An Asian MIT student asked AI to turn an image of her into a professional headshot. It made her white with lighter skin and blue eyes.::Rona Wang, a 24-year-old MIT student, was experimenting with the AI image creator Playground AI to create a professional LinkedIn photo.
That image highlights an important point, these AI produce an infinite number of images for any given prompt. It’s easy to pick one and make conclusions based on just one, like this this article did, but you’re literally ignoring infinity other images produced for the same prompt.
This was the first one I generated. I could keep generating and get different results.
I could add a LoRA and get another infinity worth of different results.
Actually though… is it infinity? Or just a very high number based on the seed?
My intuition says that it’s just a very high number, because there’s a finite variation of initial noise and a finite seed number.
Finite, but big enough that you won’t get exact repeats any time soon. For a 32-byte seed and a reasonably uniform PRNG, any time before the Sun enters its red giant phase and absorbs the Earth.