I prompted ChatGPT-4o to write a small dialogue in the style of a cult favorite movie (based on a comic book). To keep it a little meta, of course they were discussing art created by AI. It did really well at this task, coming up with quite a few answers that sounded very true to how the characters speak. (This is a fun exercise to try – prompt it with some text and ask it to rewrite it in the style of someone very distinctive.)

Next, I prompted it to draw a comic book page with these characters and one of the shorter dialogs. Kudos to ChatGPT that it immediately flagged this as a copyright violation. I switched over to another conversation with the same prompt, but this time it didn’t have the original movie name in its context, so it was not flagged. It did however repeatedly flag copyright violation when I gave the characters the same name as the characters in the movie, or when I prompted it with the names of the actresses and not the characters. Still doing well, ChatGPT.

I went through many results and tweaking of prompts. This image is the closest to what I had envisioned. Leaving aside how mind-blowing it is that it managed to generate this entire page from scratch, my conclusion from this experiment is that the model still has a lot of issues with following the instructions and rendering text. We aren’t quite at a place where you can give a detailed prompt and a script and it can generate a comic book for you effortlessly.

Can you guess who these characters are? 😉