|
Wheany posted:No, it's Stable Diffusion specifically not knowing who Bugs Bunny is. Craiyon knows who he is and is doing its best: This is how midjourney handled that prompt (slightly modified because 'Heroin' is a naughty word)
|
# ? Sep 27, 2022 15:53 |
|
|
# ? May 28, 2024 23:41 |
|
Rutibex posted:"pinup girl, large feet, made of yogurt" You had to go there, didn't you?
|
# ? Sep 27, 2022 16:16 |
|
mcbexx posted:You had to go there, didn't you? lol It's cool how AI art all of the sudden simultaneously gave us the ability to make crazy cool art and also disgusting uncanny valley poo poo where you don't really know what you're looking at but you know it's unsettling
|
# ? Sep 27, 2022 16:23 |
|
anyone got any good style prompts for photographs? I have locked in 3 painted art styles so far for my styles.csv but I struggle with figuring out what makes "real life" images more accurate or higher quality. also here are the 3 styles i use all the time (formatted for pasting in to the csv in notepad) because sharing is ftw code:
|
# ? Sep 27, 2022 16:28 |
|
mcbexx posted:You had to go there, didn't you? this is the weirdest used condom I've ever seen.
|
# ? Sep 27, 2022 16:30 |
|
AARD VARKMAN posted:anyone got any good style prompts for photographs? I have locked in 3 painted art styles so far for my styles.csv but I struggle with figuring out what makes "real life" images more accurate or higher quality. Try adding poo poo like this into the negative prompts: photoshop, render, video game, 3d, painting, art, drawing, digital art, cartoon, doll, illustration It's still finicky, like everything else, but I've gotten some absolutely photo-realistic stuff doing that.
|
# ? Sep 27, 2022 16:34 |
|
AARD VARKMAN posted:anyone got any good style prompts for photographs? I have locked in 3 painted art styles so far for my styles.csv but I struggle with figuring out what makes "real life" images more accurate or higher quality. "Canon EOS R3, f1.4, ISO 200, 1160s, 8K, RAW, unedited, in-frame, photorealistic" works well.
|
# ? Sep 27, 2022 16:53 |
|
Sedgr posted:Hope Im not spamming, shout if you want me to stop. drat these look like stuff my friend used to paint, really creepy
|
# ? Sep 27, 2022 16:59 |
|
Also avoid terms like "photo-realism", "realistic", or related terms as those are universally tagged to not-photos and will confuse the model. The best I've seen all use camera lens sizes, focal depth terms, famous photographer names, etc. So something like "award winning photograph of _____ by Steve McCurry, color, kodak film stock, 50mm lens, close shot" will get you started with some decent results. Then you can play with different options from there on shot/aperture/lens/etc.
|
# ? Sep 27, 2022 17:01 |
|
IShallRiseAgain posted:"Canon EOS R3, f1.4, ISO 200, 1160s, 8K, RAW, unedited, in-frame, photorealistic" works well. Realistic shadows, realistic reflections realistic lighting / Cinematic lighting / studio lighting these are hit or miss generally do not add more than one, as stated they are ultimately from not photos but very close. Adding one or two is generally going to help if you are getting non-photos mixed in still.
|
# ? Sep 27, 2022 17:01 |
|
Objective Action posted:Also avoid terms like "photo-realism", "realistic", or related terms as those are universally tagged to not-photos and will confuse the model. The best I've seen all use camera lens sizes, focal depth terms, famous photographer names, etc. I've had photo-realism get better results than without it.
|
# ? Sep 27, 2022 17:16 |
|
IShallRiseAgain posted:"Canon EOS R3, f1.4, ISO 200, 1160s, 8K, RAW, unedited, in-frame, photorealistic" works well. Be careful about using terms like "8K", "sharp focus", and "highly detailed" for photographs. The AI tends to go overboard and makes things look "too good" to be realistic. "Photorealistic" is tricky -- sometimes it gives good results, but just as often produces a very realistic Greg Rutkowski painting instead, so I tend not to use it. Rather than "dynamic lighting" (which tends to be a tag used more often on paintings and 3D renders), I like "well-lit", or even specifying the time of day. Using terms based on the subject can work well. If you want a close-up of a person, use "modeling photo". If you're doing animals, try "nature photography, style of National Geographic". The latter even works on not-real animals, and you can absolutely get a great photograph of a griffon or a cockatrice if you work with the prompts a little! Edit:
|
# ? Sep 27, 2022 17:25 |
|
Found a limited time offer for canvas prints, 35" x 23.6" for $22. Helloooo Etsy! Or everyone I know will get "unique AI art" for the next 5 christmases. Really tempted to see how this would look printed and if it manages to keep the vibrant colors. Edit: Oh poo poo, I only just noticed, probably not exactly this one. ("Cyberpunk Red Light District") "PICE" "PID ULT" mcbexx fucked around with this message at 18:54 on Sep 27, 2022 |
# ? Sep 27, 2022 17:46 |
|
thank you all for the tips!
|
# ? Sep 27, 2022 17:55 |
Being able to create interesting art at breakneck speed is really enjoyable but not sure exactly what to do with it all. I feel like an art miner, putting on my headlamp and taking my prompt pickaxe down into the RNG mine and cracking the stone to find the gems and precious metals. A good prompt is like a sharp pickaxe paired with a talented eye for spotting the cracks that will open up a rich vein of cool poo poo.
|
|
# ? Sep 27, 2022 18:00 |
|
Sedgr posted:Being able to create interesting art at breakneck speed is really enjoyable but not sure exactly what to do with it all. I feel like an art miner, putting on my headlamp and taking my prompt pickaxe down into the RNG mine and cracking the stone to find the gems and precious metals. I like scrolling it sometimes, maybe a thread discord where we can dump in process stuff and help people working on stuff with prompt ideas? Discord is crap as a library, good prompt smithing and good images should always make their way back to the thread if we go to a discord though. pixaal fucked around with this message at 18:26 on Sep 27, 2022 |
# ? Sep 27, 2022 18:23 |
|
Sedgr posted:Being able to create interesting art at breakneck speed is really enjoyable but not sure exactly what to do with it all. I feel like an art miner, putting on my headlamp and taking my prompt pickaxe down into the RNG mine and cracking the stone to find the gems and precious metals. Critics of this technology are crying about how people with no skills will be able to crank out infinite "worthless" art and cheapen the entire industry. But there's a lot of skill involved in prompting, generating, curating, and refining the images. I like to think I'm getting pretty good at writing good prompts, but for every 100 images I crank out there's maybe, if I'm lucky, one really great piece. The "lady in the universe dress" picture I posted a few pages ago -- it's good. I'd buy a print, or even the original, if I saw it for a reasonable price at an art show. It was the result of an hour of prompt fiddling, sorting through about 250 "garbage" images to find one I liked enough to capture the seed. Then another hour of tweaking values and generating another hundred or so versions to refine the composition and character details. Plus several upscaling runs using different methods and settings. And I haven't even gone into GIMP yet to try to fix the character's eye and smooth out the masking around her face. In all, that "worthless" image will have taken me about three hours of focused work to create. What am I going to do with it? No idea. Probably post the artwork on my social media and set it as my cellphone wallpaper. But I also won't let anyone tell me that it took no skill or talent, or that the thing I made has no value.
|
# ? Sep 27, 2022 18:24 |
|
WhiteHowler posted:It's more than that, though. You're putting in work, and the quality of the work you put into these tools absolutely affects the output. Enjoy the rewards, even if it's just posting "look at this cool thing I made" on a dead comedy forum. Yeah, if anything even this thread is a great example of AI art being no different than real art in that 99% of it is bad junk. Same with the Midjourney feed, discord/etc. I've made somewhere around 3k images to get about 50 I consider usable and high quality, but that still requires editing every one manually, upscaling with Adobe stuff, etc.
|
# ? Sep 27, 2022 18:31 |
|
Yeah I've never gotten a great image form a first prompt. That is always, always just a basis, getting your outline down. For my Halloween images I spent at least an hour each just finding an image with the right kind of composition I wanted. Also the more surprising thing I've found is simplifying a prompt can be astronomically helpful in fine tuning.
|
# ? Sep 27, 2022 18:33 |
|
Sedgr posted:Being able to create interesting art at breakneck speed is really enjoyable but not sure exactly what to do with it all. I feel like an art miner, putting on my headlamp and taking my prompt pickaxe down into the RNG mine and cracking the stone to find the gems and precious metals. Swedish Thaumocracy posted:And finally, some star-dressed women! Bonus points for guessing the inspiration. That's straight out of SD, no tweaking of variations or anything though.
|
# ? Sep 27, 2022 18:34 |
|
WhiteHowler posted:And I haven't even gone into GIMP yet to try to fix the character's eye and smooth out the masking around her face. IMHO that's also part of the fun (and "creative process", if you will) - fixing up stuff, "frankensteining" parts from different variations into a satisfying composition etc. I know I am getting a lot of practice and experience from futzing with my output. I feel "artists" who huff and puff over it don't really understand what it has to offer, especially to people who are not gifted with the manual skillset, but now have a tool to express their imagination.
|
# ? Sep 27, 2022 18:35 |
|
mcbexx posted:especially to people who are not gifted with the manual skillset, but now have a tool to express their imagination. Also this. Holy poo poo so much this.
|
# ? Sep 27, 2022 18:36 |
|
Ok dumb question about AUTOMATIC1111. I went back to tweak an image I made a few days ago. It was the fourth image in a batch of four, and if I put in the seed that was saved in the log, I can get back all of them. But if I start messing with variation seeds, it only applies to the first image of the batch so I get four variations of the first image. Essentially I guess I need to get the seed for the n-th image in a batch. Any ideas how this could be done? mcbexx posted:IMHO that's also part of the fun (and "creative process", if you will) - fixing up stuff, "frankensteining" parts from different variations into a satisfying composition etc.
|
# ? Sep 27, 2022 19:07 |
mobby_6kl posted:Ok dumb question about AUTOMATIC1111. I went back to tweak an image I made a few days ago. It was the fourth image in a batch of four, and if I put in the seed that was saved in the log, I can get back all of them. Each seed is sequential from the initial random roll so it should just be seed+n or whatever. Or you can drag it into png info tab and it should be there.
|
|
# ? Sep 27, 2022 19:11 |
|
There's also a box in settings to save a text file with every single setting with every single image. So a batch of 12 gets your 12 png files and 12 txt files. the txt files should basically be the same except have the seed for each image. I like this because I know exactly what settings I used and don't have to write them down.
|
# ? Sep 27, 2022 19:17 |
|
pixaal posted:There's also a box in settings to save a text file with every single setting with every single image. So a batch of 12 gets your 12 png files and 12 txt files. the txt files should basically be the same except have the seed for each image. I like this because I know exactly what settings I used and don't have to write them down. Assuming you didn't turn off PNG metadata in the config, the PNG Info tab will show you all of the settings used to generate an image: prompt, seed, CFG setting, sampling steps, etc. This will let you easily recreate an image, even when using the "-a" variant sampling methods. Just be sure not to lose that original output PNG when you're editing and upscaling and such.
|
# ? Sep 27, 2022 19:21 |
|
"LICFCKE" "Party's over, the po-po is here." mcbexx fucked around with this message at 20:11 on Sep 27, 2022 |
# ? Sep 27, 2022 19:46 |
|
Thanks all! I actually did check the metadata but directly in irfanview and explorer and it seemed weird that there's nothing there, but it's right in the file after the PNG header. Good to know!
|
# ? Sep 27, 2022 19:53 |
|
Trying to come up with silly examples to demonstrate multi-prompting in MJ. Ideally these would be two-word phrases where you'd get something inaccurate if you tried to combine the individual words without prior knowledge. Best one I've got so far is "crab cake" vs "crab :: cake" (I tried "hot dog" but the dogs weren't as funny as the pastry carapaces here)
|
# ? Sep 27, 2022 20:35 |
|
A colorful painting of a robot hand that is painting a human hand on paper, detailed, intricate, geometric. It didn't get the idea very much, but it pumped out some nice colors at least
|
# ? Sep 27, 2022 20:54 |
|
Automatic1111 looks amazing but looking on the site it says it has support for 4Gb cards... but I only have a 3gb 1060. Does anyone know if it works on that too or if 4gb is the bottom limit?
|
# ? Sep 27, 2022 21:03 |
|
Objective Action posted:Also avoid terms like "photo-realism", "realistic", or related terms as those are universally tagged to not-photos and will confuse the model. The best I've seen all use camera lens sizes, focal depth terms, famous photographer names, etc. You've doomed us all "award winning photograph of Hillary Clinton with clown makeup by Steve McCurry, color, kodak film stock, 50mm lens, close shot" Edit: Maybe a public spreadsheet to save all the best prompts ? I'm very random with them and at best write down some ideas on a notepad text file Comfy Fleece Sweater fucked around with this message at 21:14 on Sep 27, 2022 |
# ? Sep 27, 2022 21:11 |
|
Isomermaid posted:Automatic1111 looks amazing but looking on the site it says it has support for 4Gb cards... but I only have a 3gb 1060. Does anyone know if it works on that too or if 4gb is the bottom limit? There is a parameter for low ram systems, it is a heck of a lot slower but could potentially work.
|
# ? Sep 27, 2022 21:23 |
|
My main dreamlog.txt file, which stores all prompts along with the corresponding seed# and output filename is just a couple of bytes shy of 3 MB and contains 7931 lines. It has been created 11 days ago.
|
# ? Sep 27, 2022 21:28 |
|
Comfy Fleece Sweater posted:You've doomed us all A spreadsheet would be nice
|
# ? Sep 27, 2022 21:39 |
|
Btw someone mentioned starting a discord, I’m in a small goon run discord with some focus in art, and we just started an ai-art channel, if you’re interested in joining PM me for an invite (hope this is ok with mods, if not just edit this)
|
# ? Sep 27, 2022 21:54 |
|
Swedish Thaumocracy posted:
Beautiful pink haired bagpipers
|
# ? Sep 27, 2022 22:00 |
|
Big Chungus, style of hans zatzka, style of cyril rolando --test Big Chungus, style of Alex Grey --test Big Chungus, cosmic horror, style of HR Geiger --test
|
# ? Sep 27, 2022 22:05 |
|
Rutibex posted:Its really funny to go through individual people's profiles on midjourney. You can see every picture they made, as well as the order they are made in. Almost every profile alternates between fantastic arts and different forms of hot ladies or very buff men. You will get a few days worth of "exotic hanging garden, comic style" followed by a day of "pinup girl, large feet, made of yogurt" This is extremely true. BARONS CYBER SKULL posted:We should do a spooky season prompt-off when it rolls over to October Very much in favour of this.
|
# ? Sep 27, 2022 22:07 |
|
|
# ? May 28, 2024 23:41 |
|
AARD VARKMAN posted:
All of these prompts, with unholy blasphemy And with bagpipers
|
# ? Sep 27, 2022 22:08 |