax because some nerd fed this AI every single D&D book, and also presumably the wikipedia page about Gary Gygax. The chatbot knows who it is, and you can ask it about TSR and Dave Arneson and stuff other than D&D rules and Gygax Bot will understand what your talking about. But it just starts making things up if you ask it personal details that can't be found from a google search. You are allowed to feed it example text, but only 3200 characters. https://beta.character.ai/
|
|
#
?
Nov 2, 2022 06:53
|
|
|
KinkyJohn posted:I'm assuming the V4 of Midjourney is being beta tested by a lucky handful of so-and-sos. It's a shame they didn't use the same people from the first beta, god dammit, v4 results looking real good. Like actually better than the v1.5 of Stable Diffusion There's no beta test yet, but there's a rating page to get data on selected images made by the MJ devs. V4 is looking better than anything else I've seen so far, insane detail and coherence on some stuff. It should hopefully be out this week or early next.
|
|
#
?
Nov 2, 2022 07:12
|
|
|
Rutibex posted:"a florida man" I want endless "Florida man, why did you do [INSERT HEADLINE HERE], give me at least 800 grunts if you can't manage words".
|
|
#
?
Nov 2, 2022 07:52
|
|
|
Megazver posted:It is surprisingly hard to make AIs generate I've seen people use Jean Baptiste Monge for that. Occasionally with Brian Kesinger, which seems to add the color palette you had in those images. Couple of examples from Nightcafe. https://creator.nightcafe.studio/creation/hWJpizm43dx7DEVR3Ul7 https://creator.nightcafe.studio/creation/x2W61UuxmUnohe1cfJhU Don Bluth might work too, or at least be more consistent in color and style than trying Disney. Attack on Princess fucked around with this message at 09:05 on Nov 2, 2022 |
|
#
?
Nov 2, 2022 09:00
|
|
|
Megazver posted:It is surprisingly hard to make AIs generate "Anthropomorphic bear" worked surprisingly well for me for the ginger ale diner I tried making earlier.    E: You could then maybe use inpainting to get a more realistic head if you want. Not sure if this would work for a rabbit or mouse though mobby_6kl fucked around with this message at 11:20 on Nov 2, 2022 |
|
#
?
Nov 2, 2022 09:33
|
|
|
|
| # ? May 29, 2024 23:59 |
|
|
Thank you for the suggestions, everyone. I'll give them all a try. In other news, SD can generate sprite sheets for pixel art games now: https://www.reddit.com/r/StableDiffusion/comments/yj1kbi/ive_trained_a_new_model_to_output_pixel_art/
|
|
#
?
Nov 2, 2022 11:50
|
|