In this project, my goal is to use DALL-E 2 from ChatGPT 4 to mimic the posters of animation films directed by Hayao Miyazaki and produced by Studio Ghibli. To begin with, I selected some official movie posters of Miyazaki’s films, ranging from “Howl’s Moving Castle,” “Princess Mononoke,” “Spirited Away,” to “Ponyo on the Cliff by the Sea.” After collecting Miyazaki’s representative movie posters, I uploaded them to ChatGPT 4 respectively with the prompt “generate a similar image in the style of Miyazaki.” I decided to explore this topic since I am a fan of Miyazaki’s movies and curious about whether AI can generate a certain film/animation/contemporary art style. By exploring this topic, I hope to learn more about AI’s performance in generating artistic images, how it relates to artists’ copyrights, and how we can turn AI into a useful tool without harmful impacts.
Photo Series 1:
Inspired by “Howl’s Moving Castle”
↑ The prompt I entered: generate a similar image in the style of Miyazaki
The first image is clearly what I expected. Too complicated and does not align with Miyazaki’s style. Peoples’ faces are omitted. So I tried again by giving a more specific prompt. Here is the result:
This one is much better, with a (Japanese?) animated style, but I don’t think it follow Miyazaki’s movie style. Although it looks fine at the first glimpse, there are lots of wired and flaws sections. For instance, the girl’s pigtail looks like a tail of a cow or certain animal’s. Also, the on right part, there is a blurry head that does not make sense to us. So I asked it to generate another one:
There is no major differences with the former one. Background and arrangement are quite alike. And I have no idea why does it so obsessed with including a cat in the image. Is it because of Miyazaki’s famous character Totoro?
Photo Series 2:
Inspired by “Princess Mononoke”
By using the same prompt, I expected ChatGPT 4 to give me a colorful images. But the result is not what I want:
I have no idea why the tone of this image is so dim and the composition looks so wired: two figures almost take up all of the space and background has been cut. Also, I am so confused about the gender of the character in the front, is it a girl or a boy?
Okay, this is an interesting result. I wanted a better composition by saying “setting the image in a natural landscape,” but there might be something wrong with my description. The result is totally another one that only show the back of the characters. (But actually not really, because we can see the cat’s face from its back, which is a bug.)
Photo Series 3:
Inspired by “Spirited Away”
One of the most famous of Miyazaki, the prompt is on the screenshot↑
I do not think the style is Miyazaki: it seems like 3D instead of 2D (based on its light, shadow, and tone). Characters have rigid posture with hollowed eyes. Again, ChatGPT4 depicted another cat-like creature, with a pair of long ears. The overall style reminds me of Western science fiction films: wired aliens or robots are presented in the background.
Photo Series 4:
Inspired by “Ponyo on the Cliff by the Sea”
I selected a cute poster of Miyazaki’s film “Ponyo” from the Pinterest↑
I would say this result is an awkward imitation. The color palette is similar, but the style is definitely not in the style of Miyazaki. There two figures seem copied from some random images in its data pool: they do not have any eye contacts or interactions.
I selected another poster of “Ponyo” from Ghibli Studio. It’s a pencil drawing, and ChatGPT copied this form:
It looks okay. But compared with the original picture, this one looks way symmetrical and concentrated. I would say it lacks of creativity.
Photo Series 5:
Inspired by “The Wind Rises”
I have no idea why it stuck here. I tried ten more times and used poster from different language version, but it did not work.
Leave a Reply