Writing Images: Exploring the data behind popular text to image generation models

This website provides a brief introduction to the data and technical processes behind popular text to image models. *caveat alert: a lot of large models are proprietary meaning we can't be 100% sure what they are trained on but likely it is a combination of LAION-5b and other datasets

Below, we take a ‘tour’ of the creation of LAION 5b (one of the largest text & image datasets), making pit-stops along the way to unpack technical terms and discuss potential ethical questions that arise. A lot of what I've learned is from this paper by Abeba Birhane, Vinay Uday Prabhu and Emmanuel Kahembwe ↗.