AI Generated Content - It's all terrible and unfunny.

  • 🔧 Site instability resolved. You can report double-posts and broken attachments. For bigger issues, use the Technical Grievances thread.
    🇵🇦 Nuestro primer dominio localizado está en español en kiwifarms.pa. Our first localized domain is on Spanish on kiwifarms.pa.
  • Want to keep track of this thread?
    Accounts can bookmark posts, watch threads for updates, and jump back to where you stopped reading.
    Create account
It takes hundreds of thousands of existing headshots of furries and AI uses them to generate new headshots. Some of these come out to look hilarious. Like this potato with eyes that appears very distressed by it's own existance:
 

Archivos adjuntos

  • B3BFF824-B4FE-418C-B36F-627D372B757D.jpeg
    B3BFF824-B4FE-418C-B36F-627D372B757D.jpeg
    3.9 MB · Vistas: 98
Pretty sure the consensus on these is that most of them are fake. They didn't get a bot to generate anything, they just wrote some stupid crap and try to pass it off as if a bot wrote it (it's especially obvious with that Netflix one, no way could a bot keep the story that consistent and remember the names of the people in the story and what happened, have you ever tried AI Dungeon? It forgets stuff instantly)
 
Última edición:
It was kind of interesting when I went back to "thispersondoesnotexist" to see that the AI had learned enough to not give every person smiling three front teeth.
 
https://thispersondoesnotexist.com/ generates very realistic portraits, but has some struggles with clothes, accessories and backgrounds. Look at this dude, everything seems fine, until you take a closer look at his neck and shirt and realize that something's wrong here.
1630435744231.png

And it gets worse when there are people in the background.
1630435906458.png
1630436102260.png
 
The only thing I can add it how those those images are created, basically imagine you add a black box that converted an image (or some other form of complex input like text) to a set of number and a second black box that converted those set of numbers into the original image.
Now you just take the input to the second box and give the numbers a little nudge, this will create (in a very high probability) something new that no one has ever seen.
 
This is an image generated in artflow.ai with the search term "Average gamer":

Ver archivo adjunto 2509235
That site is currently way overloaded and wants me to wait 11½ hours to see William Shatner as Captain Picard.

Here's a fun toy that runs in your browser based on neural cellular automata: https://distill.pub/selforg/2021/textures/

Two Minute Papers on YouTube has tons of content about this kind of stuff, many times with links to interactive examples in the video description.
 
Atrás
Top Abajo