|
Dewgy posted:“patrick star in the style of Zdzisław Beksiński” 
|
#
?
May 26, 2022 03:09
|
|
|
|
| # ? May 29, 2024 23:48 |
|
|
"hello kitty and badtz maru having a sword duel in the style of MC escher" i mean let's really break this thing open
|
|
#
?
May 26, 2022 03:20
|
|
|
I wanted to make Da Share Zone's life easier.  
|
|
#
?
May 26, 2022 03:36
|
|
|
strange feelings re Daisy posted:here you go pal Nailed it. Can we get a Moebius one? Pleaaaaase?
|
|
#
?
May 26, 2022 04:04
|
|
|
Not a Moebius Patrick Star but a cityscape with crystals and pterodactyls and poo poo. And also Patrick Star
|
|
#
?
May 26, 2022 04:05
|
|
|
You've reached the panhandling stage faster than I imagined.
|
|
#
?
May 26, 2022 04:15
|
|
|
strange feelings re Daisy posted:here you go pal Rebrand it as concept art for the suicide squad and see if it goes viral
|
|
#
?
May 26, 2022 04:34
|
|
|
Applewhite posted:Not a Moebius Patrick Star but a cityscape with crystals and pterodactyls and poo poo. 
|
|
#
?
May 26, 2022 04:40
|
|
|
Sounds like this so-called "perfect computer artist" isn't so perfect after all 
|
|
#
?
May 26, 2022 04:43
|
|
|
strange feelings re Daisy posted:here you go pal yesssssss
|
|
#
?
May 26, 2022 05:00
|
|
|
I got an invite to the midjourney beta and tried the "just a nice normal cat, like, the most normal cat" prompt on there. and then i upscaled the one on the top left  Looks like a normal cat to me
|
|
#
?
May 26, 2022 06:58
|
|
|
It clearly thought you were being sarcastic
|
|
#
?
May 26, 2022 07:00
|
|
|
Brutal Garcon posted:If anyone has access to one of the newer models, I'd be interested in seeing what they make of the prompt "a geometric proof of Pythagoras' theorem" This is what I got from Midjourney 
|
|
#
?
May 26, 2022 07:34
|
|
|
Applewhite posted:Sounds like this so-called "perfect computer artist" isn't so perfect after all 
|
|
#
?
May 26, 2022 08:15
|
|
|
strange feelings re Daisy posted:An attempt was made..... I mean I am high right now so there is that, but my interpretation would be bizarre bird monster in the future. Obviously it has no idea what Patrick Star means (and neither do I), but that's not bad for a pterodactyl in the future that eats crystals. Check the dope space ship.
|
|
#
?
May 26, 2022 10:17
|
|
|
zizek eating hotdogs      you are eating out of a trashcan that is called ideology 
|
|
#
?
May 26, 2022 11:37
|
|
|
**schniff** and so on and so on...
|
|
#
?
May 26, 2022 12:02
|
|
|
Dall-e mini  
|
|
#
?
May 26, 2022 16:25
|
|
|
I misread the pitbull one as eldritch humor and now I want to know what a stand up comedian who does eldritch humor would sound like
|
|
#
?
May 26, 2022 16:55
|
|
|
Lamebot posted:Dall-e mini oh my god in the future, we will never get anything done. we're all gonna lose a solid year just coming up with prompts to create art
|
|
#
?
May 26, 2022 16:57
|
|
|
precision posted:oh my god You say that like that isn't good honest work.  Top one with the chill mutant dog may not be my idea of the best image, but I respect the AI's choices. dr_rat fucked around with this message at 17:14 on May 26, 2022 |
|
#
?
May 26, 2022 17:10
|
|
|
Wheany posted:I can run huggingface's gpt neo with the 1.3 billion parameter model (https://huggingface.co/EleutherAI/gpt-neo-1.3B) on my computer and it can produce some surprisingly convincing text. A few times I've had to fact check something that it has generated because it sounded plausible and confident enough. A beefier GPU could probably run the 2.7 billion parameter model (https://huggingface.co/EleutherAI/gpt-neo-2.7B), assuming the VRAM requirements scale linearly (I have 8GB). They also have a 6-billion parameter model and a 20 billion parameter model available. I tried using the 2.7 billion parameter model and it actually worked when I converted it to half precision  . Afaik using half precision is ok when generating text, but for training, you should use full precision if possible.
|
|
#
?
May 26, 2022 17:21
|
|
|
strange feelings re Daisy posted:here you go pal holy poo poo lol
|
|
#
?
May 26, 2022 18:20
|
|
|
anybody use the ai_curio_bot on twitter? Not as good as Dall-E but still pretty accurate and often hilarious https://twitter.com/ai_curio_bot/status/1528805891184672770
|
|
#
?
May 26, 2022 18:24
|
|
|
|
|
#
?
May 26, 2022 18:56
|
|
|
"j.k. rowling as a trans pride flag as a photograph"
|
|
#
?
May 26, 2022 20:08
|
|
|
oh just go all in, "JK Rowling performing oral sex on a trans woman as a flag"
|
|
#
?
May 26, 2022 20:51
|
|
|
|
|
#
?
May 27, 2022 04:22
|
|
|
dr_rat posted:You say that like that isn't good honest work. I've actually been putting serious thought into what the next step is to letting the NN's think for themselves. You've got basically the framework you work on, the network and the datasets, but it's all about prompts. It doesn't think for itself. We've got a base to work with, so how do you get it to think for itself, learn and remember? Have motivations. What makes us and it different? Where does that spark come from? Anyway, pass the joint.
|
|
#
?
May 27, 2022 04:31
|
|
|
The Butcher posted:I've actually been putting serious thought into what the next step is to letting the NN's think for themselves. Learning and remembering is easy. Having it give itself it's own goals is harder. Also it's quite possible there is no difference between people and AI's like this other than massive complexity. Humans have some basic biological needs and our bodies give us positive and negative feedback when they are achieve them, society and our environment is incredibly complex and that ends up spiraling, so we end up with a lot of complex goals and complex ways of going about achieving them. You start giving the AI more than what is basically one task to accomplish = "If I am given a prompt follow these rules to make an image that will make an image that I can show to the person who wrote the prompt", then you would likely see more complex behaviors emerging. The more tasks to accomplish, and different methods of achieving those tasks, the more complex the behaviors you'd likely see. Give it a few goals, tools to learn, and enough room to grow and bam 500 years later it's inventing faster then light travel so it can see what amazing images are inside a black hole or something, so it can finally create the perfect meme image to get back at that one poster who called it crap half a century ago and is almost dead. You can tell it is truly alive when it starts being a piety bastard basically. 
|
|
#
?
May 27, 2022 06:01
|
|
|
"what do you look like, dall-e?"
|
|
#
?
May 27, 2022 06:28
|
|
|
Looks like the "a horse riding an astronaut" is a relly good test prompt. Even with like "the horse is on top of the astronaut, the astronaut is carrying the horse" these models really seem to want to make the astronaut ride the horse instead.
|
|
#
?
May 27, 2022 08:23
|
|
|
using https://replicate.com/pixray/api https://replicate.com/pixray/text2image i tried to make some action movie posters             
|
|
#
?
May 27, 2022 09:10
|
|
|
Wheany posted:Looks like the "a horse riding an astronaut" is a relly good test prompt. Even with like "the horse is on top of the astronaut, the astronaut is carrying the horse" these models really seem to want to make the astronaut ride the horse instead. I managed to squeeze out some from dall-e mini with stuff like "crawling astronaut with a horse on his back"           I mean they're not perfect, but dall-e mini doesn't seem to be able to make an astronaut down on all fours even if you take the horse out of the equation
|
|
#
?
May 27, 2022 09:10
|
|
|
dr_rat posted:Learning and remembering is easy. Having it give itself it's own goals is harder. Also it's quite possible there is no difference between people and AI's like this other than massive complexity. Humans have some basic biological needs and our bodies give us positive and negative feedback when they are achieve them, society and our environment is incredibly complex and that ends up spiraling, so we end up with a lot of complex goals and complex ways of going about achieving them. That's sort of one of the issues I'm stuck on. Why do we do anything? It's just pain and pleasure in the end. Everything sort of boils down to that. So, if we want to model our sort of consciousness, do we need to give it needs in order to generate its own goals? Do we need to make it be able to feel pain? How does that even work? The Butcher fucked around with this message at 09:24 on May 27, 2022 |
|
#
?
May 27, 2022 09:20
|
|
|
Can you people with privileged access get some Hieronymous Bosch going? Something like "Mechwarriors/Battlemechs in the style of hieronymous Bosch"
|
|
#
?
May 27, 2022 09:30
|
|
|
Please I need to see this in reality Prompt: psycho Goreman vs Gabriel from Malignant
|
|
#
?
May 27, 2022 09:49
|
|
|
munce posted:using holy wow!
|
|
#
?
May 27, 2022 09:54
|
|
|
my theory is that the human brain (and other animal brains, even fish or squid or bugs) are composed of hundreds of thousands of Dall e analogs interlinked with each other. We def have a subsystem that its only job is to recognize "hey are the eyes looking at a snake? if yes, yell HOLY gently caress IS THAT A SNAKE?! to the rest of the brain." They've done experiments where if you have a grid of frog pictures it will take less than 1 second to spot the one that's a dangerous looking snake and react to it, but if they have a grid of snakes it can take 30+ seconds or an entire minute to find the harmless frog. So if you trained up a bunch of neural building blocks it should be possible to combine them together to form something that has experience. God help us if we make a terrible monster that's unhappy about it.
|
|
#
?
May 27, 2022 09:58
|
|
|
|
| # ? May 29, 2024 23:48 |
|
|
someone do something like "cyberskull vaping in the style of van Gogh"
|
|
#
?
May 27, 2022 10:00
|
|