NAILED IT
It’s been a while since I had noteworthy (or fun) bloopers to share. AI art has improved by leaps and bounds since I began this project in mid-2023, but fear not! AI is still providing the crazy. Last week I was trying to create fairly complex images that included three people and a flying robot. On top of that the setting was a dark cave. OpenAI/DALL-E 3 was not quite up to the task.
I was having such a hard time that I decided to focus on one character at a time. My prompt: “Create a graphic novel image of a 50 year old woman with long black hair. She is inside a dark cave.”
It gave me this:
Not bad. But what are those shadows on her face that look like dirt? I simply wrote “Do it again.” And got this:
Now the dirt is . . . hair? Why does she have a five o’clock shadow up her nose?
DALL-E now shares a note when it creates an image. With this one it wrote “Here’s another graphic novel-style illustration of the 50-year-old woman in the dark cave. Let me know your thoughts or if you’d like further adjustments!” I assume it used an exclamation point to demonstrate its enthusiasm for hairy woman.
I repeated, “Do it again.” And voila . . .
Very good. Very Raiders of the Lost Arc. But now she is too young. (I run into this a lot where the AI fixes one thing only to backtrack into other mistakes.) I told DALL-E to “Make her look 50.” The result made me laugh at loud.
Listen, I am the first to admit that middle age has made me a bit more . . . let’s not say hirsute. Let’s say “peach fuzzy,” but this is ridiculous. So the problem was not the lighting in a dark cave, it was about accurately showing a middle age woman with a beard she hasn’t plucked?
DALL-E’s note: “I've refined the illustration to ensure the woman clearly appears to be 50, with visible wrinkles, fine lines, and mature facial features reflecting her life experiences. Let me know if further adjustments are needed!”
Frustrated, I went back and looked at all my prompts so I could figure what might need to change. That’s when the penny dropped: DALL-E was conflating a woman in a cave with a cave woman!
At this point, I had spent nearly an hour working on the one image, so I started a brand a new thread with DALL-E that said “Create a graphic novel image of a 50 year old woman with long black hair. She is inside a dark space.” The resulting image was decent and I used it.
I moved on to my heroes trying to catch a tram. AI kept showing them running from a tram. I love how in this one everyone else in the scene is running, too, and the woman on the left is using a chair as a walker?
And then there is this one: Harry Potter has used magic to turn himself into a tram so that he might crush his enemies.
I will leave you with this creepy ad that came up in my Instagram feed. The product is language software, but the ad uses AI to make famous people talk about the software (I have demonstrated this ability to make a still photo move and talk here). You can choose any of their celebrities to be your language tutor.
It was not the personalities but the uncanny valley effect that made me watch through the whole thing. The Steve Jobs is pretty good but the Benedict Cumberbatch will give you mightmares.
Hilarious as always. Thank you for spreading laughter when it is needed so badly!