|
IShallRiseAgain posted:F*** off! wow I would watch this show
|
#
?
Jan 27, 2023 05:26
|
|
|
|
| # ? May 30, 2024 08:58 |
|
|
I can't believe someone murdered Charlie Kirk!
|
|
#
?
Jan 27, 2023 09:28
|
|
|
This seems pretty interesting to generate concepts. This is all pretty new to me but could you use these to generate sci-fi concepts? Like spaceships and the like? If so, what models would I need to use to get some good results? It doesn't have to be photoreal, but interesting shapes would be awesome. The examples you guys posted are pretty mindblowing (the planet stuff).
|
|
#
?
Jan 27, 2023 10:29
|
|
|
Mr Shiny Pants posted:This seems pretty interesting to generate concepts. This is all pretty new to me but could you use these to generate sci-fi concepts? Like spaceships and the like? If so, what models would I need to use to get some good results? https://civitai.com/models/4775/planit-a-documentation-embedding https://civitai.com/models/4897/concept-sheet https://civitai.com/models/5271/style-info-an-embedding-for-infographic-style-art
|
|
#
?
Jan 27, 2023 11:18
|
|
|
Mr Shiny Pants posted:This seems pretty interesting to generate concepts. This is all pretty new to me but could you use these to generate sci-fi concepts? Like spaceships and the like? If so, what models would I need to use to get some good results? literally just type "a spaceship" into the prompt. if you get sailing ships in space say "a spacecraft." iterate from there
|
|
#
?
Jan 27, 2023 16:26
|
|
|
Good pieces are about not describing what you want always especially if you just want a set on a theme. Throw related words into the positive prompt leave the negative blank except maybe like captcha erotic or something else since it does better with something in there. If you see a lot of something you don't want add it to the negative if you want more of something add another related word to the stack. If it looks like poo poo add quality increasing words to the postive and quality reducing to the negative (this takes some playing with and will change the type of pictures you are getting) things like highly detailed, fine details, 4k, 8k, HDR, Raytracing, bad, lovely, ugly, burned, damaged remember things like burned and damaged don't just describe the work itself but may also describe items inside. If you are going for a battle damaged ship you need to be careful to get a good quality image with a damaged ship and not like a picture of a ship that has been burned and has a coffee stain. What I like to do is add words one at a time and if one I want is reducing quality I try synonyms or other ways to go about asking for that. You sometimes find magic words which have extremely high quality images associated with them primarily. Usually a concept that is new. Vantablack is an excellent quality word *If* you want a dark image. Your blacks are going to be maxed out. bioluminescence is a great word if you want glowing things but it does tend to make the lighting blue and pink. If you want blue and pink but not glowing gender reveal is the keyword for you. So I used P: space craft, drive plume, battle damage, solar system, asteroid belt, highly detailed, 4k, scarred N: captcha, out of focus, blurred  added bioluminescence to the positive  removed that from the positive and added to the negative lowpoly, low poly  adding bio back into the positive  combined I think it's losing the subject of the ship at this point. There's a balance but hopefully this gives you an idea of how to find which words to add where. This isn't an attempt to be exhaustive or how to get great images from your first generations. Anything that's really good looking is someone figuring out a good set of words and slamming them. You will eventually have some favorite setups I really like how it doesn't seem to hurt anything by adding captcha and makes words way more readable if they show up.
|
|
#
?
Jan 27, 2023 17:47
|
|
|
Tried to get a picture of Stone Cold Steve Austin, and I think MJ took the cold part too literally lmao 
|
|
#
?
Jan 27, 2023 17:48
|
|
|
That's going to be fun to get out, negative of ice winter and snow for starters? MJ supports negative right? I'm using I think SD 2.1 path says the folder it's running it from is called 2.1 768
|
|
#
?
Jan 27, 2023 17:52
|
|
|
KakerMix posted:https://www.youtube.com/watch?v=G08hY8dSrUY I really don't get the explanation at 6:00. Yes, its copyright infringement, but ppl aren't jerking off to the thumbnails so its fine?
|
|
#
?
Jan 27, 2023 18:01
|
|
|
unzin posted:I really don't get the explanation at 6:00. Yes, its copyright infringement, but ppl aren't jerking off to the thumbnails so its fine? Like the video touches on, the difference between "straight copyright infringement" and "copyright infringement but it's okay because it qualifies as fair use" is basically whatever lawyers can convince the judge of/wring a settlement out of. IP Law is Very Dumb.
|
|
#
?
Jan 27, 2023 18:06
|
|
|
pixaal posted:That's going to be fun to get out, negative of ice winter and snow for starters? MJ supports negative right? I'm using I think SD 2.1 path says the folder it's running it from is called 2.1 768 The previous v.3 did. V.4 didn't allow negative weights at first but might by now. I just got a good chuckle from him looking like Mr Freeze.
|
|
#
?
Jan 27, 2023 18:08
|
|
|
                 
|
|
#
?
Jan 27, 2023 18:09
|
|
|
unzin posted:I really don't get the explanation at 6:00. Yes, its copyright infringement, but ppl aren't jerking off to the thumbnails so its fine? Literally yes, the thumbnails were used as a directory, in an informational sense. Google is (was  ) all about connecting people with what they were looking for, so they uses copyright infringing little copies of the magazines to facilitate people's searches. That's what the court decides. This area of the law is (as has already been described) very silly and dumb because drawing lines around creativity because of capitalism is very dumb and you get weird poo poo all the time decided by people arguing in a court. This also explains why 'copyright infringement' doesn't actually say much without further details, i.e. fair use or derivative. Context of the infringement is everything, not infringement itself.
|
|
#
?
Jan 27, 2023 18:11
|
|
|
pixaal posted:That's going to be fun to get out, negative of ice winter and snow for starters? MJ supports negative right? I'm using I think SD 2.1 path says the folder it's running it from is called 2.1 768 Yeah, MJ uses '--no (what you want removed)' and you can stack those but you have to add the '--no' prefix to each negative
|
|
#
?
Jan 27, 2023 18:54
|
|
|
How long until we have access to the ultimate bot? The bot that doesn’t argue with you and will answer any question truthfully no matter how unethical? We’re due for another paradigm shift.
|
|
#
?
Jan 27, 2023 18:59
|
|
|
If you work for one of the companies that back OpenAI, right now!
|
|
#
?
Jan 27, 2023 19:03
|
|
|
Nigmaetcetera posted:How long until we have access to the ultimate bot? The bot that doesn’t argue with you and will answer any question truthfully no matter how unethical? We’re due for another paradigm shift. Lawyers have existed for hundreds of years.
|
|
#
?
Jan 27, 2023 19:13
|
|
|
Humbug Scoolbus posted:Yeah, MJ uses '--no (what you want removed)' and you can stack those but you have to add the '--no' prefix to each negative pretty sure you can just --no [something], [something], [something]
|
|
#
?
Jan 27, 2023 19:16
|
|
|
KillHour posted:Lawyers have existed for hundreds of years. Lawyers still refuse to roleplay my anal vore scenarios though
|
|
#
?
Jan 27, 2023 19:17
|
|
|
pixaal posted:That's going to be fun to get out, negative of ice winter and snow for starters? MJ supports negative right? I'm using I think SD 2.1 path says the folder it's running it from is called 2.1 768 With MJ I think all you'd need to do is either call him "Stone Cold" or "Steve Austin".* "Stone Cold Steve Austin" can easily be interpreted as a Steve Austin who is stone cold Or "Stone Cold(WWE)" or something maybe. *Of course it all depends on the rest of the prompt deep dish peat moss fucked around with this message at 19:21 on Jan 27, 2023 |
|
#
?
Jan 27, 2023 19:19
|
|
|
KillHour posted:Lawyers have existed for hundreds of years. My lawyer won’t tell me how to make ricin, in fact he’ll lecture me about why I shouldn’t it at length. I hope ChatGPT is such a paradigm-shifting game-changer that the very concept of education has to be utterly bulldozed and rethought from the bottom up. Hopefully in a way that’s more conducive to me learning in an institutional setting. Sorry for any inconvenience this causes y’all educators.
|
|
#
?
Jan 27, 2023 19:20
|
|
|
Dreambooth model I'm working on, I don't really know what I'm doing, here it is at 30k steps on 576 images. DND_Battlemap, a battlemap interior of a tavern with tables, chairs, a bar, inn  DND_Battlemap, an isometric battlemap of a dungeon with chains, cages, and a bubbling cauldron as the centerpiece, high quality, best, beautiful, excellent  DND_Battlemap, top down map of a wizards tower with grid  DND_Battlemap, top down map of a wizards tower interior with grid 
|
|
#
?
Jan 27, 2023 19:22
|
|
|
Thanks, this is all pretty slick. I'll give it a shot. You running it on your own machines?
|
|
#
?
Jan 27, 2023 19:24
|
|
|
Nigmaetcetera posted:My lawyer won’t tell me how to make ricin, in fact he’ll lecture me about why I shouldn’t it at length. If you wouldn't enter it in Google, I don't suggest entering it in ChatGPT either. You can ask, and you can find out, but the FBI will subpoena your questions.
|
|
#
?
Jan 27, 2023 19:28
|
|
|
KillHour posted:If you wouldn't enter it in Google, I don't suggest entering it in ChatGPT either. You can ask, and you can find out, but the FBI will subpoena your questions. A theoretical one that will run local though you could ask anything deny it access to the internet then snuff it out before connecting anything that touched it to the internet. I think that's the fear of a local assistant AI for the masses. Because once it's out that version that will tell you how to make dangerous stuff, will always be out there someplace. The dataset will age but chemical reactions are always the same.
|
|
#
?
Jan 27, 2023 19:30
|
|
|
KillHour posted:If you wouldn't enter it in Google, I don't suggest entering it in ChatGPT either. You can ask, and you can find out, but the FBI will subpoena your questions. Omfg the FBI doesn;t care about me, I know about this for a fact.
|
|
#
?
Jan 27, 2023 19:35
|
|
|
You can already run the PygmalionAI chatbot locally. You of course need a beefy Nvidia card (so us AMD users are still out of luck) and its current best model is only 6 billion whatevers (which is paltry compared to even CharacterAI, much less GPT-3), but you can do it.
|
|
#
?
Jan 27, 2023 19:36
|
|
|
So I've been tuned out of the AI art stuff for a few months but want to give it another shot, especially since folks seem to be talking about a bunch of new models and forks of SD that look pretty capable. What are the best ones to try right now?
|
|
#
?
Jan 27, 2023 23:29
|
|
|
Is there a chatbot thread or is this it? I haven’t been following along for a couple months now.
|
|
#
?
Jan 27, 2023 23:38
|
|
|
Snowy posted:Is there a chatbot thread or is this it? I haven’t been following along for a couple months now. There is a Seinfeld one, but its fine posting here too. https://forums.somethingawful.com/showthread.php?threadid=4019584
|
|
#
?
Jan 27, 2023 23:57
|
|
|
Mr Luxury Yacht posted:So I've been tuned out of the AI art stuff for a few months but want to give it another shot, especially since folks seem to be talking about a bunch of new models and forks of SD that look pretty capable. What are the best ones to try right now? Welcome back. https://civitai.com/ is where everyone are throwing these models so feel free to check that out. Seems to be down right now though that's kind of normal. Protogen models tend to be more realistic, Orange models are anime as hell. You'll see people have been busy while you were away lol. Also worth noting, .ckpt have mostly been replaced with .safetensors. They are the same thing but you can't run arbitrary code on safetensors. https://huggingface.co/blog/lora LoRAs seem to be the next big thing, smaller and easier to train with overall better quality, however they require at least 8gb vram currently. My 1660 super isn't enough to run them so I'm planning out my next computer. Question for everyone, I've seen "48gb of vram to train" thrown about in a few conversations but I can't find more info on that. That's a ridiculous target (comedy 3x Intel Arc) so I need to source that and know exactly what I'd be doing training wise, if anyone can point me in a better direction it'd be appreciated.
|
|
#
?
Jan 28, 2023 02:32
|
|
|
Anyone know if Google has a way to access the MusicLM stuff yet? This looks really fascinating, I'm wondering if it supports some of the generation like the art tools have where you can feed it an existing song and have you generate off that https://google-research.github.io/seanet/musiclm/examples/
|
|
#
?
Jan 28, 2023 03:05
|
|
|
Chainclaw posted:Anyone know if Google has a way to access the MusicLM stuff yet? This looks really fascinating, I'm wondering if it supports some of the generation like the art tools have where you can feed it an existing song and have you generate off that https://google-research.github.io/seanet/musiclm/examples/ Google never lets the public access their fancy ML stuff.
|
|
#
?
Jan 28, 2023 03:06
|
|
|
This entire AI explosion is kinda a direct result of them releasing the foundational stuff like tensorflow in the first place.
|
|
#
?
Jan 28, 2023 03:31
|
|
|
KwegiboHB posted:Welcome back. https://civitai.com/ is where everyone are throwing these models so feel free to check that out. Seems to be down right now though that's kind of normal. Protogen models tend to be more realistic, Orange models are anime as hell. You'll see people have been busy while you were away lol. Also worth noting, .ckpt have mostly been replaced with .safetensors. They are the same thing but you can't run arbitrary code on safetensors. Thanks! The safetensor thing is great since ever since I heard .ckpt models could execute random code I avoided everything but the official SD models. Civitai seems like it has some neat art styles I'd like to give a shot.
|
|
#
?
Jan 28, 2023 06:27
|
|
|
grotesque utility monster steals a bicycle Using Stable Diffusion    https://imgur.com/45t5YZE  https://imgur.com/CeBeaKo  https://imgur.com/fxd56JE  https://imgur.com/dPiKC9c Edit: I cannot figure out how to add thumbnail images  Second edit: I could figure it out Lord of the rivets fucked around with this message at 10:39 on Jan 28, 2023 |
|
#
?
Jan 28, 2023 10:35
|
|
|
mcbexx posted:Yep. I found that once you use that, it's really rough to go back to SD 1.x. do you just mean vanilla 1.x? Isn't Protogen itself built off 1.5, not 2.x?
|
|
#
?
Jan 28, 2023 11:35
|
|
|
Cabbages and Kings posted:do you just mean vanilla 1.x? Isn't Protogen itself built off 1.5, not 2.x? Yes, plain 1.4/1.5.
|
|
#
?
Jan 28, 2023 15:56
|
|
|
Positive central theme: 4k, vantablack, Pink Floyd, nebula, prism, lighting bolt                
|
|
#
?
Jan 29, 2023 04:53
|
|
|
|
| # ? May 30, 2024 08:58 |
|
|
pixaal posted:Positive central theme: 4k, vantablack, Pink Floyd, nebula, prism, lighting bolt I saw those in Spencer's Gifts back in the 80s.
|
|
#
?
Jan 29, 2023 16:17
|
|