Twitter in Japan has been buzzing about attempts made to generate an image of Sephiroth from Final Fantasy VII swimming using image-generating AI tools. However, the AI doesn’t seem to understand the concept of “swimming” and only makes images of Sephiroth standing on water.
Illustrator @deon_talk has been using the AI image generator Midjourney to create a series of fictional manga starring Final Fantasy VII’s Sephiroth. The manga features Sephiroth getting lost in a castle made of candy, serving ramen in the Shinra company cafeteria, and other settings far removed from anything the official Sephiroth has been seen in.
@deon_talk also tried to make an AI illustration of Sephiroth swimming. However, no matter what they tried, the AI didn’t seem to understand the concept of swimming. Sephiroth proved to be stubborn and @deon_talk posted the results of Sephiroth only standing on the water to Twitter.
The reason for this is thought to be the data used by AI tools. These image-generating AI tools study already existing illustrations and create images. In other words, if it isn’t a concept present in the illustrations it learns from, it can’t recreate it very well.
Drawing someone swimming is difficult, and those kinds of illustrations aren’t very common. And there certainly aren’t any scenes of Sephiroth swimming from The Compilation of Final Fantasy VII that come to mind. Because of the overwhelming lack of data, generating an AI image of Sephiroth swimming looks to be a tough task. There are many reports on Twitter of users trying to generate AI images of people swimming to lackluster results.
@deon_talk even tried directing it with specific poses and posted the frankly shocking results to Twitter. Sure enough, using AI to generate images of people swimming isn’t as simple as it sounds.
Other users reacted to seeing Sephiroth standing on water with comments such as, “Sephiroth doesn’t swim,” and “He’s just like Moses.” While there were some discussing the nature of AI images, most were just having a laugh looking at the odd pictures. Some users half-jokingly said that maybe the AI doesn’t want to force Sephiroth to swim.
Sephiroth is certainly an adored character with his one wing and extremely long sword. Even the AI might consider it blasphemous to take such a beloved character and try squeezing out an image of him swimming.
Afterwards, @deon_talk tried to have Sephiroth surfing, but he still persisted to defiantly stand on the water instead. The AI seems to see Sephiroth as too refined for even a surfboard.
In addition to swimming, there are other sets of images demonstrating what AI has trouble generating. One Twitter user posted the strange results when they tried generating an image of The Idolmaster Shiny Colors character Madoka Higuchi eating ramen. It looks like holding bowls and holding chopsticks while slurping noodles is difficult for AI.
The pros and cons of AI art are often brought up on social media. While it can bring about these sorts of fun posts, they can also be harmful to the image of the original works, so it’s probably best to just enjoy them as something akin to fanart. As we continue to watch trends in AI art, hopefully society will get to a place where the technology isn’t misused.
Written by. Nick Mosier based on the original Japanese article (original article’s publication date: 2022-10-18 20:23 JST)