How programmers transformed the internet into a painting tool. Imagen from DALL-E 2 is explained.
Since January 2021, advances in AI research have resulted in a plethora of deep-learning models capable of generating original images from simple text prompts, effectively extending the human imagination. Researchers at OpenAI, Google, Facebook, and others have created text-to-image tools that have not yet been made public, and similar models have proliferated online in the open-source community and at smaller companies like Midjourney.
Because they eliminate the need for technical labor in the image-making process, these tools represent a massive cultural shift. Instead, they look for creative ideation, linguistic skill, and curatorial taste. The ultimate consequences are difficult to predict, but these algorithms herald a new, democratized form of expression that will spark another explosion in the volume of imagery produced by humans, much like the invention of the camera and, later, the digital camera. However, they, like other automated systems trained on historical data and internet images, have unresolved risks.