I decided to spruce up my blog by finally adding some cover photos to my articles. I didn't want to use Unsplash, so given that DALL-E is now openly available, I decided to give it a try.
Frankly, what I got wasn't what I had in mind, and many of these are unusable. Perhaps it's because my queries are highly specific. I settled on using abstract art styles to cover for the AI's misses on the specific things I wanted in the shot.
Regardless of the rough edges, I am thoroughly impressed by how easy it is now to generate unique cover photos for articles. I no longer have to rely on Unsplash or OG image generators that everyone else seem to be using. Now, I could make my posts more distinct with unique cover art.
Here are some queries I tried, and their results.
"a landscape with the hasura logo and a purple hot air balloon"
"a landscape with gentle rolling hills in purple gradients with a sky using hex color 1599e5, with a black and white striped hot air balloon prominently in the foreground"
For this next query, I was expecting something in the style of a professional illustrator, not something made in MS Paint:
"a cartoon with a shark that looks like the ikea blahaj swimming in a bright blue ocean"
I tried specifying a more specific art style. I wanted to challenge the AI a bit, so I prompted it to use the art style of a more niche artist like Joan Cornella
"a shark that looks like the ikea blahaj swimming in a bright blue ocean in the style of joan cornella"
In fairness to DALL-E, Joan Cornella's subjects are mostly all people, so it would be extremely challenging to know how he would illustrate a shark.
"a stick figure trying to exit a door in the style of basquiat"
I tried challenging DALL-E futher by asking for paintings of HTML code. Let's just say that DALL-E is not very good with text:
"an impressionist painting of html code"
"a colorful painting of html code in the style of basquiat"
"a colorful painting of html code in the style of Wassily Kandinsky"
"a colorful illustration of a busy TV studio, focusing on the mixers and monitors in the foreground, with the cameras and set in the background in the style of Wassily Kandinsky"
DALL-E also seems to struggle with highly-specific requests. In the following illustrations, I needed to display a composite RCA cable, but instead kept getting USB cables:
"a colorful 3d model of the raspberry pi computer with a composite rca video cable hanging from its 3.5mm jack"
"a basquiat style painting of the raspberry pi computer with a composite rca video cable hanging from its 3.5mm jack, with a sky blue background"
With a more generic prompt that is more open to interpretation, DALL-E performs much better:
"a robot painting in renaissance chiaroscuro style"
