I have no knowledge how it all works, so I'd love to know.
A I Quicksand scenes
- Herzeleid
- Posts: 794
- Joined: Wed Apr 15, 2009 1:15 pm
- Location: Eastern UK
A I Quicksand scenes
Hi everyone, just been looking at some A I quicksand scenes on Deviant Art, and I was very impressed with how real they looked. I know that they are A I generated, but my question is : Are the models featured based on real people, as they look pretty real to me , or is it ALL A I produced ?
I have no knowledge how it all works, so I'd love to know.
I have no knowledge how it all works, so I'd love to know.
"Gott weiss ich will kein engel sein" ( God knows I don't want to be an angel )
_ Rammstein, "Engel"
_ Rammstein, "Engel"
-
papa-lenin
- Posts: 34
- Joined: Sun Jul 19, 2009 5:24 pm
Re: A I Quicksand scenes
Basically, yes.
AI cannot create anything new, it can only pull from the images which it was "fed" with during it's creation process called training.
So the way it works the companies making these models feed their models with billions of pictures scrubbed off internet, and that becomes their data set and each picture is categorized and tagged, so that the model understands what each word means <-- this is also why it is so controversial tech because it of course contains countless copyrighted images.
When you type a prompt, you create sort of a filter for this vast knowledge base so that it pulls from exactly the images of the thing you named in the prompt. And that obviously contains also real people, you can even type some famous actors by name and it'll often give you a very similar lookalike.
And additionally you can create what is called a LoRA - a small mini-model which lands on top of the main one, and that one you can train yourself on for example a real life person you want to feature.
This is really oversimplified and kinda not 100% true explanation, but it should be the gist of it lol.
AI cannot create anything new, it can only pull from the images which it was "fed" with during it's creation process called training.
So the way it works the companies making these models feed their models with billions of pictures scrubbed off internet, and that becomes their data set and each picture is categorized and tagged, so that the model understands what each word means <-- this is also why it is so controversial tech because it of course contains countless copyrighted images.
When you type a prompt, you create sort of a filter for this vast knowledge base so that it pulls from exactly the images of the thing you named in the prompt. And that obviously contains also real people, you can even type some famous actors by name and it'll often give you a very similar lookalike.
And additionally you can create what is called a LoRA - a small mini-model which lands on top of the main one, and that one you can train yourself on for example a real life person you want to feature.
This is really oversimplified and kinda not 100% true explanation, but it should be the gist of it lol.
Return to “General Discussion”
Who is online
Users browsing this forum: No registered users and 0 guests