|
Boba Pearl posted:My Prompt Where is the best place to read up on AI for text stuff I'm sitting on a super rare really interesting piece of polish sci fi social commentary from the 70s that's kind of been lost to time and history and was never translated to English (author died 20+ years ago too*). I transcribed it to ASCII/Unicode and ran it though Google translate, about 200 lines at a time (you get rate limited eventually, and it's a 250 page book) and so you can sort of read it but it's super rough 1) would like to maybe re-translate it from the original polish 2) feed it through 2-3 AI, then play "pick the best phrase/sentence/paragraph" between them all 3) publish it for others to read*, ** *doesn't look like any of his work was ever reprinted except his one or two most popular novels, and the family hasn't really attempted to monetize his body of work, if they even know they technically own the rights since he was like a c level author at his peak
|
#
?
Nov 4, 2022 07:25
|
|
|
|
| # ? May 29, 2024 18:37 |
|
|
Rutibex posted:i trained an AI to be Gary Gygax to give me advice for my D&D campaign. hes pretty good What tools did you use to accomplish this
|
|
#
?
Nov 4, 2022 07:55
|
|
|
It's character.ai
|
|
#
?
Nov 4, 2022 08:12
|
|
cowboy
|
https://www.youtube.com/watch?v=3ApMSKql23I
|
|
#
?
Nov 4, 2022 08:40
|
|
|
sadly, seems like the 'sprite sheet' model is basically just overtrained on stardew valley sheets. I tried making some and was pretty unimpressed    two of these are supposed to be batman, and two are supposed to be mario. Since that's an extremely rigid and constrained design space, they all end up looking very samey. You honestly don't need a NN to get better results than this to make this sort of sprite algorithmically, a paperdoll style system with a few tweaks and randomized hues would get more consistent results.
|
|
#
?
Nov 4, 2022 08:45
|
|
|
Sure, but it's more of a proof of concept. In a couple of years, I suspect you'll actually be able to produce enough assets with AI tools to make a game with them.
|
|
#
?
Nov 4, 2022 11:10
|
|
|
Hadlock posted:Where is the best place to read up on AI for text stuff Don't know about the other stuff, but use DeepL for the first draft translation. It tends to be quite a bit better than Google. Like, here's the first paragraph of Strugatskiy brothers' The Doomed City translated directly from Russian. DeepL posted:The tanks were rusty, wrinkled, with loose lids. There were scraps of newspaper sticking out from under the lids, potato peels hanging down. It looked like the jaws of an unkempt, indiscriminate pelican. They looked unbearably heavy, but, in fact, the two of them, together with Van, would have been able to pull one up to Donald's outstretched arms and set it on the edge of the reclining side. All they had to do was to guard their fingers. Then they could fix their mittens and breathe through their noses a little while Donald tugged the tank and set it in the back of the body. Google posted:The tanks were rusty, dented, with lagging lids. Scraps of newspapers stuck out from under the lids, potato husks hung down. It looked like the mouth of an untidy, unscrupulous pelican. They looked unbearably heavy, but in fact, together with Van, it was not worth anything to jerk such a tank up to Donald's outstretched hands and fix it on the edge of the folded side. I just had to watch my fingers. After that, you could straighten your mittens and breathe a little through your nose while Donald turns the tank, setting it in the back of the body.
|
|
#
?
Nov 4, 2022 11:25
|
|
|
There are hard limits to translation AI, which can't really be solved with current AI technology. It can really be good at getting the gist of the text, but without really being able to use context, its going to miss a lot and be inaccurate. This is probably going to be especially true for a work of Sci-Fi which introduces new concepts. There are a lot of people who try to use AI to translate novels and the results are not good and you can always tell.
|
|
#
?
Nov 4, 2022 13:15
|
|
|
I think the AI art boom is already starting to hit the lower rungs of artists who do commissions. Got cold messaged by one on Discord a couple of days ago: That was the day I figured out how to run SD on my PC, so in the next window on my desktop I'm, like, already cranking out gigs of portraits for my OCs. I was overcome with such pity I didn't even report her to the server.
|
|
#
?
Nov 4, 2022 13:25
|
|
|
That Arcane model is a loving treasure:           
Megazver fucked around with this message at 13:35 on Nov 8, 2022 |
|
#
?
Nov 4, 2022 17:07
|
|
|
Then I decided to test how many different characters I could squeeze out of a single actor, so I went Full Hamm:               
|
|
#
?
Nov 4, 2022 20:52
|
|
|
"my latest steam profile avatar"
|
|
#
?
Nov 5, 2022 00:41
|
|
|
edit: nm im stupid
tote up a bags fucked around with this message at 08:21 on Nov 5, 2022 |
|
#
?
Nov 5, 2022 08:14
|
|
|
tote up a bags posted:These are insane, love it. I'm trying to set this up locally but I can't seem to find a ckpt file on the huggingface page, if i clone the git i don't get one either - is there a way to generate one? Here's the official Stable Diffusion 1.5 model: https://huggingface.co/runwayml/stable-diffusion-v1-5 If you're not going to do your own training, you can get the smaller v1-5-pruned-emaonly.ckpt file.
|
|
#
?
Nov 5, 2022 08:22
|
|
|
Victorian Foxes eating pizza        Bonus: Fish eating pizza  
|
|
#
?
Nov 5, 2022 12:24
|
|
|
Megazver posted:Then I decided to test how many different characters I could squeeze out of a single actor, so I went Full Hamm: These are from the 1.4 model I think... Steampunk Jon Hamm:   Demonic Cara Delevigne:  Cyberpunk Olivia Wilde:  Steampunk Tom Holland:  Cyberpunk Vanessa Hudgens: 
|
|
#
?
Nov 5, 2022 12:59
|
|
|
Hmm https://twitter.com/Valdevia_Art/status/1588875453410205698?t=d_rgFqqdDfvd6wopBpDEbw&s=19
|
|
#
?
Nov 5, 2022 14:50
|
|
|
Man, not a single model knows how to have a single cataract eye or a big scar through the eye. What the hell.
|
|
#
?
Nov 5, 2022 15:43
|
|
|
Dick Trauma posted:These are from the 1.4 model I think... This one is p.good
|
|
#
?
Nov 5, 2022 16:45
|
|
|
TheWorldsaStage posted:Hmm this is why I have been hammering midjourney nonstop to get as much AI stock art as possible. someone's gonna come for the AI art services soon, i can feel it
|
|
#
?
Nov 5, 2022 17:17
|
|
|
 https://twitter.com/JanelleCShane/status/1588923299505274881?s=20&t=g81B35BeerXnl4eVO69C6g
|
|
#
?
Nov 5, 2022 17:41
|
|
|
Rutibex posted:this is why I have been hammering midjourney nonstop to get as much AI stock art as possible. someone's gonna come for the AI art services soon, i can feel it locally hosted SD stuff, cold dead hands, etc
|
|
#
?
Nov 5, 2022 20:09
|
|
|
frumpykvetchbot posted:locally hosted SD stuff, This is kind of the big thing about trying to legislate AI art is that so much of it can be done without external services and it got released to the public before the corporations could get their hands on it, so I really don't see how it could be pushed back against in anyway. Maybe a law banning commercial use of AI art, but that's about it.
|
|
#
?
Nov 5, 2022 21:08
|
|
|
Fuzz posted:Man, not a single model knows how to have a single cataract eye or a big scar through the eye. What the hell. It should work if you prompt a popculture character known for a messed up eye. That'll bring its own problems though, since it's hard to override their appearance with other descriptions.
|
|
#
?
Nov 5, 2022 21:17
|
|
|
Vlaphor posted:This is kind of the big thing about trying to legislate AI art is that so much of it can be done without external services and it got released to the public before the corporations could get their hands on it, so I really don't see how it could be pushed back against in anyway. Maybe a law banning commercial use of AI art, but that's about it. Then you get to argue about what is AI art or not. How can you tell? Are we going to have art inspectors everywhere making sure there is no ILLEGAL PROMPTING happening? There is no AI finger print or talent trail to follow. That we can't tell means it's absolutely unenforceable no matter what. It's absolutely way beyond even thinking about containing, and AI art has already infiltrated and been commercialized and we all see it regularly and don't even realize.
|
|
#
?
Nov 5, 2022 22:06
|
|
|
Oh dear god Midjourney v4 is delicious! It is better than Dalle 2, this is amazing i'm sorry I ever doubted you Midjourney. I have saved all of my "fast" time like elixirs at the end of final fantasy, and my account renews tomorrow. So today I get to use all of my months credits on the new version 4  First thing I tried was to fancy up my D&D character. v4 (unlike test version) supports image prompts. It worked amazingly! The image I am using as a prompt is the same as my avatar here on SA. I used a clip interrogator AI to get some good keywords to use for the text prompt, but it works really well with stuff I just make up too!    Bonus, some elegantly dressed animals eating pizza version 4 style     
|
|
#
?
Nov 5, 2022 22:07
|
|
|
Do you have, like, any comparisons for same seed and prompt, v3 vs v4? I don't know if that's possible. But yeah, despite getting into SD I still really like Midjourney. Hope they make their model available to run on your own PC one day. On another note, I found an amusing blind spot in SD. For some strange reason, Grace Jones is just too freaky for it to paint properly. It does other attractive black ladies just fine, but Grace Jones is like "woah WOAH WOAH, my human friend, what the hell is THAT?!"   vs Naomi Campbell - 4 out of 4 are pretty good. 
Megazver fucked around with this message at 22:28 on Nov 5, 2022 |
|
#
?
Nov 5, 2022 22:18
|
|
|
Megazver posted:Do you have, like, any comparisons for same seed and prompt, v3 vs v4? I don't know if that's possible. I think there is a setting for the same seed but I don't know exactly how to use that! I just did a bunch and picked the best looking ones as a more fair comparison. The prompt is "lisa ann as buffy the vampire slayer full body concept art" Midjourney v3 this feels very primative going back to it!    Midjourney v4    Midjourney Test (I think this is Stable Diffusion?)   Edit: My old friend Junji Ito in the Style of Jon Burgerman is giving some very interesting results in v4!                  
Rutibex fucked around with this message at 23:16 on Nov 5, 2022 |
|
#
?
Nov 5, 2022 22:47
|
|
|
(cyberpunk) daguerreotype is a powerfully weird prompt, even if you have to add some negatives to cleanse the neon.        Tempted by that dog av.
|
|
#
?
Nov 6, 2022 02:26
|
|
|
Midjourney v4 is released now. To celebrate, here's "Guy Fieri as a WH40k Ultramarine --v 4"
|
|
#
?
Nov 6, 2022 10:41
|
|
|
Midjourney v4 is very good for turning your friends into characters and also putting them on the moon and in hell          
|
|
#
?
Nov 6, 2022 11:28
|
|
|
Rutibex posted:Midjourney v4 is very good for turning your friends into characters and also putting them on the moon and in hell Is that just using their photo as a image prompt? How does it actually work, what's the prompt? i tried doing that with MJ and the results are... not the same.
|
|
#
?
Nov 6, 2022 11:46
|
|
|
Megazver posted:Is that just using their photo as a image prompt? How does it actually work, what's the prompt? i tried doing that with MJ and the results are... not the same. Yeah, every one of those pics is the exact same image prompt. For text I put in " walking on the moon" or "venom symbiote" or whatever and this is the result. Make sure you have it set to version 4
|
|
#
?
Nov 6, 2022 12:08
|
|
|
Yeah, hot drat, MJ4 is phenomenal at image prompts. My cat:  MJ's take on that photo:   
|
|
#
?
Nov 6, 2022 12:46
|
|
|
Killer robot posted:(cyberpunk) daguerreotype is a powerfully weird prompt, even if you have to add some negatives to cleanse the neon. Wow! I love these    
|
|
#
?
Nov 6, 2022 17:13
|
|
|
(skipping ahead like 50 pages, I've been busy) I use SD but Midjourney's community Showcase is looking more ArtStation level now https://www.midjourney.com/showcase/  This latest upgrade seems to be a big deal. I hope SD doesn't fall too far behind, I can clearly see a difference in the art quality now. I like SD because it was stupid easy to install locally, but Midjourney looking good
|
|
#
?
Nov 6, 2022 18:46
|
|
|
Yeah I would say Midjourney v4 is the best AI available right now. Its only PG-13 and has word filters and no in-painting, but its the smartest one and understands the prompts a lot better. It does hands and faces really well, and can even do multiple subjects in the same picture without too much issue. Best of all it does absolutly fantastic pixel art. Its as good as Dall-e 2, which up until now was the only AI which could do decent pixel art (and its super expensive!) "Mr. Angry and All His Favorite Shapes, 1985, punk album cover, bizarre album covers"  "an oil painting of the old gods walking amongst us"  "the love of my life wrestling a bear in oil"   "a young man praying to Satan in an old abandoned dilapidated derelict barn with a stone altar and moonlight streaming through the barn doors, smoke, mist, demons"  "a crowd of muscular drunk gay guys in a locker room dancing to Madonna"  "Gorgeous girl demon cooking a steak on a fantasy BBQ in Hell, massive chest, Norman Rockwelll, boris vallejo"  "drunk old man with a shaved head and white beard drinking whiskey sitting at a typewriter, moody, gloomy, bright moonlight streaming in the window, full moon, candles[monochrome, chiaroscuro]"  "two rural policeman passionately kiss while beating a man to death"  "old drunken punk rocker in a bar watching an old punk band fall off the stage in wheelchairs"  "Lemmy with his pet cowboy raccoon, post-apocalyptic cowboy town, black cowboy hat, chaps, leather vest"  "a mountain of human skulls is on fire while a muscular barbarian stands beside it, perfect abs, long black hair, clean shaven, holding a sword"  "pixel art witch" 
|
|
#
?
Nov 6, 2022 19:26
|
|
|
Have any of you seen this MagicPrompt thing? It's a text model trained on the data from lexica.art and krea.ai etc to generate stable diffusion prompts.  This plugin script has an option to use it locally in your prompt generations. I put in "a masterpiece fantastical magical nighttime landscape painting, landscape painting of" and let the magicprompt thing complete it:     For example, this one was "a masterpiece fantastical magical nighttime landscape painting, landscape painting of a hazy and cloudy cloudy cloudy sky with various low poly mountain buildings and waterfalls flying in the sky, award winning landscape painting by Alex Grey of a vibrant royal blue and gold royal"  And this one was "a masterpiece fantastical magical nighttime landscape painting, landscape painting of a hazy and rainy mega city with a european style tower and golden ornaments in the sky, archdaily, wallpaper, highly detailed, trending on artstation."  It's a neat way to let the AI surprise you and get some random variety. Of course it adds gets totally different subject and artstyle keywords too so it's not too easy to get it to stay on theme. Here was "A portrait of Danny Devito" plus additions from MagicPrompt in the trinart-characters model: 
|
|
#
?
Nov 6, 2022 19:51
|
|
|
the new "how would you look like as a human, MJ" MJ Girl: In the previous version of MJ: 
Megazver fucked around with this message at 22:59 on Nov 6, 2022 |
|
#
?
Nov 6, 2022 21:19
|
|
|
|
| # ? May 29, 2024 18:37 |
|
|
As someone who hasn't been able to play around with too much cutting-edge stuff because I have a crappy computer and am too dumb to figure any of the more complicated stuff out, MidJourney v4 is blowing my goddamn mind. What's the best in style transfer these days? If I put some time into it, would I be able to reasonably transfer a portrait photograph into the style of, say, Robert McGinnis or Mort Künstler? Or is that sort of thing still difficult for it?
|
|
#
?
Nov 7, 2022 02:32
|
|