clues to movies that don't exist
clues to movies that don't exist
sorry for the spam, but i am so amazed at how good AI image creation is now since i last played with it that i've been doing nothing for the past two days but playing around making imaginary actors and film stills for films someone should have made, and i have to post them somewhere.....
firstly, silent medusa....
but then i add futuristic robot medusa to the prompt and got:
and then cuz i liked the hoops so much i did a series of satyr characters:
and then just for fun, since it's so good with coils, i did a couple of southern gothic farm pics
firstly, silent medusa....
but then i add futuristic robot medusa to the prompt and got:
and then cuz i liked the hoops so much i did a series of satyr characters:
and then just for fun, since it's so good with coils, i did a couple of southern gothic farm pics
it's terribly kitsch, but also, there are a really lot of hot imaginary men, and you can add a moustache to anyone....
here's the hollywood version of medusa:
and the film of diana and actaeon seems to be a hilarious comedy that i definitely would watch, lubitsch was probably involved:
more from the satyrs:
and then also, to prove it's not completely racist, every so often, but not that often, it offers a non-white response:
actually it has a strange sense of identity. looking at other people's efforts, i guess it recognises contemporary celebrities, but i could not get it to recognise a single one i wanted, not even chaplin or buster so i just had to recognise whoever popped up....
this is jaque catelain?
lars hanson?
pierre etaix?
here's the hollywood version of medusa:
and the film of diana and actaeon seems to be a hilarious comedy that i definitely would watch, lubitsch was probably involved:
more from the satyrs:
and then also, to prove it's not completely racist, every so often, but not that often, it offers a non-white response:
actually it has a strange sense of identity. looking at other people's efforts, i guess it recognises contemporary celebrities, but i could not get it to recognise a single one i wanted, not even chaplin or buster so i just had to recognise whoever popped up....
this is jaque catelain?
lars hanson?
pierre etaix?
but for some reason, what really really worked was 'czech'
how czech are my czech medusas?
it works for anything, here are my czech pierrots/mimes/clowns:
although tbf it also captured french quite well:
(sulky rather than depressed czechs:D)
how czech are my czech medusas?
it works for anything, here are my czech pierrots/mimes/clowns:
although tbf it also captured french quite well:
(sulky rather than depressed czechs:D)
i don't know what data set the ai is using but it really must have sucked up a lot of czech fairy tale movies or something.....who wouldn't want to watch a colourful czech fantasy starring these characters and settings?
(NB you can see that i did get highly distracted by just making houses by the end, but still they all seem insanely czech-magical to me)
(NB you can see that i did get highly distracted by just making houses by the end, but still they all seem insanely czech-magical to me)
and finally, it's terrible at reproducing magpies, these are clearly pigeons disguised as magpies
but i at least have my next two avatar pics:
ok i'll stop now, will probably get bored by next week
but i at least have my next two avatar pics:
ok i'll stop now, will probably get bored by next week
don't summon a cthulhu please there's quite enough going on right now
i adore those pierrot houses ♥
i adore those pierrot houses ♥
lol, well, actually i was trying to create some spirit photography but haven't quite got the hang of it yet....at the minute it's just going into the wallpaper....
what tool are you using to make these?
Wow, love them all Sally! I still haven't tried AI tools. Maybe I should have someone riding a dinosaur for my avatar.
I love the seasonal banner pics we have for scfz, but wouldn't it be fun to try mixing some visuals/frames from three or four films (from the same year, say 1931, or make a mix from a few radically different years/eras instead) into the one pic, using AI tools, to use as one of our banners?
https://www.bing.com/create
pabs wrote: ↑Thu Oct 12, 2023 4:08 am I love the seasonal banner pics we have for scfz, but wouldn't it be fun to try mixing some visuals/frames from three or four films (from the same year, say 1931, or make a mix from a few radically different years/eras instead) into the one pic, using AI tools, to use as one of our banners?
i can't do that with the ai image generator i'm using. each one has its strengths....i think you can input images and then tweak generated ones in midjourney, but i haven't tried that.
the main advantage of bing is that it's free and from reading online, and also what i also quickly discovered, is that you can't use existing images or recycle ones it generates for you, each image is a one-off, you can't add negative prompts (ie you can't exclude things) and it's designed to be really good with objects. which it is. for example, it LOVES the 'weird still life painting' genre, anyone can knock one of those out in one go:
apparently there's some trick to stop it chopping off the sides of images with its stupid aspect ratio, but i haven't figured it out yet.
and of course, everyone loves weird medieval monsters:
also, it likes detail. here's some psychedelic lichen i made after rewatching shepherd xeudi for the 73 poll:
but it is TERRIBLE with people. and even if you do manage to get someone that looks human, i think in AI world all the women are murdered at 30, i had to wrangle it for ages to get a woman not in excessive make-up, who was normal looking, or older than 20....
but apart from that, it is incredible really. and FREE. at the minute.
anyway, i was continuing the mythical movie theme with icarus, but lithuanian como snuck in, and then i just got more intrigued by the buildings again, and the limits of what you could do to a horse
thanks sally
i can see this being a bit addictive...
i can see this being a bit addictive...
yes! maybe just ignore this thread for.....a while
today i went on my own walk through H and mildly terrified myself...
today i went on my own walk through H and mildly terrified myself...
what did AI do to sancho????
(nevermind, it did a lot of other very pleasing versions)
(nevermind, it did a lot of other very pleasing versions)
"Joao Cesar Monteiro as Charlie Chaplin" was flagged for some reason.
lol, although I'm guessing anything Chaplin related is flagged...
ah you reminded me this thread existed....gonna dump some here, although i've gone more now from being interested in making pictures of movies that don't exist, via the occasional pretty boy porn (in which i discovered the above posted spirit animal, who i named bardo) to now just being more interested in creating pics that only AI could have created, which usually means deformed (to us) bodies but also mixing proportions and planes (although that is harder to get it to do).....
these are gorgeous
i've been playing around with different ai's there's quite a difference between them all at the moment. the three main underlying systems available to idiots like me are dalle, midjourney and stable diffusion. midjourney is $$$ and on discord and i haven't the heart to get to grips with all that.
from seeing what other people produce, stable diffusion makes the best looking pictures and you can prompt with images as well as words, although the prompt understanding is minimal and you just have to be happy with a bunch of really nice looking random stuff that wasn't what you asked for, but the best thing is it's open source and you can train your own models, so it's free with the caveat that your PC has a good enough dedicated GPU (mine doesn't so i'm buggered) but there are a few places online where you can either pay for some partial access or get a free number of credits every day.
bing is based on dalle and is free, it has the best prompt understanding, you usually get what you ask for, but there's a lot of restrictions and the image quality is increasingly rubbish, plus the dataset it was trained on presumably only had grotesque fish-lipped russian prostitutes as examples of 'women'..if i want anyone normal looking i have to prompt 'ugly german housewife from 1953'
bing also doesn't really recognise many people or art styles (although i'm guessing no one has a dataset that includes enough monteiro to reproduce him, i can't even get any george osborne!) but the stable diffusion ones will let you get approximations of some people and art styles - which is where it gets ethically icky as in exploitation of artists....i do try and only use dead artist mash-ups although i gave in the other day and blasted everything with martin parr because it gave such good lighting to the images that i couldn't get any other way, sorry martin!
anyway, if anyone wants to play with the SD-based side of it, there's limited free access that refreshes daily on
https://app.leonardo.ai/auth/login
https://replicate.com/stability-ai/sdxl
from seeing what other people produce, stable diffusion makes the best looking pictures and you can prompt with images as well as words, although the prompt understanding is minimal and you just have to be happy with a bunch of really nice looking random stuff that wasn't what you asked for, but the best thing is it's open source and you can train your own models, so it's free with the caveat that your PC has a good enough dedicated GPU (mine doesn't so i'm buggered) but there are a few places online where you can either pay for some partial access or get a free number of credits every day.
bing is based on dalle and is free, it has the best prompt understanding, you usually get what you ask for, but there's a lot of restrictions and the image quality is increasingly rubbish, plus the dataset it was trained on presumably only had grotesque fish-lipped russian prostitutes as examples of 'women'..if i want anyone normal looking i have to prompt 'ugly german housewife from 1953'
bing also doesn't really recognise many people or art styles (although i'm guessing no one has a dataset that includes enough monteiro to reproduce him, i can't even get any george osborne!) but the stable diffusion ones will let you get approximations of some people and art styles - which is where it gets ethically icky as in exploitation of artists....i do try and only use dead artist mash-ups although i gave in the other day and blasted everything with martin parr because it gave such good lighting to the images that i couldn't get any other way, sorry martin!
anyway, if anyone wants to play with the SD-based side of it, there's limited free access that refreshes daily on
https://app.leonardo.ai/auth/login
https://replicate.com/stability-ai/sdxl
i can't even make a banner with genuine screencaps from real movies, the proportions are just too weird, so no chance with AI
anyway had more important things to do making more hot boy porn
anyway had more important things to do making more hot boy porn