|
SpaceCadetBob posted:a girl with headphones on in a neon city with neon lights and neon lights behind her is a neon background, Alena Aenami, synthwave style, cyberpunk art, computer art Base SD 1.5 is still great for some classical painting styles but really can't keep up compared to fine tuned models and model blends for anything modern or realistic. Telephone looks like it'll just go around in circles, but here's what I got out of the Interrogator webUI extension, probably same but different compared to the img2img interrogate button: quote:"woman wearing headphones in a neon city, cyberpunk art by Ilya Kuvshinov, cgsociety, computer art, girl wearing headphones, cyberpunk headset, beautiful cyberpunk girl face"  I spent way too long on this, faffing about with different models, samplers, CFG values, and using img2img for upscaling & loopback and I'm still not satisfied, AI artgen isn't for would-be perfectionists who lack editing skills. I ended up settling for these, though I still couldn't pick just one:   Perhaps running Stable Diffusion locally is for people who don't mind using Linux. The auto1111 webui has a very gentle learning curve, but it does have one if you're trying to get the results just as you want them. I ended up merging models to try and clear up some of the clutter that the first attempts were getting, with some success. Roman posted:This is definitely a project I want to make real, AI or not. Hope everyone can enjoy the adventures of my gals someday. I can imagine that at least some of the next generation of writers will be using AI artgen to help visualize/storyboard their works, plus the randomness of txt2img may help rattle the can when writer's block sets in.
|
#
?
May 8, 2023 15:53
|
|
|
|
| # ? May 29, 2024 01:05 |
|
|
The Sausages posted:These are amazing and look like a fun thing to make as well. https://twitter.com/PatrickJDoody/status/1651072860205764610
|
|
#
?
May 8, 2023 16:02
|
|
|
someone wanted a very specific image mostly because it was very specific and not very likely to be possible a manatee, a shrimp, and a lobster doing lines of cocaine in a brothel I forgot about the brothel but I think I got it    Beach Party, white powder on noses, manatee with raging biceps, shrimp with weak arms, lobster in the white powder, 4k, photograph, white powder on a silver tray
|
|
#
?
May 8, 2023 16:11
|
|
|
Roman posted:I did come across an industry guy on Twitter who was anti-AI but now uses it for pitch decks and for explaining stuff to his concept art people. Adobe Firefly is gonna fix some hearts.  Um..... I have some bad news about this dude.... Sure you can Opt Out, but how many people (including that writer) know they have to do so? 
|
|
#
?
May 8, 2023 17:35
|
|
|
I just want one generator that doesn't train on stolen work. AI is very interesting technology but too many folks are rushing head-first and not giving a single thought to how it's hurting others. I love goofing around with these things (not making a product out of them, though) and would like to do so with a clear conscience.
|
|
#
?
May 8, 2023 17:39
|
|
|
Inzombiac posted:I just want one generator that doesn't train on stolen work. Sorry, what's the difference between training on protected images and text? GPT doesn't train on stolen work. Why do the image generators?
|
|
#
?
May 8, 2023 17:45
|
|
|
https://www.reddit.com/r/StableDiffusion/comments/13bvbps/ive_created_200_sd_images_of_a_consistent/ Good-sounding guide for generating consistent characters with SD.
|
|
#
?
May 8, 2023 17:49
|
|
|
Well, the way it works now, the thing people consider "stealing," is everyone having access to the global dataset. The "ethical" version is large corporations owning their own gigantic datasets which individuals just can't compete with. The corps will own the copyrights to those datasets. But like what was posted, they may have stuff in those datasets they don't own the copyright for, but opting out and even knowing you have to opt out is the problem. So all you as an individual artist will have copyright for is your small Artstation portfolio, which will somehow have to compete with giant corps with giant datasets who can make art with the press of a button, and are probably gong to get way more business and be better at selling their content than you are. Although Disney probably won't be in the commissioned furry porn market, so you might be safe there. Also there's the issue of open source art, which is only going to get better, running on consumer hardware in someone's bedroom and eventually on phones when phones are powerful enough. And you have to somehow be able to tell with 100% accuracy that it's not copyright breaking AI art, so you can compete with that as well. Then there's even the questions of whether copyright and renting IP is a good thing. I'm not an educated man, so I'm not sure what the difference is between being a landlord/rent-seeker and owning copyright, and it confuses me why the people who say the former is bad support the latter.
|
|
#
?
May 8, 2023 17:56
|
|
|
nothing about the underlying technology is proprietary. all of the change is mere quantity of data. this is another instance of rich suttons bitter lesson. quantity cannot be negotiated away for getting quality from these models
|
|
#
?
May 8, 2023 18:09
|
|
|
         On some kind of "faded polaroid of a bittersweet lost love" sorta vibe idk.
|
|
#
?
May 8, 2023 18:11
|
|
|
Nothing is being stolen and anyone saying otherwise is either a moron or intentionally lying to you.
|
|
#
?
May 8, 2023 18:12
|
|
|
nothing is decided either way and anyone who says otherwise hasnt been paying attention to the lawsuits lol
|
|
#
?
May 8, 2023 18:14
|
|
|
Court decisions have no retroactive impact on objective reality.
|
|
#
?
May 8, 2023 18:16
|
|
|
the definition of theft cannot be objective reality, come on and unless you got ahold of a lotta guns artillery nukes and men lately the state has a say in what theft is too
|
|
#
?
May 8, 2023 18:21
|
|
|
If you want to wring your hands and debate the ethics of AI art, kindly go do it somewhere else. That's not what this thread is for, and making GBS threads up the thread with the same tired argument every 10 pages ain't it.
|
|
#
?
May 8, 2023 18:24
|
|
|
PlaysGamesWrong posted:Nothing is being stolen and anyone saying otherwise is either a moron or intentionally lying to you. My question was just whether the copyright laws are different in letters and art, because people are talking IP issues for visual art and the LLM people seem to have best practices in place that everyone so far follows, and nobody talks about "stealing." (Not my term.)
|
|
#
?
May 8, 2023 18:31
|
|
|
Not a debate thread.
|
|
#
?
May 8, 2023 18:34
|
|
|
bing does some fun stuff with "alien potatoes" 
|
|
#
?
May 8, 2023 18:39
|
|
|
today I shall be doing polaroids 
|
|
#
?
May 8, 2023 18:40
|
|
|
AARD VARKMAN posted:today I shall be doing polaroids 
|
|
#
?
May 8, 2023 18:44
|
|
|
The Sausages posted:Telephone looks like it'll just go around in circles Possibly! No way of knowing what happens until you try though! Inzombiac posted:I just want one generator that doesn't train on stolen work. Got a week and $50k? You're in luck! Back in January they posted it took 13 days and $160k and in less than 5 months optimizations have brought it down to half the time and a third the cost. I have no idea how much further it will come down in another 5 months, probably in the range a small studio could easily do. This is for training billions of images, an important thing a lot of people are finding is, the largest dataset isn't always the best. It's looking more and more like small bespoke models laser focused on whatever topic out perform large general models by far. A much smaller set trained on 'just' millions of images is going to be in the range of everyone "soon" (tm). The distributors are the ones crying the loudest, the middle-men farthest from whoever pushes pencil or pen or brush. They are the ones demanding a dollar before they pass on a few cents to the actual artists. Never forget that. If you need it, I personally give you a worldwide unrevocable royalty-free license to just have guilt-free fun. Roman posted:Then there's even the questions of whether copyright and renting IP is a good thing. I'm not an educated man, so I'm not sure what the difference is between being a landlord/rent-seeker and owning copyright, and it confuses me why the people who say the former is bad support the latter. https://questioncopyright.org/promise This is a longer piece on the history of censorship and the origins of copyright. It's well worth a read because the more things change, the more they stay the same. bob dobbs is dead posted:nothing is decided either way and anyone who says otherwise hasnt been paying attention to the lawsuits lol You mean the lawsuits that might get thrown out because the people suing didn't register their works with the copyright office so no infringement actually occured? lol.
|
|
#
?
May 8, 2023 18:53
|
|
|
theres like 6 ongoing ones of very varying legal merit yes
|
|
#
?
May 8, 2023 18:57
|
|
|
AARD VARKMAN posted:today I shall be doing polaroids I don't know if anyone can top that 70s goon there drat.                                
|
|
#
?
May 8, 2023 19:09
|
|
|
Ok, no more debate, except maybe debating why AI art, which I keep being told is a magic Do Anything You Want button no human can compete with, has problems making me a picture of a girl in a futuristic exosuit in a Vogue photoshoot. One of my characters wears a time travel suit. MJ can crank out amazing looking stuff all day just by mentioning "exosuit." Example:  So I thought it would be amusing to have her wearing her time travel exosuit in a glamorous Vogue photoshoot. Stole a prompt off Twitter, changed it, and this happened: glamour photo, medium overhead shot from above, sharp focus low-light photography shot on Fujifilm, capturing Zoey Deutch with tired eyes resting on a tufted couch, wearing a high-tech exosuit, luxurious Parisian mansion, dim chandelier lighting, elegant, glamorous, sophisticated. It's been a long night. --ar 16:9 --v 5  Ok, what the gently caress. Got a bunch of those while playing with prompts, this was the best I could get: capturing Zoey Deutch wearing a high-tech exosuit. Zoey has tired eyes. Background is a luxurious Parisian mansion, dim chandelier lighting. medium overhead shot from above, sharp focus low-light photography shot on Fujifilm. --ar 16:9 --v 5  So they're "fancy" exosuits, and some like they're made of photography equipment like you would see in a behind the scenes of a Vogue photoshoot. It keeps trying to make stuff from the same little Vogue world, instead of combining things from two different settings. MJ 5.1 with default and with --style raw is a little better, because they at least look sci-fi, but still "fancy" and not like the one in the very first image.   tried specifying styles of No Man's Sky and Fallout Power Armor. Still not right, still "fancy."    So while the pics are interesting, I would still have to get better at using PS and/or find an artist to actually make the specific thing I wanted.
|
|
#
?
May 8, 2023 19:16
|
|
|
AARD VARKMAN posted:today I shall be doing polaroids     
|
|
#
?
May 8, 2023 19:20
|
|
|
Still trying AI Telephone, if I can get to at least 16 I'll attempt animating them all. polaroid photo of woman wearing headphones in a neon city, cyberpunk art by Ilya Kuvshinov, cgsociety, computer art, girl wearing headphones, cyberpunk headset, beautiful cyberpunk girl face Steps: 48, Sampler: DPM++ 2M Karras, CFG scale: 11, Seed: 3981350073, Size: 512x512, Model hash: 7f16bbcd80, Model: dreamshaper_4BakedVae, Denoising strength: 0.69, Hires upscale: 1.5, Hires steps: 48, Hires upscaler: 4x-UltraSharp
|
|
#
?
May 8, 2023 19:28
|
|
|
Roman posted:
You really need to combine with Stable Diffusion to tweak your images. For just firing off 'lets do polaroids lmao' MidJourney is great, and for getting baselines it's also great, but it's extremely restrictive. The reason it doesn't want to power armor your posing photoshoot woman is because women don't wear power armor in vouge photoshoots, they wear pretty gowns and lounge on sofas.  This is one single pass after cropping your image in Stable Diffusion via inpainting and 'Realistic Vision V20' checkpoint, using Auto1111. I say this because I did no effort at all. You can go so much further if you just reach in and tweak it a bit, change their facial expressions or whatever too. It's incredibly powerful for photo editing KakerMix fucked around with this message at 19:44 on May 8, 2023 |
|
#
?
May 8, 2023 19:41
|
|
|
Did you just add in the Polaroid bit manually? I don't know if we have the users needed to get that high without people taking multiple turns, which feels like something you don't do in telephone? I may have misunderstood.
|
|
#
?
May 8, 2023 19:41
|
|
|
PlaysGamesWrong posted:Did you just add in the Polaroid bit manually? I did. I think misunderstanding is integral to the telephone game lol. I just want enough images that I can mess with this RIFE framegen animation software.
|
|
#
?
May 8, 2023 19:47
|
|
|
At first I thought the idea of telephone was use image to prompt to get a prompt from the image in various programs and generate away letting the next person do the same in their software. There's no way to do it in Bing, but I thought MJ had a way, I know SD has 2 ways (at least) and they produce different results and use different methods.
|
|
#
?
May 8, 2023 20:10
|
|
|
pixaal posted:At first I thought the idea of telephone was use image to prompt to get a prompt from the image in various programs and generate away letting the next person do the same in their software. MJ uses the describe function
|
|
#
?
May 8, 2023 20:15
|
|
|
Roman posted:Ok, no more debate, except maybe debating why AI art, which I keep being told is a magic Do Anything You Want button no human can compete with, has problems making me a picture of a girl in a futuristic exosuit in a Vogue photoshoot. I kind of liked the power armor photoshoot concept....   The armor is still more dainty and fashionable but I also didn't try to tweak things very hard. Edit: made a couple more attempts  
hydroceramics fucked around with this message at 20:37 on May 8, 2023 |
|
#
?
May 8, 2023 20:23
|
|
|
hydroceramics posted:I kind of liked the power armor photoshoot concept.... I like this set looks like a scene of someone waiting all day in the waiting room to see someone, including a part where they take a nap. e: oh I made images recently that are neat I guess. half grass, half dog, Polaroid photo  
pixaal fucked around with this message at 20:36 on May 8, 2023 |
|
#
?
May 8, 2023 20:25
|
|
|
Roman posted:Ok, no more debate, except maybe debating why AI art, which I keep being told is a magic Do Anything You Want button no human can compete with, has problems making me a picture of a girl in a futuristic exosuit in a Vogue photoshoot. Start with this prompt: Zoey Deutch posing for a photoshoot wearing a mechanical high-tech exosuit, tired eyes and mood lighting, lounging on a leather chaise, photographed in a parisian mansion in the evening, full length portraiture --ar 3:2 To get this in a grid:  Using the variation button with remix mode on in Midjourney, change the prompt to: Zoey Deutch wearing a armored mechanical high-tech exosuit, tired eyes and mood lighting, lounging on a leather chaise, photographed in a parisian mansion in the evening, full length portraiture --ar 3:2  Remix the variation again: Zoey Deutch wearing a armored mechanical high-tech exosuit with orange elements, tired eyes and mood lighting, lounging on a leather chaise, photographed in a parisian mansion in the evening, full length portraiture --ar 3:2  or Zoey Deutch wearing a heavily armored mechanical high-tech exosuit with hunter orange accents, tired eyes and mood lighting, lounging on a leather chaise, photographed in a parisian mansion in the evening, full length portraiture --ar 3:2  or Zoey Deutch wearing bulky and heavily armored mechanical high-tech exosuit, the exosuit is hunter orange and gunmetal grey, tired eyes and mood lighting, lounging on a leather chaise, photographed in a parisian mansion in the evening, full length portraiture --ar 3:2  In short, set up the pose/setting with the first prompt, then use remix to remove keywords that seem to have weighting in the direction you're not really aiming for (in this case, removing "posing for a photoshoot" tones down the fanciness it seems, allowing some flexibility with the suit). These could still be improved upon, but the basis for a better exosuit seems to be there.
|
|
#
?
May 8, 2023 21:01
|
|
|
Got something reasonable in Bing with "woman wearing lightweight power armor, laying on the sofa, bulky attachments" that last bit on bulky attachments doesn't seem to be much but something should be there the first too feel like they are missing something. Maybe just flagging some lighting or camera direction would be enough. Oddly it *really* likes to put feet in the image set.
|
|
#
?
May 8, 2023 21:10
|
|
|
 The only prompt I gave it was "Ron Desantis".. 
|
|
#
?
May 8, 2023 21:18
|
|
|
Soulhunter posted:In short, set up the pose/setting with the first prompt, then use remix to remove keywords that seem to have weighting in the direction you're not really aiming for (in this case, removing "posing for a photoshoot" tones down the fanciness it seems, allowing some flexibility with the suit). These could still be improved upon, but the basis for a better exosuit seems to be there.
|
|
#
?
May 8, 2023 21:22
|
|
|
Roman posted:This is great, thanks! I'm not great at prompts so I need all the help I can get. Hope it helps, I've done an absolutely stupid amount of Midjourney generations, but I still feel like a novice skimming the surface most of the time. There's so much to learn, so many styles to utilize, and the tech is just rapidly evolving. Very rewarding though. PlaysGamesWrong posted:The only prompt I gave it was "Ron Desantis".. NOP!  Who would've known Ron participates in the republican drag entertainers club as a side gig? DeSantis   McConnell   Cruz    Trump   Graham (Midjourney really did not want to attempt a drag outfit for Graham)   and the best for last, Giuliani  
|
|
#
?
May 8, 2023 21:45
|
|
|
Roman posted:This is great, thanks! I'm not great at prompts so I need all the help I can get.  This is the basic idea of ordering for strength. Whichever token is first gets most emphasis. I don't know if Midjourney allows token weighing like Stable Diffusion does, (token:1.2) would be a 20% taller starting curve with longer fall off for instance.
|
|
#
?
May 8, 2023 21:48
|
|
|
|
| # ? May 29, 2024 01:05 |
|
|
the Strange Head lora is very cool with wildcards:     
|
|
#
?
May 8, 2023 21:59
|
|
