Table of contents
No headings in the article.
If you follow trends in twitter couples months back, you may notice that there a lot of machine generated image pass over your social media timeline. especially, since OpenAI's DALL E 2 took off the as beta product at price tag 155 image (generations) for 15 USD. Pretty expensive if you have no idea how to prompt it. some folks people trying to make open sources version of it, so here you go Stable Diffusion from Stability AI, Compvis, LAION.
The idea here is not new. since the AlexNet break record of ImageNet dataset. people are starting to believing that computers can "see" given enough data to train the model. fast forward in time, people are wandering if it could do otherwise not, auto-captioning alt text. but generate image from text or its captions. you may think "it easy, just gather a lot of image data from internet with alt text", well you not entirely wrong it was happen. but, it is not as easy as it seems.