so the seed (currently set at -1) controls how it will interpret it... if you set the seed at a specific number other than -1, it should fix it at that seed. -1 means it allocated a seed at random. i don't really understand this part of it tbh, you'll need to look it up. if you find out, lmk
Hearthleaving
- Thread starter constant escape
- Start date
Clinamenic
Binary & Tweed
So -1 would make this probabilistic then, if it just means a random number.
Pepe the alt right frog devouring kermit with Boris Johnson's face | in the style of Goya's Saturn
Clinamenic
Binary & Tweed
That is amazing. One can discern a pseudo-Johnson.
sufi
lala
& so if you repeat the same set of initial conditions it will not create the same result?
Clinamenic
Binary & Tweed
The seed component will come in handy for optimizing aesthetics around a fix text input.
sufi
lala
that is revolting 👍
Clinamenic
Binary & Tweed
If I understand it correctly, yes, you should be able to rerun the same text input and get different outputs, with -1 as the seed.& so if you repeat the same set of initial conditions it will not create the same result?
Clinamenic
Binary & Tweed
For example, I'm getting different output than you did with the same "robot art | Betye Saar" input.
Clinamenic
Binary & Tweed
It seems a person's name, being taken as input, territorializes the person's face. Makes sense, just seems difficult to abstract an artists style from their face, in terms of influencing the output.
Clinamenic
Binary & Tweed
I remember going crazy about this three years ago and not being able to explain to anyone why it struck me as being so profound.
yeah, exactly. if you leave it at -1. but if you fix it at another number, they should come out similar. i just use -1 and have not investigated fully& so if you repeat the same set of initial conditions it will not create the same result?
Clinamenic
Binary & Tweed
A very Deleuzian technique, this.
Clinamenic
Binary & Tweed
Especially with the temporal dimension. A lot to philosophically analyze regarding semiotics. Organs are defined by functional distinction, rather than perceptual distinction, but watching various semiotically distinct objects coming in and out of existence, ones you can attribute to input texts, is very much in this Deleuzian mindspace.
also you can change the model to a different one - that would also give totally different results. it's got so much random built in.
and also the very nature of how it works... GAN = generative adversarial network... i read a little bit of an article about it and it is actually two networks working together/against one another ie VQGAN has a go at doing what you've asked and CLIP says yes or no in terms of the prompt and guides it. so it's like acting/directing a bit.
and also the very nature of how it works... GAN = generative adversarial network... i read a little bit of an article about it and it is actually two networks working together/against one another ie VQGAN has a go at doing what you've asked and CLIP says yes or no in terms of the prompt and guides it. so it's like acting/directing a bit.
Clinamenic
Binary & Tweed
Didn't know GANs were behind thispersondoesnotexist.com
Clinamenic
Binary & Tweed
Although the site does display a little text bubble that says so.
sufi
lala
The serach results are like an AI essay on memeism already https://duckduckgo.com/?t=ffab&q=go...travel,x1000,right-pad,750x1000,f8f8f8.u1.jpg
