|
Closing in on 5000 images on MJ and I have about 75-100 pieces I'm ready to upscale for a print project. My plan is to start with Adobe Lightroom's Super Resolution, which gives you 4x resolution with the best quality I've seen compared to Topaz and similar PS plugins. Anyone else tried that with AI art? I'll post some samples to compare this evening.
|
#
?
Sep 2, 2022 22:37
|
|
|
|
| # ? May 14, 2024 06:57 |
|
|
Bula Vinaka posted:https://www.youtube.com/watch?v=L9hlfc16qg0&t=565s This is a new medium. It may not be other artists' medium of choice, but it is a tool to turn ideas into images and, yes, ultimately, art. Everyone can take a paint brush and splash paint on a canvas like Pollock did - or in a way that resembles his work. Why is his work considered art and mine most likely wouldn't be? Did he control every single random splash to get the exact outcome he intended? Where do you draw the line between randomness and intent when it comes to art? Is reproducibility an issue? Warhol used screen printing to mass produce his art. It was part of his process.
|
|
#
?
Sep 2, 2022 22:45
|
|
|
mobby_6kl posted:If you don't have an nVidia GPU I don't even know what to say That I've been using ATI/AMD GPUs since the 90s and up until now, I have yet to find anything that I actually cared about that specifically needed an nVidia card ? Famed comic artist Dave McKean (Sandman covers and tons of other stuff) talking about what he thinks of this new-fangled nonsense...http://chimeraobscura.com/vm/episode-497-dave-mckean
|
|
#
?
Sep 2, 2022 22:50
|
|
|
Bad Purchase posted:now let me describe the idea for a game i've had for the past 20 years and see if it can generate one I have had this idea of doing a game based on a Soviet Submarine which almost launched a nuclear torpedo during the Cuban missile crises. Got lots of pics of the particular class of sub. This was originally going to be a VR title in 2017 but funding died. Now I could see using the images and making it into a Visual Novel Adventure with this tool. Do you have to pay for the rights to use the images generated?
|
|
#
?
Sep 2, 2022 23:55
|
|
|
VideoGameVet posted:
with stable diffusion, no. In dream studio (their hosted version) they say that every generated image is public domain but I'm not sure if that applies if you only use the weights locally. With dall-e 2, no (besides the fee to use it) but they have a non-exclusive license to use it for whatever they want with midjourney, no (besides the fee to use it) unless you make more than a million in revenue, and they have a non-exclusive license to use it for whatever they want lunar detritus fucked around with this message at 00:19 on Sep 3, 2022 |
|
#
?
Sep 3, 2022 00:16
|
|
|
Pretty happy with how the first images out of SD local came out, based on "a cat in a mech suit, studio photography, intricate, highly detailed, sharp focus, trending on artstation, 4K, award winning, shadows":
|
|
#
?
Sep 3, 2022 01:39
|
|
|
realistic reflections might be fun with that last one.
|
|
#
?
Sep 3, 2022 01:41
|
|
|
 Bonus Space Meowrine 
|
|
#
?
Sep 3, 2022 02:08
|
|
|
mcbexx posted:This is a new medium. It may not be other artists' medium of choice, but it is a tool to turn ideas into images and, yes, ultimately, art. I think it's going to come down to a few things - people like art to be a display of skill. You often hear "my kid could have done that!" to disparage art, even ignoring AI art. With AI art you're definitely going to get even more people saying "I could have typed in prompts until I got a picture as nice as that!" The skill just isn't immediately apparent at this point in time. There's also the intention vs. randomness thing. That's tricky because even if the artist claims intent, it will gnaw at people that maybe the artist is bullshitting and they just picked out a pretty random picture that the AI spit out. Hell, this is also a problem people have with existing art, that the artist might just be bullshitting some intent behind a pretty picture they drew. Now it's exacerbated by the fact that there might literally be near zero intent behind a relatively complex image. I think it will probably evolve as its own thing a bit like photography. Artists with consistent output won't have to argue that they meant to create a certain piece. I could see in something like an art competition having a predefined theme to enforce intent. As people get more accustomed to AI generated images they will also get more accustomed to their shortcomings, which means if you can do something like get it to spit out a clear image of dragons fighting F-16s, or a crowd of people where their limbs and hands aren't all hosed up people will be impressed by that. Of course that will shift as the AIs improve. Funnily enough I could see AI generated art having the same impact even for non-digital art. The AI isn't just good at drawing pictures, it's also good at color, composition and occasionally anatomy (I'm assuming this will improve). If you wanted to win a painting competition you could generate pictures until you found something really nice, then just copy it in (physical) paint. There's a little more technical skill there, but I'm sure people would still be pissed if they found out an artist just copied an AI piece. I'm eagerly awaiting the first "mildly famous painter found to be just copying random AI pictures!" controversy.
|
|
#
?
Sep 3, 2022 02:48
|
|
|
The badly composed faces and hands are almost identical to how people appear in every dream I've ever had. Creepy as gently caress. Edit to add content:    yoloer420 fucked around with this message at 02:58 on Sep 3, 2022 |
|
#
?
Sep 3, 2022 02:52
|
|
|
Nice Van My Man posted:people like art to be a display of skill. This is not true in the slightest. In fact, the majority of pop culture is dominated by a race to the bottom, lowest common denominator stuff, be it movies, music, artists on social media, etc. That’s not new either, and goes back centuries at least. Art snobs aren’t even wholly in that camp, as entire movements are based around “easy” or “bad” art from a technical perspective. Counter culture is often made up of this kind of thing that swells until it becomes mainstream. Bottom Liner fucked around with this message at 03:37 on Sep 3, 2022 |
|
#
?
Sep 3, 2022 03:35
|
|
|
mobby_6kl posted:I haven't done many realistic images with DALL-E, but the doggo's head here seems fine, pretty photorealistic without any weirdness really. I've use the watermarking function in Photoshop, and it only adds barely perceptible noise. It seems like it's mostly just how their algorithm works sometimes, but admittedly I don't the details of what's causing it. And everything I would like to be? For I can fly higher than a beagle And you are the wind beneath my wings...
|
|
#
?
Sep 3, 2022 05:22
|
|
|
mcbexx posted:Everyone can take a paint brush and splash paint on a canvas like Pollock did - or in a way that resembles his work. You have a point. Pollock's 17A sold for 200 million in 2015. I also think of Cy Twombly... what he did, in the years he did it, play a role I think. Someone making similar artwork today wouldn't have the success that he did, I don't think. So I guess I opened a "what is art" can of worms, which I didn't intend. Sorry about that. 
|
|
#
?
Sep 3, 2022 05:35
|
|
|
Bula Vinaka posted:You have a point. Pollock's 17A sold for 200 million in 2015. Now with AI we have the debate of "what is artist" which fits right in with art itself.
|
|
#
?
Sep 3, 2022 06:29
|
|
|
It’s art if you can swindle some rich moron into buying it
|
|
#
?
Sep 3, 2022 06:36
|
|
|
'art' is a syntactically meaningless series of letters, and the sooner we acknowledge that the sooner we can stop talking about it.
|
|
#
?
Sep 3, 2022 08:49
|
|
|
BARONS CYBER SKULL posted:It’s art if you can swindle some rich moron into buying it https://en.wikipedia.org/wiki/Artist%27s_Shit
|
|
#
?
Sep 3, 2022 09:08
|
|
|
I'm sure they've been posted in this thread before but I like this channel's videos of taking song lyrics, putting them into one of these generators, and making a slideshow of them set to the song: https://www.youtube.com/watch?v=-_VYGvHpED8&hd=1 I'm partial to the Shaq with Shaq's mouth on his own stomach https://www.youtube.com/watch?v=elQGj6uar1A&hd=1 Bo Burnham - Welcome to the Internet https://www.youtube.com/watch?v=ueTH6wqA9K4&hd=1 gotta say, wasn't expecting Bo Burnham calling me a whore to result in this:  but maybe he knows me better than I know myself 
|
|
#
?
Sep 3, 2022 09:18
|
|
|
I was looking at how SD would interpret some of my Craiyon prompts, and I accidentally set the wrong aspect ratio when im2img-ing one of the results but that gave me this great image of dystopian high rises reinterpreted as a traffic jam of trams based off this  based on this prompt 
|
|
#
?
Sep 3, 2022 10:37
|
|
|
So with img2img I don't think describing what's in the scene matters as much as the style you want to see, it's already looking and has decided on what the shapes most of the things in the scene are going to be, and it really didn't seem like changing the keywords (at .5 denoise) did much to change up the ability to style transfer. I've been testing this out to make comic backgrounds, and have had a lot of successes. Maybe this prompt matrix will help someone figure out their own stuff? code:
|
|
#
?
Sep 3, 2022 12:37
|
|
|
low poly is a cool aesthetic in its own right, but it can also be used with illustrations to subtly or not so subtly affect the artstyle. it mixes well with some artists.  __________________ a low poly golden retriever wearing sunglasses, gta5 box art, artstation, by rhads, andreas rocha, akoto shinkai, laurie greasley, lois van baarle (modified from tote up a bags)   __________________ low poly close up of woman in dystopian city, gustave moreau alphonse mucha, rebecca guay, studio ghibli, 4k, sharp, Katsuhiro Otomo, Hiroya Oku Production I.G   __________________ Stanisław Wyspiański close up of low poly peasant woman in dystopian city, gustave moreau, Mamoru Oshii studio ghibli, 4k, sharp, Katsuhiro Otomo, rebecca guay   __________________ Hiroya Oku Production I.G low poly close up of woman in hong kong dystopian city, Mamoru Oshii, studio ghibli, 4k, sharp, Katsuhiro Otomo, gustave moreau alphonse mucha, rebecca guay Boba Pearl posted:So with img2img I don't think describing what's in the scene matters as much as the style you want to see, it's already looking and has decided on what the shapes most of the things in the scene are going to be, and it really didn't seem like changing the keywords (at .5 denoise) did much to change up the ability to style transfer. I've been testing this out to make comic backgrounds, and have had a lot of successes. I think that largely depends on the denoising strength. Sometimes you want more of a variation of an image, but still keep the same subjects, which might disappear as you increase denoising if you're just inputting the artstyle.
|
|
#
?
Sep 3, 2022 13:03
|
|
|
True, my use case is wanting as little variation as possible from the scene I have, basically using it as a more advanced style transfer. I've also learned if you use edge detection to get lines on the image and send it through again, you'll get lined art.
|
|
#
?
Sep 3, 2022 13:07
|
|
|
Boba Pearl posted:So with img2img I don't think describing what's in the scene matters as much as the style you want to see, it's already looking and has decided on what the shapes most of the things in the scene are going to be, and it really didn't seem like changing the keywords (at .5 denoise) did much to change up the ability to style transfer. I've been testing this out to make comic backgrounds, and have had a lot of successes. You can change things significantly at 0.7 Here's a link to my post when I was playing with it. I kept playing and made a few more posts with more examples from the same image over the next day or so. https://forums.somethingawful.com/showthread.php?threadid=4000251&userid=47135&perpage=40&pagenumber=2#post525750613  
|
|
#
?
Sep 3, 2022 13:11
|
|
|
Any way to help Dall E to make eyes consistent? I always have the problem that it draws the pupils differently. Like one is without reflections, one eye with, one has a spark, the other doesn't -- it's unsettling.
|
|
#
?
Sep 3, 2022 13:21
|
|
|
|
|
#
?
Sep 3, 2022 14:35
|
|
|
https://i.imgur.com/0Kjdbh9.mp4 Alt ending  Frames should be in order https://imgur.com/a/0zUN11x pixaal fucked around with this message at 15:23 on Sep 3, 2022 |
|
#
?
Sep 3, 2022 15:00
|
|
|
I know there's a lot of information about running SD locally but it's distributed, please bear with me: Is there some guide emerging for a user friendly stable diffusion script for local running? I can do a little bit of python but I'd prefer something that was easy to use. Also is there a site or blog that collects some insides into the effects of various prompts? I know some users here have been reporting the effects of including punctuation, different syntaxes, and so forth. pixaal posted:https://i.imgur.com/EjyGpRY.mp4 Wild poo poo! How did you make it? Just variations of the same image? Feeding it into img2img repeatedly?
|
|
#
?
Sep 3, 2022 15:11
|
|
|
Lord Stimperor posted:I know there's a lot of information about running SD locally but it's distributed, please bear with me: I will write you one my friend. First Requirements for the thing I'm using: Windows, Nvidia card there's linux and amd distros, but I don't know how to use that. Ok, go here: https://github.com/AUTOMATIC1111/stable-diffusion-webui/ Unzip this to it's own folder. You download it by clicking Code and then download zip It has all the following links, but it doesn't super explain it You're going to need two models, you want the Full EMA I believe, though someone will tell me the differences I'm sure. https://huggingface.co/CompVis/stable-diffusion-v-1-4-original https://github.com/TencentARC/GFPGAN/releases/download/v1.3.0/GFPGANv1.3.pth Put this file in the same folder as the files from the git link So it should be in the same directory as all the files from stable diffusion. As in webui.py webui.bat and your model file should all be in the same directory. Also, rename the 1.4 model you downloaded to model.ckpt. GFPGANv1.3.pth should remain the name it has. You need to run and install python 3.10.6 https://www.python.org/downloads/windows/ You need to run and install git (64 git bit for windows setup) https://git-scm.com/download/win and you'll need the Cuda Toolkit 11.3 (Windows, x86_64, version 10) https://developer.nvidia.com/cuda-11.3.0-download-archive?target_os=Windows&target_arch=x86_64 run webui.bat from Windows Explorer.
|
|
#
?
Sep 3, 2022 15:26
|
|
|
Lord Stimperor posted:I know there's a lot of information about running SD locally but it's distributed, please bear with me: The initial is "A Kitten eating the sun" with a handful of artists that do psychedelic work. Then I gave it a kitten eating the sun at like 0.4 and each frame is one run at like 50 steps the last step is 0.6 and 200 steps. Each frame was part of a batch of 20. Oh some of the later frames I changed it to a cat with butterfly wings when the background started to crinkle, and it reminded me of butterfly wings like the last 3 or 4 only. pixaal fucked around with this message at 15:39 on Sep 3, 2022 |
|
#
?
Sep 3, 2022 15:36
|
|
|
Bottom Liner posted:This is not true in the slightest. In fact, the majority of pop culture is dominated by a race to the bottom, lowest common denominator stuff, be it movies, music, artists on social media, etc. That’s not new either, and goes back centuries at least. I was just listing it as one way to get someone recognized as an artist that's impacted by AI, I didn't mean to imply it was required to be an artist. For the lowest common denominator art you listed I'd actually say the opposite, that most of it is technically outstanding even if it doesn't do much otherwise.
|
|
#
?
Sep 3, 2022 15:41
|
|
|
Lord Stimperor posted:Any way to help Dall E to make eyes consistent? I always have the problem that it draws the pupils differently. Like one is without reflections, one eye with, one has a spark, the other doesn't -- it's unsettling. I think people are mostly passing the faces through another program to fix them. Here's one you can try: https://arc.tencent.com/en/ai-demos/faceRestoration.
|
|
#
?
Sep 3, 2022 16:12
|
|
|
TIP posted:I was having the same issue and aside from the reboot I also made one change to the installation that might be what fixed it. During startup of SD I noticed a warning that said something like "LDSR is missing" and said that it was missing a "latent-diffusion" folder. Will give that a try, awesome thanks! Seeing all the things you all do with img2img is inspiring. I need to experiment with it more. I wanna work towards eventuallt wielding it rather than lean on it.
|
|
#
?
Sep 3, 2022 16:24
|
|
|
Rinkles posted:
I'm gonna start stealing people's prompt crafting now. Hiroya Oku Production I.G low poly close up of the corn cob king, Mamoru Oshii, studio ghibli, 4k, sharp, Katsuhiro Otomo, gustave moreau alphonse mucha, rebecca guay     (DALL-E)
|
|
#
?
Sep 3, 2022 16:53
|
|
|
Goddamn Stable Diffusion is fun as heck.  Cadillac Brougham 911  1964 Chrysler Imperial Fiat 500, if Fiat owned Chrysler in the 60s  Neon bullnose Ford  1977 Dodge Viper, imagine that V-10 in the late 70s  1954 Audi TT  Porsche 911 and BMW M1 hybrid  Dodge Aries (block 80s car) Toyota 2000 GT mix It's the late 90s again and I've got a bunch of paper and my colored pencils and am just drawing cars all night
|
|
#
?
Sep 3, 2022 17:07
|
|
|
Those are great. Did you have to cherry pick a lot (nothing wrong with that), cause my attempts for similar photographs tend to have a relatively high failure rate?
|
|
#
?
Sep 3, 2022 17:28
|
|
|
Rinkles posted:Those are great. Did you have to cherry pick a lot (nothing wrong with that), cause my attempts for similar photographs tend to have a relatively high failure rate? Sort of? I don't have a particular image in my head, just a general idea, then I see what comes out then refine towards what it shows me. What I want goes hand in hand with what it shows me and I'll go in different directions, sometimes going back and scraping what I just did for the last 5 minutes. I'm away from my computer at the moment but I'll share the prompt I use, just gotta swap in whatever you want pictures of. It's really similar to the one I shared before with all the neon. I tested it with 1.5 when that ran for a short while on discord and it's all different, can't wait to relearn how to prompt with that update!
|
|
#
?
Sep 3, 2022 17:38
|
|
|
Boba Pearl posted:[instructions] Unfortunately, the potato I'm working with here is a mess of old versions of python stuff, so it's throwing errors whenever I try to install anything like this. Even if I do, I only have 2gb of vram. Can I run this at terrible resolution, or is 4gb a strict cutoff?
|
|
#
?
Sep 3, 2022 17:40
|
|
|
Brutal Garcon posted:Very helpful.
|
|
#
?
Sep 3, 2022 17:50
|
|
|
I heard you can run it on a CPU, but it’s really really slow.
|
|
#
?
Sep 3, 2022 17:54
|
|
|
|
| # ? May 14, 2024 06:57 |
|
|
Drive Through RPG has put up a big flashing warning to publishers that they must tag any publications that contain AI generated art. Im guessing the lawyers are getting a bit worried about what's gonna happen with copyright. They want any easy way to pull everything if the conservative supreme court rules AIs are an afront to Jesus or something.
|
|
#
?
Sep 3, 2022 17:58
|
|