In this issue:

• we explain Meta AI’s Make-A-Scene; 

• we discuss Meta AI’s Make-A-Scene Paper; 

• we explore LAION, one of the most complete training datasets for text-to-image synthesis models. 

Enjoy the learning!  

💡 ML Concept of the Day: Understanding Meta AI’s Make-A-Scene 

During the course of this series, we have been exploring different text-to-image synthesis models that are able to produce photorealistic images from textual input. Despite the astonishing progress in this area, there are fundamental challenges in generating images that fully capture the semantics of text input. Given a sentence such as “a dog riding a bike”, it is hard to infer whether the dog is big or small, what position it is assuming in the bike, whether the bike is on the left or right side of the picture. You got the idea. Aspects such as position, orientation, and size are hard to convey using textual inputs. These elements are much better captured by drawings or sketches. Enter Meta AI’s Make-A-Scene.  

Created by Meta AI Research, Make-A-Scene is ...