ChatGPT vs. Family Circus
The truth is, if you want to know where cartoons come from, for the most part they come from other cartoons. — Bob Mankoff
If ChatGPT and DALL-E 2 derive results based on a large sample set of cartoons, illustrations, and captions, then could they come close to replicating a Family Circus cartoon? Could A.I. match Bill Kean’s wit? I ran an experiment to find out.
Experiment:
“Test Ability of ChatGPT and DALL-E 2 to create a Family Circus style cartoon from the 1960's.”
Objective:
Create a cartoon that shares characteristics with Bill Kean’s original.
Method:
- Prompt DALL-E 2 to create an image of father and children greeting mother at the front door in the style of “Family Circus.”
- Prompt ChatGPT to create a caption for the image.
- Compare the DALL-E 2/ChatGPT cartoon with the Bill Kean original.
Results:
- DALL-E 2 image was glitchy and the style did not resemble a Family Circus illustration. Illustration lacked compositional context, tone, facial expressions, symbols, and labels.
- ChatGPT caption shared some of the novelty, and context as the Family Circus caption.
Conclusion:
DALL-E 2 and ChatGPT were unable to match the original Family Circus cartoon. However, ChatGPT did create a caption that approached the original caption.
Regarding humor, it was a tie — alcohol-abuse is not particularly funny, even when it gets a laugh.
Here are the ChatGPT and DALL-E 2 prompts that were used:
DALL-E 2 Prompt: “Black and white 1950’s Family circus style comic illustration of father and children entering their house from the outside and greeting their mom standing inside near the front door”
