What does it mean for image generation to be responsible?
Recently, large multi-modal models that can generate images based on textual descriptions (also called text-to-image generation models) have exploded in popularity. Research community announced multiple models and the public started to pay attention to these models that provide creative image generation and editing capabilities. These models are trained on large datasets of images and aligned texts and employ various deep learning methods. With all the impressive capabilities of text-to-image generation technology, they present many risk factors and it is imperative to understand these risks and find solutions. In this talk, I will go over Google’s text-to-image models Imagen and Parti, and discuss what it means for image generation models to practice responsible AI.