Coming Soon: AI for Artwork - Learn Stuff

Artificial Intelligence of varying levels has become part of our everyday lives. Most of use interact with Siri, Google, or Alexa on a regular basis. We are scanned by very sophisticated facial recognition AI without even knowing it. The next new thing is going to be AI that creates artwork.

DALL·E 2 is on the way and it can create artwork in various styles based on just a few keywords. It can also edit or re-create existing images.

If you think this sounds kind of wild, I agree with you. There is a basic demo at https://openai.com/dall-e-2 that demonstrates what this thing can do.

As an expert in digital manipulation with decades of experience, I have a few questions.

Some of the output examples include what are clearly actual photographs. It`s one thing to program line-drawing output, for instance, but are they telling me this AI can create photographic-level realism?

Inserting an object into an image is relatively easy (even accounting for shadows or reflections), but removing objects can be quite a challenge. I do it all the time, and the problem is you don`t have the data from behind whatever you are removing. As a human, I use textures, patterns, or sometimes just paint-in the gaps. According to the demo page, Dall·E 2 does this accurately with little effort? What?

Obviously the demo is just a demo and could easily be completely fake. If Dall·E 2 can do half of what they say, the average person will be able to do what used to require years of practice and, you know, actual skill.