I love this photograph. It’s shot by my niece, with my dad, sister and her husband surrounding my Mom. She didn’t hear that this was a “silly pose” picture, and was composing herself before flashing her usual family portrait smile when my niece shot the photo. She wasn’t humorless, but a lot of funny stuff left her unmoved. Now, every time I see it, it makes me smile.
I was casting around, looking for something to do on a slow Good Friday. Playing with AI image creation? Sounds good.
I started slow, using chatCPT 4o, using the prompt:
Please draw this image as a cartoon in the style of South Park characters, without risking a copyright violation.
Right off the bat, I had something that was pretty damned good:
Next, I turned to a Peanuts-style cartoon, using the same AI. Here’s that prompt:
Could you draw the image again in the style of a cartoon, this time in the style of a Peanuts cartoon. Again, please do so without breaking copyright rules.
This time, the results were pretty good, but not as impressive as the South Park version:
I decided to give Google’s Gemini Advanced 2.0 Flash a try with the same prompt.
It did a better job of catching the style, but it totally broke the “taken as a selfie” composition, instead showing a picture of a group taking a selfie. How meta…
I decided to give Dr. Suess a try with the following prompt:
Please draw this image in the style of a Dr. Seuss cartoon, without breaking copyright rules.
ChatGPT didn’t get the assignment at all, drawing a generic looking cartoon, far from the style I asked for. I think ChatGPT did a decent job. I’m not sure who the photo-bomber is.
At this point, I got cocky, and tried to tell Gemini to switch places around in the image, and threw in a metaphor:
Please take the image and create a new realistic image in which each of the participants is shifted, like musical chairs.
That kind of poisoned things in that conversation. From that point on, all of the images it generated featured everyone fighting over a chair. I couldn’t make it forget it.
I shifted strategy. Let’s try for an artistic drawing:
Please render the attached image as an artistic drawing in pencil. Please make the drawing look like the original image as much as possible. In other words, keep each of the people posed in the same way, and with the same expressions they had in the original image.
Everyone got very serious, mostly older, and my neice was cloned, with different haircuts. Which do you like better?
I went back to ChatGPT, and started playing with other styles.
Please render this image in the style of an Elizabethan portrait.
Here's another idea. Using the original image, could you replace the background so that is appears that they are riding in a subway car in New York city?
Here's another. Change the background on the original image so that they look like they are standing at the top of Mount Everest, with appropriate clothing
Please do another, but make all of the people in the photo dressed as cavemen in animal-skin clothing, as if in a play, with dinosaurs and volcanoes in the background.
Please dress each of the people as if they were in "The Sound of Music" and change the background to the Alps.
For some reason, the dinosaur came along!
Mom’s kept a smile on my face all day now!
I hope you’ve enjoyed my AI image generation experiments!
That is great! Mom stayed in character for every pose. Needed to put a smile on her face but that would take her out of character.