Graphic Novel

Küss den Frosch

Which AI is best for image generation?

For this project, we were supposed to create a graphic novel in groups and use AI to generate the images. I worked in a group with Chris, Lynn, and Johanna. So, the first step was to decide which AI we wanted to use for our project. Each group gave a short presentation on one AI tool. My group presented ChatGPT, but we also heard presentations about MidJourney, Adobe Firefly, Imagen, and Stable Diffusion. This way, we were able to learn something about each of the tools, and it turned out that MidJourney and ChatGPT Plus would be the most suitable for our project. Our group decided to go with ChatGPT.

Story development

The next step was to come up with the story for our graphic novel. We had the idea of taking an existing fairy tale and rewriting it a bit. We chose The Frog Prince. But instead of a happy ending with a prince, the protagonist ends up in a psychiatric hospital. The story was supposed to be funny and slightly disturbing.Here are some screenshots from our figma board that we created during our creative process (we wrote the story in German):

Image generation

Once we had outlined the storyline, we divided the scenes among us and started generating the images. This part could sometimes be frustrating. Often, ChatGPT needed up to five minutes to generate a single image, and then it didn’t even look the way we wanted. But after several attempts, we learned how to send more specific prompts to the AI and were able to achieve some continuity in the characters. Still, the protagonist doesn’t look exactly the same in all the images. Another thing is that all images generated with ChatGPT had a slightly brownish tone—but that didn’t bother us too much since it somehow matched the style we were going for. A final problem was the text. We wanted to have text visible in some images, for example in speech bubbles. But ChatGPT is really bad at generating text, so we had to edit all images with text afterwards in Photoshop.

Website development

As a last step, we programmed the website where we wanted to showcase the images. Our goal was to make the website look like a physical graphic novel. So the website looks like an open comic book, and you can flip forward or backward using two buttons. The design of the website is kept very simple, since we didn’t want to distract from the content.

Conclusion

This project was the most fun for me, because we were able to work in groups, coming up with the story was really entertaining, and generating the images was also a lot of fun—it was cool to see how ChatGPT brought our vision to life. On top of that, I learned a lot about different AIs in terms of image generation, which will definitely be useful again in the future.

repository