|
Mozi posted:cool! threw that into stable diffusion and got nailed it
|
# ? Aug 21, 2022 19:07 |
|
|
# ? May 30, 2024 14:07 |
|
Got ahold of that Stable Diffusion leak, was curious about "Guy Fieri fighting a dragon vray" Something about that third picture just loving kills me "ethereal crystal cave, trending on tumblr, mystical vibe" Vlaphor fucked around with this message at 20:01 on Aug 21, 2022 |
# ? Aug 21, 2022 19:22 |
|
All 3 of these are some variation of "Path through an evil forest" (with a few different modifiers and different seeds) and all 3 share strange similarities. Using the leaked Stable Diffuser Especially of note is the way the several of them start out in the same place, and several of them curve in the same way.
|
# ? Aug 21, 2022 22:24 |
|
|
# ? Aug 22, 2022 00:42 |
|
legit incredible. as thumbnails ai images look totally real. wondering what kind of technology they'll eventually need to prove an image isn't generated.
|
# ? Aug 22, 2022 00:50 |
|
I was inspired, same prompt in Stable Diffusion Questions abound, like why does the guy in the first pic have a ghost hand on his shoulder The worlds largest man with the world's largest stromboli on the 4th pic Finally, props to the double neckbeard in pic 3.
|
# ? Aug 22, 2022 00:52 |
|
this one is great, andre the giant is in the background, the guy is cupping the strombols.
|
# ? Aug 22, 2022 00:53 |
|
Vlaphor posted:I was inspired, same prompt in Stable Diffusion I'm guessing because lots of people have someone's hand on their shoulders in photos so the model might've learned that.
|
# ? Aug 22, 2022 00:58 |
|
|
# ? Aug 22, 2022 01:25 |
|
I can't stop pushing the make art button, now that I can make it locally. A couple of "Gateway between life and death" with a few modifiers and "human hand with too many fingers, spooky vibes, oil on canvas"
|
# ? Aug 22, 2022 01:27 |
|
Vlaphor posted:All 3 of these are some variation of "Path through an evil forest" (with a few different modifiers and different seeds) and all 3 share strange similarities. Using the leaked Stable Diffuser Are you sure it is using different seeds? All six slots line up across different prompts.
|
# ? Aug 22, 2022 01:37 |
|
Vlaphor posted:A couple of "Gateway between life and death" with a few modifiers and "human hand with too many fingers, spooky vibes, oil on canvas" This makes me want to see some stuff inspired by Caspar David Friedrich’s paintings
|
# ? Aug 22, 2022 01:43 |
|
Vlaphor posted:"Gateway between life and death" I thought it would be bigger. PS: v cool imagery.
|
# ? Aug 22, 2022 02:10 |
|
lunar detritus posted:This reddit comment has instructions on how to run an optimized version locally Is the optimized version supposed able to generate 1024x1024 images on a GPU with 12GB of VRAM? I keep trying using the arguments: --W 1024 --H 1024 --n_samples 1 and I'm getting this error each time even with the optimized scripts, so not sure if I'm doing something wrong: RuntimeError: CUDA out of memory. Tried to allocate 8.00 GiB (GPU 0; 12.00 GiB total capacity; 4.73 GiB already allocated; 4.88 GiB free; 4.79 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF
|
# ? Aug 22, 2022 03:19 |
|
Does the leaked home version have the same restrictions as the current legit version, or is it pretty much open to whatever goblin dongs you want to make?
|
# ? Aug 22, 2022 03:33 |
|
dot communist posted:Does the leaked home version have the same restrictions as the current legit version, or is it pretty much open to whatever goblin dongs you want to make? "Goblin penises by Dan Witz, oil on canvas", I don't think it's NSFW, but I'm gonna play it safe. {Maybe NSFW, but not really} Edit: Second attempt (with "spooky atmosphere" added, not at all nsfw This art is beyond bizarre and I love it. Also, it doesn't seem to want to make goblin dongs. Vlaphor fucked around with this message at 04:15 on Aug 22, 2022 |
# ? Aug 22, 2022 04:10 |
|
Vlaphor posted:I can't stop pushing the make art button, now that I can make it locally. What's your gpu, if you don't mind sharing?
|
# ? Aug 22, 2022 04:15 |
|
Rinkles posted:What's your gpu, if you don't mind sharing? 3080 10gb. edit: For some reason, putting "3080 RTX" in place of goblin penises gets this weird trippy art that has nothing at all to do with video cards. I'm loving this kind of art. The right kind of hazy and spooky. Vlaphor fucked around with this message at 04:20 on Aug 22, 2022 |
# ? Aug 22, 2022 04:17 |
|
"Empire of skulls, painted by zdzisław beksiński, paint on black canvas" gives some metal as gently caress results. That mountain in the first pic is amazing.
|
# ? Aug 22, 2022 05:06 |
|
Beep beep
|
# ? Aug 22, 2022 05:30 |
|
I don't mean to keep spamming this thread, but I keep getting jaw dropping results. This is a couple different iterations of "gateway between life and death" with a wide range of modifiers That's enough for now, I gotta get some sleep.
|
# ? Aug 22, 2022 05:34 |
|
https://twitter.com/abcdentminded/status/1561211951161569280
|
# ? Aug 22, 2022 06:06 |
|
Symphoric posted:This is probably no help because I have no idea what I'm doing, but I got it to run using those same directions. The first time I ran the optimizedSD script I got the same error as you, I then ran the regular (non-optimized) python script which hung for a VERY LONG time before finally it ran out of memory and errored out. I then ran the optimized script again and it suddently worked for no reason I can explain, and it's been working fine since. Well it works now. I nuked everything and started over with this. I think an issue was that some stuff failed to install automatically and I did it manually with the pip installer, but that hosed up the environment creation. No idea why it would fail to find the OptimizedSD module though as that has nothing to do with installed packages. loving python. Anyway it's also extremely slow on a 1070. Only uses 3.7 gigs so maybe there's quite a bit to be gained by actually using all 8 gigs. E: ah ok the default prompt makes 20 images. So about 1 minute / image. At least it's free The default prompt is pretty poo poo but it did get some cool results too. mobby_6kl fucked around with this message at 11:31 on Aug 22, 2022 |
# ? Aug 22, 2022 11:22 |
|
mobby_6kl posted:Well it works now. I nuked everything and started over with this. I think an issue was that some stuff failed to install automatically and I did it manually with the pip installer, but that hosed up the environment creation. No idea why it would fail to find the OptimizedSD module though as that has nothing to do with installed packages. loving python. You managed that stuff on your 1070?? I have a 1080 ti and just assumed getting a local instance of this was out of reach.
|
# ? Aug 22, 2022 13:26 |
|
The many horrible faces of "shrek self portait, 8k, magical vibes, realistic, vray lighting"
|
# ? Aug 22, 2022 13:55 |
|
Vlaphor posted:The many horrible faces of "shrek self portait, 8k, magical vibes, realistic, vray lighting" The first one is blowing bubblegum with his eyes! That's a neat trick
|
# ? Aug 22, 2022 13:56 |
|
pixaal posted:The first one is blowing bubblegum with his eyes! That's a neat trick Nope, Shrek is a deadite in that one.
|
# ? Aug 22, 2022 14:03 |
|
A Strange Aeon posted:You managed that stuff on your 1070?? I have a 1080 ti and just assumed getting a local instance of this was out of reach. It's definitely not as good as the latest live version, I couldn't even get a single anime Kissinger out of it. But it's still possible to get some really cool stuff.
|
# ? Aug 22, 2022 14:46 |
|
I might try installing stable diffusion. Does it work with AMD cards or do I need those nVidia CUDA cores?
|
# ? Aug 22, 2022 14:53 |
|
The dark world of "guy fieri cooking people, drawn by zdzisław beksiński, sinister vibe" Sadly, cooking people didn't really make the cut, but what did is amazing. That last pic is so good.
|
# ? Aug 22, 2022 14:59 |
|
Nice Van My Man posted:I might try installing stable diffusion. Does it work with AMD cards or do I need those nVidia CUDA cores? on the 13th the lead dev said "oh yeah just got it working on a AMD RX6800" The 4chan leaked version predates this though I think, so you might need to wait for the official release. Emad (I think that's the lead dev) is counting down to something later today on Twitter. It's either publicly announcing the website, releasing the github, or a test for the audio AI. I guess it could be something else but I can't think of anything else that would be major enough to count down in this project.
|
# ? Aug 22, 2022 15:05 |
|
That local version spits out png’s that I can see thumbnails of but are invalid images when I open them
|
# ? Aug 22, 2022 15:07 |
|
Vlaphor posted:"Goblin penises by Dan Witz, oil on canvas" Huh. Well, next time we have asparagus my girlfriend is going to wonder why I'm giggling
|
# ? Aug 22, 2022 15:22 |
|
SD is going public in 10 mins, maybe? /edit/ apparently: https://huggingface.co/CompVis/stable-diffusion-v-1-4-original Popoto fucked around with this message at 19:05 on Aug 22, 2022 |
# ? Aug 22, 2022 18:50 |
|
https://twitter.com/EMostaque/status/1561777122082824192 it's out!
|
# ? Aug 22, 2022 19:12 |
|
lunar detritus posted:I found this neat collab, that basically asks CLIP how it'd describe an image. Then, you can use that prompt to generate your own. Yeah, it's actually super useful, like just taking the prompt it gives you and putting it into craiyon immediately gets pictures that are surprisingly close to the vibe at least. edit: modifiers like "photo taken with nikon d750", sure whatever, it was taken with a nikon d750 lol Wheany fucked around with this message at 19:54 on Aug 22, 2022 |
# ? Aug 22, 2022 19:48 |
|
Trying the new model that just got released ( I just swapped out the old model that was there and renamed the new one to model, that seems to work), tried a couple of things. "guy fieri building a snowman, whimsical vibes, 8k" Came out fairly well. I especially love that last image, since it feels on brand with him "shrek building a snowman, whimsical vibes, Hyper Detail, 8K, HD, Octane Rendering, Unreal Engine, V-Ray, full hd" That came out a bit different. So far it feels basically identical to the old model (no changes that I can tell right away), except the model is 3gb smaller and image creation seems to take slightly less time. Edit: Threw in the true test, "Gateway between life and death" Same seed on all 3 "gateway between life and death, sinister vibes, Hyper Detail, 8K, HD, Octane Rendering, Unreal Engine, V-Ray, full hd" "gateway between life and death, sinister vibes, painted by zdzisław beksiński" "gateway between life and death in the middle of a field, magical vibes, painted by zdzisław beksiński" I think they look alright. Vlaphor fucked around with this message at 20:21 on Aug 22, 2022 |
# ? Aug 22, 2022 20:12 |
|
Wheany posted:Yeah, it's actually super useful, like just taking the prompt it gives you and putting it into craiyon immediately gets pictures that are surprisingly close to the vibe at least. A lot of the device modifiers I think work by what type of photo that camera most was used for. I assume certain models become popular in certain types of photography so it will have trained on that type of image in higher quality on an expensive camera an amateur wouldn't buy.
|
# ? Aug 22, 2022 20:16 |
|
Whoa poo poo that's sooner than expected! I'm certainly not suddenly regretting loving around with python all weekend to make the leak work Vlaphor posted:Trying the new model that just got released ( I just swapped out the old model that was there and renamed the new one to model, that seems to work), tried a couple of things. Neat, looks great! It's getting pretty late here though so I think I'll wait until tomorrow until the model is on torrents or something
|
# ? Aug 22, 2022 20:20 |
|
|
# ? May 30, 2024 14:07 |
|
"forest of streetlamps, snowy lighting, 8k, magical vibes" Different seeds though. Old model New model
|
# ? Aug 22, 2022 20:28 |