/ic/ - Artwork/Critique

>>6934584
>ehehe ur just using a semantic trick!!
no.
generalization is just what it is.
it is detecting patterns across many many different images with similar tags.
if the AI copies a 1000 bottles during training, then it will see that blurry starting shape always looks a certain way and the details also always look a certain way.

so when it creates outside of training, it will know the probably shapes to start with and the probable details to make up.

and i say shapes, but it's not just shapes. it's any complex relationship.
like
>sunny images always have certain colors
>night time images always have certain possible colors
>water reflects surrounding colors
>etc etc

>>6897054
you think learning a style is the same as recreating an image?
>pffftbwahahahahahahahahahaaaaa
:)
see, you people really don't have the slightest clue about any of this.

and that is the scary part about this, isn't it? that the AI can actually do this.
though realistically speaking, it would probably take a lot of effort to make it actually fully learn a style in a meaningful way instead of just having a surface understanding of it.