@davidaugust @slightlyflightyone@sergal.org I am aware of this problem and was eager to see what people think about this approach and of the paper "Extracting Training Data from Diff. Models". Currently working with an open source lawyer who describes the copyright issues of AI-generated images in an article. The abstract ways in which images from the training database are represented within the models' latent space does not allow a copyright attribution at present, afaik. Our focus is the output, though.