Images generated by AI models trained on massive datasets often can’t be traced to specific training images, MIT CSAIL researchers found. Removing individual images from the dataset didn’t change outputs, complicating copyright questions.
I don’t think this is really all that surprising if you understand how diffusers actually work. They don’t just draw contours and shit. They apply sequential transformations across a matrix of pixels in an iterative fashion, the transformation varies each iteration based on prompt embeddings and the surrounding pixels of each respective transformation, typically starting from a seed of arbitrary static. A lot of images share overwhelmingly identical techniques, with a tiny sliver being responsible for the truly creative deltas. Diffusers effectively learn techniques so it tracks that dropping singular training data here and there doesn’t change much.
I don’t think this is really all that surprising if you understand how diffusers actually work. They don’t just draw contours and shit. They apply sequential transformations across a matrix of pixels in an iterative fashion, the transformation varies each iteration based on prompt embeddings and the surrounding pixels of each respective transformation, typically starting from a seed of arbitrary static. A lot of images share overwhelmingly identical techniques, with a tiny sliver being responsible for the truly creative deltas. Diffusers effectively learn techniques so it tracks that dropping singular training data here and there doesn’t change much.