Author: Philipp Gayret
This entry is about some funny images generated with OpenAI DALL·E, the queries used to generate them and how to "speak" to an AI. All art posted here was created in the span of 2 or 3 hours.
OpenAI DALL·E is, to quote Wikipedia; "(...) transformer models developed by OpenAI to generate digital images from natural language descriptions."
Today I received access & AI is serious business, so I tried it out and immediately and ended up using all my free credits bringing some of my online friends' avatars to life. Here's two of my favorites that took many attempts to make them look funny;
I will provide no further context.
During the process of learning how to talk to it, I shared some images with one of my favorite business-oriented internet forums. The query for the second row of images was“evil bear laughing maniacally from the shadows as he hides behind a transparent computer screen displaying stocks with downwards trends in red”
I received a few comments including suggestions from the forum on what to further generate. Note that the forum is ephemeral in nature so the discussion by now is already dead and gone, but one comment stood out which said to describe an image as if it exists and everyone knows it. I applied this to a request of a “laughing mermaid looking at sales charts”
Instead of generic laughing mermaids, I got something really cool back. Now it does look "AI-ish", but so does everything examined up close from OpenAI DALL·E. The query used reads; “Screenshot of the mermaid laughing at sales charts on a computer screen from the movie Mermaid Stock Trader by Hayao Miyazaki”. OpenAI DALL·E can be nudged to draw more complex or rich imagery, by describing something as if it was a photo or screenshot that really exists. Presumably, that's how it was trained!
Note that Hayao Miyazaki never made any movies about stock trading. Probably didn't make any about mermaids either. What's really incredible about the specific generation for this query is that the chart she's looking at has the shape of a whale, which has its own meaning in this context.
The people requested more images to be generated. Below are the best images, of the best sets of images, of the best queries people suggested. My favorite is the "Wheels of Progress" painting, that said, many of the others below are crazy good as well. Especially considering the time spent making them; Minutes, or seconds even.
You're 40% on the scrollbar, but that was the whole blog!
OpenAI granted me the exclusive rights to reproduce and display these pieces of art; so behold!!
“Benevolent and charitable bull handing you dollar bills from through a transparent computer screen displaying stocks with upwards trends in green”
“A cartoon black swan with glowing red eyes laughing maniacally in front of a burning planet earth, cinematic photograph”
“A toad with a euphoric facial expression and dollar signs over his eyes behind a bright glowing blue arrow blasting upwards accompanied by 3 white cubes, painted by rembrandt”
“Cross section of new york, upper part is wall street, lower part bulls and bears in masks stealing bags of money, night sky blue tint”
“Screenshot of an anime mermaid laughing at sales charts on a computer screen from the movie Mermaid Stock Trader by Hayao Miyazaki”
“Artists having their careers crushed by the wheels of progress. Oil on canvas”
Author's note: Open AI DALL·E is non-violent to its fullest extent. So much in fact that for many things you'll receive a warning rather than a set of images. Maybe one day we'll be able to see how violent the wheels of progress really are.
“Zdzisław Beksiński's famous work depicting painters in despair as machine creates a flawless shining painting” (Generation 1)
“Zdzisław Beksiński's famous work depicting painters in despair as machine creates a flawless shining painting” (Generation 2)
Click here to go back up to the funny cats.