Hello, that is Denis. Recently a guy I know has got access to Dally 2 algorithm. Basically, this AI was trained on billions of text and image pairs and is now capable of generating an image from any text, you can read about it more here.
We immediately started to generate “robotic doom like hamsters” because why do we need AI if not for that. Dalle 2 could generate up to 8 versions of the same topic in 1024x1024 resolution in less than a minute.
I have slightly animated generated results of the Doom-Alike-Hamster-Demon-Hunter:
(turn sound on for more demonic effect)After enormous satisfaction, we wanted to try something new: Dalle 2 also provides tools that could “generate” the rest of the image. You fill in a blank where AI should complete the rest, and AI fills the rest with new data - as simple as that sounds.
So, I have prepared those images: One for the “Mona Lisa” painting by Leonardo da Vinci.
And one for the “Girl with a Pearl Earring” painting by Johannes Vermeer.
The idea was to see will be Dalle 2 able to generate “full-body” versions of the famous paintings, and it looks like AI was superb in dealing with it (except hands, hands are complicated even for humans, I can’t draw hands too for example).
Mona Lisa full-body versions
Girl with a Pearl Earring full-body versions
As you can see, there are some glitches in results for sure, but all those images have been generated in ~2 minutes or so; practically, glitches could be reduced if we will generate a few thousands of paintings and just cherry-pick the best looking ones.
It was a fantastic small experiment, which I personally will remember as “2022 will be a year when humanity has hacked art as a conception”; even if those results do not bring much value to the art world, they still look interesting and inspirational.
BTW, you can apply for a Dalle 2 wait list here.