|
"A neon gate of pain and sadness, Photoshoot, White Balance, Tetrachromacy" No idea where the pain and sadness fit in, but this gate looks cool as poo poo. Upscaled with GFPGAN I love how AI art makes these otherworldly places seem real. This would be an amazing art exhibit. "breakdancing tournament on the sidewalk, new york city, thomas kinkade, dusk, streetlamps" This is amazing art Vlaphor fucked around with this message at 03:35 on Aug 25, 2022 |
# ? Aug 25, 2022 03:22 |
|
|
# ? May 30, 2024 05:35 |
|
Im curious to see if AI generated content is going to be accepted as "real" art alongside manmade pieces and not just a tech curiosity at some point. Personally, Ive enjoyed some of the things DALL-E 2 and the like produce more than the average art Ive encountered on the internet.
|
# ? Aug 25, 2022 03:54 |
Monglo posted:Im curious to see if AI generated content is going to be accepted as "real" art alongside manmade pieces and not just a tech curiosity at some point. Personally, Ive enjoyed some of the things DALL-E 2 and the like produce more than the average art Ive encountered on the internet. I'm thinking that if there was an intent behind, sure. Someone made a comic using stable diffusion https://twitter.com/abubu_newnanka/status/1562032093873393664 Apparently they redrew a lot of the characters and it took 300 images but still, we're in very early stages, people are still figuring out how to best use it. https://twitter.com/abubu_newnanka/status/1562119036342743040 https://twitter.com/abubu_newnanka/status/1562289318772850688
|
|
# ? Aug 25, 2022 04:06 |
|
|
# ? Aug 25, 2022 04:07 |
|
KakerMix posted:Apologies is this has been posted before but this site Not all of them, at least not yet. Hopefully they're still populating their database, but it's missing most of the prompts I came up with. For example it's missing all the images I made of "x celebrity cosplaying as y at DragonCon" as well as "yoda sitting in a high chair at a table in the restaurant outback steakhouse, annoyed, 3 5 mm, f / 2. 8 " -W 768 -S 3516339913" Searching that site for "outback steakhouse" returns exactly one result: "girl in bikini on ipad at outback steakhouse"
|
# ? Aug 25, 2022 04:34 |
|
lunar detritus posted:I'm thinking that if there was an intent behind, sure. And we’re only at the infancy stage. I’m sure there’ll eventually be tools to train and/or lock in specific character designs and other elements for re-use. You can already sort of do that with smart img2img massaging.
|
# ? Aug 25, 2022 04:38 |
|
Doggles posted:... "yoda sitting in a high chair at a table in the restaurant outback steakhouse, annoyed, 3 5 mm, f / 2. 8 " -W 768 -S 3516339913" I was inspired...except I replaced Yoda with Henry Rollins That last one is fairly close to my memory of him, except for the lack of tattoos
|
# ? Aug 25, 2022 04:47 |
|
presenting, Mona Lobster!
|
# ? Aug 25, 2022 05:03 |
|
Vlaphor posted:I was inspired...except I replaced Yoda with Henry Rollins Looks more like Harley from the Cro-Mags
|
# ? Aug 25, 2022 05:10 |
|
|
# ? Aug 25, 2022 05:48 |
|
random prompts on unrelated images can be fun this painting of Aphrodite it made earlier + "a derelict space ship" (+ the usual artstation etc quality booster) which gave the outputs a nice color palatte. one ship reminded me of Peter Elson, and the results are great I also like going on image2image journeys, where the ai's output steers the directions of my input in about 25 steps I went from this to this Ever wonder what a Jill sandwich looks like?
|
# ? Aug 25, 2022 05:50 |
|
Emma Watson and Steve Buscemi share a coke...? edit: Mozi posted:presenting, Mona Lobster! Cable Guy fucked around with this message at 07:18 on Aug 25, 2022 |
# ? Aug 25, 2022 07:15 |
|
I finally got K diffusion working. It's marginally faster. I still can't make images bigger than 512*512, though. Also the GFPGAN tool (face cleanup) works pretty well, but I'd recommend being careful with it, and only applying it manually, because by default it can excessively "beautify" the subject "woman by caravaggio" The top image, without GFPGAN, looks more like a real Caravaggio. The bottom face looks a bit airbrushed, and the eyes and eyelashes are a bit too photographic. Also it seems to like making blue eyes brownish
|
# ? Aug 25, 2022 10:20 |
|
Not necessarily news, but punctuation matters. This sequence was made by (painstakingly, manually) consecutively adding " ," (space and comma) to the prompt (until I ran out of space; there's a limit to how long the prompt can be) from beautiful woman on tachikoma, in dystopian city, caravaggio to beautiful woman on tachikoma, in dystopian city, , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , caravaggio https://i.imgur.com/9lhTRGJ.mp4 >> I think it's deemphasizing carvaggio? Usually the effect is subtle, but when I was using the Prompt Matrix function, I noticed that one of the results looked very different than when I typed it out by hand. Turns out I was missing a space and comma, because I didn't realize how the function was naming the outputs motoko_kusanagi,_beautiful_woman,_in_dystopian_city_,_sophie_anderson__ motoko_kusanagi,_beautiful_woman,_in_dystopian_city_sophie_anderson__ (underscores instead of spaces because I'm copying filenames) The only difference between the two is a single instance of "_,". Usually the difference isn't as big, but this really stood out. Rinkles fucked around with this message at 14:48 on Aug 25, 2022 |
# ? Aug 25, 2022 14:39 |
|
so , should be another prompt you are basically telling it to think of a blank image for each comma I think?
|
# ? Aug 25, 2022 14:43 |
|
|
# ? Aug 25, 2022 15:00 |
|
"tree in the shape of a face, insanely detailed and intricate, hypermaximalist, elegant, ornate, hyper realistic, super detailed, F/2.8" This is the kind of cursed content I love making with AI Art,
|
# ? Aug 25, 2022 15:32 |
making giant crabs is weirdly hard "traffic accident caused by huge monster crab, grainy, photo journalism"
|
|
# ? Aug 25, 2022 17:04 |
|
That Stable Diffusion "open-source" stuff. That for real or some sort of leak they made to work?
|
# ? Aug 25, 2022 17:07 |
|
Combat Pretzel posted:That Stable Diffusion "open-source" stuff. That for real or some sort of leak they made to work? most of the posts here the last few days are using it. or rather the official open source release.
|
# ? Aug 25, 2022 17:09 |
|
Oh okay, I can continue installing it then. --edit: Can't hold a candle to DALL-E so far. --edit: lol "pink hairy fluffy Godzilla standing at the beach, surrounded by world war 2 tanks, photorealistic, cinematic lighting" (DALL-E blocks this one for whatever reason.) Combat Pretzel fucked around with this message at 17:47 on Aug 25, 2022 |
# ? Aug 25, 2022 17:19 |
|
It might be "war", I had some prompts rejected for trying to do Civil War photos. Maybe try "WW2" or "WWII"?
|
# ? Aug 25, 2022 18:10 |
|
It was the "hairy" adjective. The gently caress. That said, no tanks whatsoever.
|
# ? Aug 25, 2022 18:28 |
|
Combat Pretzel posted:It was the "hairy" adjective. The gently caress. Maybe fluffy or fuzzy?
|
# ? Aug 25, 2022 18:29 |
|
Cate Blanchett is Tor Johnson Joe Biden making it rain Mozi fucked around with this message at 18:55 on Aug 25, 2022 |
# ? Aug 25, 2022 18:31 |
|
Got 512x512 working finally by just cutting the number of samples to 1 and running more iterations. It really is the practical minimum IMO, images just look so much better than any resolution below that.
|
# ? Aug 25, 2022 21:07 |
|
I really like how this one turned out. "Still of Dr. Jordan B. Peterson as Dr. Frank N. Furter in the Rocky Horror Picture Show (1975)"
|
# ? Aug 25, 2022 21:14 |
|
Mr Luxury Yacht posted:Got 512x512 working finally by just cutting the number of samples to 1 and running more iterations. It really is the practical minimum IMO, images just look so much better than any resolution below that. Don't go above it either, it duplicates things. It was trained at 512x512 and not much has been done yet to help other sizes not look hosed up. It *should* get better with time. Don't be afraid to use an upscaler after to take it to 2000x2000 or 8000x8000
|
# ? Aug 25, 2022 21:15 |
Love the generate 250 images in a row take the best slowly refine them in img2img and then do 250 img2img from the best of that.
|
|
# ? Aug 25, 2022 21:42 |
cheesetriangles posted:Love the generate 250 images in a row take the best slowly refine them in img2img and then do 250 img2img from the best of that. Yeah, it's amazing after getting used to disco diffusion's 10 minutes per image.
|
|
# ? Aug 25, 2022 21:50 |
|
a hunka hunka hunka hunka hunka burning love
|
# ? Aug 25, 2022 22:10 |
|
I've found that usually 30 sampling steps is good enough, and it really speeds up rendering. Especially for just for browsing possibilities. You can always re-render at with more steps (if you save the details!). sometimes the ai just nails it Rinkles fucked around with this message at 22:29 on Aug 25, 2022 |
# ? Aug 25, 2022 22:26 |
|
Elmo the Barbarian by Frank Frazetta
|
# ? Aug 25, 2022 22:55 |
|
So can you add to the existing training sets? That is, if I wanted to beef up a local version of this thing with a better grasp of my own OC Sonic creations, could I annotate a bunch of my art and feed it back to the AI in some way?
|
# ? Aug 26, 2022 00:56 |
|
Gromit posted:So can you add to the existing training sets? That is, if I wanted to beef up a local version of this thing with a better grasp of my own OC Sonic creations, could I annotate a bunch of my art and feed it back to the AI in some way? I believe so, but you'd need a lot to make a difference
|
# ? Aug 26, 2022 01:10 |
|
To experiment with prompts, I've been working on Ben Franklin's YouTube channel.
|
# ? Aug 26, 2022 03:01 |
|
"sheep wearing evil masks, hyperdetailed, sinister vibes," somehow got me really weird, but really realistic, sheep masks.
|
# ? Aug 26, 2022 04:03 |
|
i had a bunch of free Dall-e credits so i used them to generate pictures of wizards in the style of various famous artists
|
# ? Aug 26, 2022 05:30 |
|
what are all these people doing? was there an accident or something? oh hey what's that over... aaaaaaaaaaaaaaaaa gently caress
|
# ? Aug 26, 2022 06:11 |
|
|
# ? May 30, 2024 05:35 |
|
I used the same with stable diffusion seed trying to refine my prompt and was really not getting the results I was going for. Then I tried a bunch of random seeds with the same prompt and it turns out the prompts were fine, I just had chosen an unfortunate seed value.
|
# ? Aug 26, 2022 07:39 |