/ic/ - Artwork/Critique

(2/2)

How it works is pure math. The AI understands that graphic data is supposed to be arranged in a certain way, pixel by pixel, via tensorization- when fed image data, in the case of image gen. When prompted, it will try to re-create the ((PIXEL PATTERNS)) weighed by the keywords you prompted (which are also used in the learning process). It knows that certain patterns (aka pixel placement) are expected, but unlike a human, is unaware of the validity of said pattern in the output, because it doesn't know what a cat is, biologically.

In your research, you might've misunderstood the term "3D tensors" which is what is used to get all data from a colored image- and interpreted it as the AI learning "things" in a 3D aspect, with full spatial visualization and projecting and translation into a 2D plane like how human artists work. 3D tensors are actually just a plain image processed in raw data for learning purposes; it contains pixel intensity information for all channels and is necessary for colored images. The AI literally scrapes the image of all mathematical information it has- but it doesn't know what it is. We literally aren't capable of copying raw data to such a degree by just looking at an image- which is the crux of the problem.

(I also think I forgot to sage the last post, my bad.)