View Tweet
great thread. read.
nothing ai-generated art does could be considered plagiarism unless we stretch the definition to absurd lengths.
ai art models do not store or reproduce any specific images. their training results in a highly complex statistical model that contains large-scale statistical data about the collective set of images (without retaining any data about individual images).
to draw a somewhat strained analogy, i'll borrow this analogy from someone else. imagine you had an idea for creating the next van gogh painting. you compiled a comprehensive file of van gogh's works and made an extensive excel spreadsheet.
in this spreadsheet, you tracked recurring elements: each cell might note "vibrant colors" or "swirling skies" and the frequency these elements appear in his paintings.
from this, you drew correlations between these cells -- what percentage of paintings with swirling skies also featured cypress trees? what percentage included starry nights?
if you then instructed someone with moderate painting skills to use your spreadsheet to create a van gogh-like painting, it would likely resemble a van gogh piece: but any accusation of "plagiarism" would be ridiculous.
a gross simplification, of course, but it’s much closer to how ai art functions compared to the "collage machine" analogy that many people concerned about plagiarism use. and even if it were a collage machine, it still wouldn't be plagiarism because collages aren’t plagiarism.
naturally, the original quoted thread is by someone who actually knows what they're talking about and goes into much more depth.
Generated by Thread Navigator
Press ⌘ + S to quick-export