In the post introducing AI Art Generators (HERE), I mentioned there was one that I was waiting on. It’s by invitation only, and I finally got in . . .
It’s the DALL-E AI Generator, and I had great hopes for it as it seemed to generate more ‘realistic’ renderings than most. So, naturally, I tried duplicating some of the stuff I tried before.
For instance, generating a graphic for my Rainman Story (previous attempts HERE).
I didn’t play with these, so they are presented exactly as rendered. I supposed I could ‘improve’ them if so motivated. For instance, I could do this to the above rendering.
But, for now, I’ll show what the AI put out without messing with it.
I won’t show all the concepts (they are all in THIS SmugMug gallery, and there’s a link to a slideshow at the end of the post), but enough to give a feel for the tool. As a reminder, the prompt for these was: Man standing in arid landscape next to a clean and shiny white 1960 cadillac, thunderstorm in the distance, photorealistic, ultra-realistic, detailed.
Mind you, I’ve not learned all the bells and switches peculiar to this tool (every tool has different qualifiers and modifiers), and I just shot from the hip, as it were.
Right off, I notice the renditions are closer to what I envisioned when I gave the prompt than the offerings from the three other AI Art Generators. Still not rendering people’s faces, but at least I recognize it’s a human figure.
As mentioned, I’m not showing all the stuff DALL-E generated, but all the images are in the SmugMug gallery, and they’ll all be in the slideshow at the end.
Next up, I made a mistake I didn’t notice until later, so I couldn’t work out why I got what I got. Can you spot the mistake?
Prompt: Chictopher Columbus caravelle in space with planet as a backdrop.
There are a couple more along the same lines, and I imagine the AI interpreted my mistake as referencing a space shuttle. As I said, I didn’t realize I’d misspelled the name, so — after accusing the AI of being an idiot — I gave it another prompt.
Prompt: Sail ship in space with planet as a backdrop.
I thought those were pretty good, so I modified the prompt slightly . . .
Prompt: three masts sailing ship in space with planet as a backdrop, high detail, bright, photorealistic, realistic.
There’s some ‘extra’ stuff in some of these (I don’t understand why unless the AI interprets the words separately for each variation, thinking it I might want three ships), but these, again, are a lot closer than what I got from the other AIs.
I then asked for variations on what I thought was the closest to what I wanted, the second offering . . .
I tried a slightly different prompt to maybe further limit what the AI would interpret.
Prompt: three masts wood sailing ship in space with planet as a backdrop, high detail, bright, photorealistic, realistic.
Again, I’d like to know for sure what modifiers are available and how the AI might interpret them. But, still shooting in the dark, I did another prompt:
Prompt: 3d render of a wood three-mast sailing ship in orbit around jupiter, high quality photo.
So, not a bad ship, but where’s Jupiter? So, thinking the AI might just not like Jupiter, I modified the prompt.
Prompt: 3d render of a wood three-mast sailing ship in space with Saturn in the background, high quality photo.
OK, so no rings, but now we might have something like Jupiter.
At this point, I got tired of playing the game and tried another prompt: color drawing of sailing ship falling off the edge of the earth with sea serpents and dragons flying above
Here, again, the AI went off in a totally different direction. THIS is what I was looking for, and given that it easily comes up with a simple search, I’m disappointed the AI — all the AIs — failed to generate something like those depictions.
This brings me back to what I observed about AIs . . . it’s not intelligence. They can certainly generate some amazing graphics, but there’s no “understanding” there. Heck, there’s no there, there.
Of course, this is also a problem with humans; we’re rarely able to convey our mental image to someone else using words. As someone who tries to write, I’m aware that my choice of words affects the interpretation of my writing, but I can’t reliably predict how different readers will interpret the same words.
Much like with the AI, I rely on people to have a general idea of what a sunset is — or a sailing ship — but when it comes to specifics, what individuals envision can be significantly different than what I intended. I view describing things to AIs similar to describing colors to a deaf person or sounds to a blind person . . . and unless you know the definition of blind and deaf, you won’t know I switched the terms.
Anyway, here’s the slideshow of the DELL-E results to date:
That’s it. This post has ended . . . except for the stuff below.
Note: if you are not reading this blog post at DisperserTracks.com, know that it’s copied without permission, and likely is being used by someone with nefarious intentions, like attracting you to a malware-infested website. Could be they also torture small mammals.
Note 2: it’s perfectly OK to share a link that points back here.