|
neato burrito posted:Sifl and Olly doing the Titanic pose together on a canoe, ultra realistic. I don't think stable diffusion knows who sifl and olly are 
|
#
?
Aug 7, 2022 04:34
|
|
|
|
| # ? May 29, 2024 20:31 |
|
|
Stable Diffusion is fast as gently caress. Spits out results in like 2 seconds. Nicholas Cage in Dune, cinematic  
|
|
#
?
Aug 7, 2022 04:44
|
|
|
TIP posted:I don't think stable diffusion knows who sifl and olly are Aw, thanks for trying anyway 
|
|
#
?
Aug 7, 2022 04:55
|
|
|
!dream "Vibrant colored poster for a rockband, time traveling spiders, glossy finish, 4k, octane render, 1983" -H 960 -s 80 -S 2921744743 !dream "Vibrant colored poster for a rockband, time traveling spiders, glossy finish, 4k, octane render, 1983" -H 960 -s 80 -S 2921744743 --cfg_scale 12  
|
|
#
?
Aug 7, 2022 04:57
|
|
|
ymgve posted:(this one makes me really suspect it sometimes "remembers" trained photos instead of generating them) I think it's generating the photos. It's just "memory" is actually a generative process (even in humans). When you "remember" things you are not seeing a .mpg file replay, your mind is doing whatever Dall-E is doing and reconstructing the training data. Like data compression
|
|
#
?
Aug 7, 2022 04:57
|
|
|
This one was too good to not share: Damien Hirst art installation titled Meat, 4K photo
|
|
#
?
Aug 7, 2022 05:03
|
|
|
|
|
#
?
Aug 7, 2022 05:43
|
|
|
 “butter sculpture of a golden retriever”
|
|
#
?
Aug 7, 2022 06:48
|
|
|
Neat idea by my wife.
|
|
#
?
Aug 7, 2022 07:06
|
|
|
Right so Stable Diffusion is now the new hotness, just full-stop incredible stuff pouring out from it. It is bad at whimsy and outright fantastic, asking it for things like bannanas boxing or a can of soup on a facetime call doesn't give good results, or at least I don't know the prompts for them. It's real good at other things though: 1955 Pontiac Firebird concept as drawn by Syd Mead  Same prompt but a station wagon this time  Children's book about a 1955 Pontiac Firebird by Dr. Seuss  Robocop as Guy Fieri cinematic Eastman 5384 film  Robocop as mayor of Flavortown cinematic Eastman 5384 film  Robocop as mayor of Flavortown cinematic Eastman 5384 film   Famous heavy metal glam rocker Bernie Sanders with a neon mohawk, ripped denim jacket, and lots of jewelry       Famous heavy metal glam rocker 20 year old Bernie Sanders with hair his mom would not approve of studio photograph for his new metal album    Film still of Nicolas Cage in Mad Max 2 Fury Road as an evil raider (not my prompt nor my images but had to share)      a scary filmic wide shot color ground level angle movie still 35mm film photograph of the full body of a dangerous shape shifting alien creature, with multiple mutated snarling drooling human faces with a grotesque variety of human and animal limbs protruding from its lower torso inside a lab, in the style of nature documentary footage, The Thing 1982 (also not mine but gently caress me, look at those results)  macready winning at chess (  )Stole 'The Thing' prompt from above and modified it a bit:    filmic wide shot color ground level angle movie still 35mm film photograph of the full body of an obese man eating mayonnaise out of a jar inside a lab     Nicolas Cage as god of chaos in a hood dark fantasy, intricate, smooth, artstation, painted by Wayne Barlowe, zdislav beksinski    Nicolas Cage as god of chaos in a hood dark fantasy, intricate, smooth, artstation, painted by Wayne Barlowe (notice no longer has the beksinski bit in this prompt)             Monstrous screaming in agony Nicolas Cage dark fantasy, intricate, smooth, artstation, painted by Wayne Barlowe, zdislav beksinski       Took the above prompt and added "Garbage Pail Kids trading card" on the front       Monstrous screaming in agony Van Gogh dark fantasy, intricate, smooth, artstation, painted by Wayne Barlowe, zdislav beksinski  Above prompt but 'Marylin Monroe"  Nicolas Cage as god of flames (plus the rest of the prompt from above)       Mister Rogers as god of happy hugs dark fantasy, intricate, smooth, artstation, painted by Wayne Barlowe, zdislav beksinski I am more convinced than ever that we're in for a wild, wild ride for ai art generation across the board.
|
|
#
?
Aug 7, 2022 07:11
|
|
|
KakerMix posted:Right so Stable Diffusion is now the new hotness, just full-stop incredible stuff pouring out from it. It is bad at whimsy and outright fantastic, asking it for things like bannanas boxing or a can of soup on a facetime call doesn't give good results, or at least I don't know the prompts for them. It's real good at other things though: Midjourney is my favorite for paintings by far. Vague prompts still get you gorgeous paintings with a wonderful texture and style most of the time. Stablediffusion is now my favorite for photos. It does a great job representing film, camera, and lens types. If you ask it to put people in Stranger Things or Mad Max or whatever it usually gets the visual style of the source film correct.
|
|
#
?
Aug 7, 2022 07:49
|
|
|
https://twitter.com/RoboArtistry/status/1556130211145535488
|
|
#
?
Aug 7, 2022 08:03
|
|
|
Feeling a bit sad to have missed the stable diffusion invite, but I'm sure the next months/year will bring even better models. What a ride. Meanwhile, here's Mona Lisa doing shopping, 1988 polaroid by midjourney
|
|
#
?
Aug 7, 2022 08:13
|
|
|
   Dingus. 
|
|
#
?
Aug 7, 2022 08:40
|
|
the Mods, they knew!
|
"Posters for the cancelled Mortdecai sequels"     
|
|
#
?
Aug 7, 2022 10:27
|
|
|
 
Lamebot fucked around with this message at 12:23 on Aug 7, 2022 |
|
#
?
Aug 7, 2022 11:52
|
|
|
A few more from Stable Diffusion. SD seems to have much wilder swings than Midjourney. The highs are higher, but it will also routinely make sub-Craiyon level trash. SD also seems to like long, detailed descriptions. You can feed Midjourney something like "Verdant forest with pond", and get a decent, visually interesting result. With SD, you really want to feed it as much info as possible. Write an entire paragraph describing your scene, the detail you want, artist inspirations, rendering style, etc etc. Just go nuts describing what you want. That's not to say you can't get a good result with a simple description, but it's a crapshoot whether you receive something great or something awful. All of the best stuff I've seen after several hours of watching output has had really long, detailed descriptions.  
|
|
#
?
Aug 7, 2022 12:56
|
|
|
I'm messing around on Stable Diffusion and I'm not quite getting the results I'm hoping for. When I look at other people's prompts I see at the end they usually include a bunch of numbers and letters, but I have no idea what any of that means or where to find out. Does anybody have any ideas where I can look? Like others have said, discord is a really sub optimal platform for this kind of thing. On a related note, I think it would be interesting/helpful if we were to compile our combined knowledge into something more easy to read and search than a thread. I've learned a lot reading everybody's posts, but it's easy to lose track of stuff when I want to read helpful posts again.
|
|
#
?
Aug 7, 2022 14:54
|
|
|
-W width (must be multiple of 64) -H height (must be multiple of 64) -n number of images -i split up images into different files instead of collaging them -S seed (regenerates same image with same prompt. NOTE: CASE SENSITIVE) -s number of steps the model takes (controls grainyness and detail) -C cfg scale, controls how hard it tries to optimized for the prompt. google doc showing effects of cfg and seed https://docs.google.com/spreadsheets/d/1SYQhyJaKkkY0cmPd0WQvPwEX188l5FZxzukkC7IQDw4/edit#gid=0
|
|
#
?
Aug 7, 2022 15:29
|
|
|
Some stable diffusion horrors of previous prompts: "a swarm of cheesy rats with cheesy bread stick tails in a multitude of boxes of cheesy bread stick rats being served by a rat waiter to a family at an intergalactic Pizza Hut on the outside of a space station" -n 9  "photorealistic baby seal eating a disgusting amount of chicken wings and living mutant pasta fazool surrounded by an audience of biomechanical pizza ducks laughing and dripping with mayonnaise in a Washington area forest" -n 9  "baby seal disguised as a loaf of bread making repairs on the outside of the international space station" -n 9 
|
|
#
?
Aug 7, 2022 16:20
|
|
|
One thing I noticed about Stable Diffusion is that going beyond the default 512x512 resolution often fails and creates multiples of your focus object, probably because they do a form of auto-inpainting with a sliding window, so the extra generations don't "see" that you already have for example a full face in the image, so it creates an extra one. It works well with landscapes and other stuff that don't have a singular focus, though. I do hope they release the full model pretty quickly so I don't have to fear getting kicked out of the beta for using it too much
|
|
#
?
Aug 7, 2022 16:43
|
|
|
I don't have Stable Diffusion access yet but I'm kinda into just watching the feeds of people making stuff. Just the sheer variety of high quality, new, frequently bizarre images is awesome in the "awe" sense. A year ago I would have said there was no way I'd be this in to computer generated art, but the human element is still present in the actual creative side, and I guess that's enough apparently. 
|
|
#
?
Aug 7, 2022 17:30
|
|
|
Tunicate posted:google doc showing effects of cfg and seed Play with these per image with -C and -s as you get close lock in the seed and change these values slightly to dial it in remember the seeds are always in the same order so if you like images 2 and 3 from a set you can generate a set of 2 using seed 2 and get both! This job took 4.2seconds !dream the creepy doll in the box asking if you will fix it a snack too while you are up, studio lighting, warm colors -s 30 -n 2 -C 12 -S 1532310911  I had maybe 6 generations before this with the 1-minute limit on submitting right now that means this took me 6 minutes of work assuming I thought of all corrections in the time period (I was waiting the timer every time on this one, the concept came together surprisingly quick). I also clearly drew inspiration from a song on random. some concepts need out there settings! I couldn't get a universe in a snowglobe to look right at all until -C 60+ that 's an insane level.
|
|
#
?
Aug 7, 2022 17:32
|
|
|
"a swarm of biomechanical ducks made of cheese pizza taking over a hospital maternity ward -n 9" LOL at the beds just covered in pizza, or are actually pizza themselves? No clue what is going on in the center one, looks like a tower of nachos with olive eyes or something. Mercury_Storm fucked around with this message at 18:16 on Aug 7, 2022 |
|
#
?
Aug 7, 2022 18:12
|
|
|
 https://twitter.com/RoboArtistry/status/1556184627605151744?s=20&t=QfHr8vR4VU0Aau1mCdaDSQ "Hong Kong street in Bladerunner" 
|
|
#
?
Aug 7, 2022 18:40
|
|
|
https://twitter.com/genekogan/status/1555184488606564353
|
|
#
?
Aug 7, 2022 18:43
|
|
|
"Inside Fox News headquarters in Manhattan, The Pursuer boss from Dark Souls 2 smashes desks and destroys sound sets, high quality realistic live camera footage -n 9"
Mercury_Storm fucked around with this message at 21:57 on Aug 7, 2022 |
|
#
?
Aug 7, 2022 19:24
|
|
|
Yeah thats the kind of tools Im hoping to see become feasible. Using the ai to generate individual elements and then merge them all coherently. Be interesting once they get to the point where 3d geometry instructions and pose uploading and stuff becomes common.
|
|
#
?
Aug 7, 2022 20:43
|
|
|
I've managed to get into Dall-E, Midjourney, and Stable Diffusion, and I've been having a lot of fun with them, but I'm also starting to understand Frank Herbert's whole Butlerian Jihad thing
|
|
#
?
Aug 7, 2022 20:52
|
|
|
"Occupy Wall Street except the protestors have assault class nuclear weaponry, lava chainsaws, and the graviton gun from Syndicate Wars, high quality news footage"
|
|
#
?
Aug 7, 2022 20:55
|
|
|
Willem Dafoe as The Joker, still from the Dark Knight, detailed, 4k    
|
|
#
?
Aug 7, 2022 21:14
|
|
|
I wonder if it's possible with something like stable diffusion to use a simple object prompt, then reuse the seed with rotation information prompts, and pull images to reconstruct the 3d object with photogrametry. I'd guess the info is visually mostly correct but may fail to generate enough true physical characteristics to pull the data back together.
|
|
#
?
Aug 7, 2022 21:17
|
|
|
I tried to get Craiyon to draw me the picture that Jeremy is drawing in the Pearl Jam song.   
|
|
#
?
Aug 7, 2022 21:29
|
|
|
Tunicate posted:-W width (must be multiple of 64) Thank you!
|
|
#
?
Aug 7, 2022 21:29
|
|
|
Sedgr posted:I wonder if it's possible with something like stable diffusion to use a simple object prompt, then reuse the seed with rotation information prompts, and pull images to reconstruct the 3d object with photogrametry. It doesn't work in a 3D space so that's not possible. You don't have precision enough to say "[some object] camera to the left 45 degrees" or something, at least not with current models.
|
|
#
?
Aug 7, 2022 21:44
|
|
|
"baby seal disguised as a loaf of bread engages the enemy faction and fires the graviton gun from Syndicate Wars while at Mar-a-Lago, view from a news truck, satellite lasers strike down in the distance, high quality news footage" -n 9
Mercury_Storm fucked around with this message at 21:51 on Aug 7, 2022 |
|
#
?
Aug 7, 2022 21:47
|
|
|
ymgve posted:It doesn't work in a 3D space so that's not possible. You don't have precision enough to say "[some object] camera to the left 45 degrees" or something, at least not with current models. there is one system that generates video clips, I imagine that could be used to describe an object/person rotating and you might be able to get something half usable out of that it's still pretty wobbly and weird, but these things progress quickly
|
|
#
?
Aug 7, 2022 21:50
|
|
|
   
|
|
#
?
Aug 7, 2022 22:58
|
|
|
Doing stupid things in Stable Diffusion. Here's Danny DeVito as the Saint of Killers Danny DeVito having a thought  Danny DeVito as the Doom Slayer 
|
|
#
?
Aug 7, 2022 23:29
|
|
|
|
| # ? May 29, 2024 20:31 |
|
|
Ogdred Weary posted:Danny DeVito as the Doom Slayer I thought it was the cyberdemon that had huge guts, not the doom slayer!
|
|
#
?
Aug 7, 2022 23:34
|
|