‘Blue Archive’ EP Rejects Generative AI Art In “Cute Girl Games”, Says Players Dislike “Images That Are Simply ‘Clicked Into Existence'”
As the capabilities and availability of generative AI continues to expand, Blue Archive Executive Producer Yongha Kim has warned that while said technology may have some beneficial applications in the field of video game development, using it to create art assets is not one of them – especially when it comes to gacha-style games centered on ‘cute girls’.
Kim, who has overseen the development of Blue Archive ever since it was first proposed in 2018 by Nexon Games CEO Park Yong-hyun, offered his thoughts on the topic of generative AI in video games as part of a recent open lecture given to participants of South Korea’s 2025 Content Creative Talent Mentoring Program.
Aptly titled ‘Developing Cute Girl Games in the AI Era’, Kim’s talk was not a one-sided condemnation of AI, but rather an in-depth reflection on using it effectively, intelligently, and in a way that wouldn’t torpedo a given studio’s reputation.
“Users who enjoy character games do not prefer AI-generated content. I also feel a strong aversion to spending money on images that are simply ‘clicked’ into existence,” he told the crowd, as recapped by South Korean news outlet GameMeca and machine translated via DeepL. “Therefore, I would like to share my thoughts on how to approach AI in this genre.”
From there, the producer recalled even further instances in which AI would only cause trouble for a given developer, as based on his own personal AI-video game experimentation.
These included level design:
“I thought it would be great if AI could handle level design, so I tried having it create puzzles. I had it try the [classic Japanese] puzzle game Sokoban, where you move boxes to fill in spaces, but even the latest models struggled when the difficulty level increased slightly. If it can’t solve puzzles, creating them is even more difficult.”
The animation process for in-game models:
“Rigging must always be done before animation, but since it’s repetitive work, I tried it based on a paper claiming that AI could do it well. When I actually tried it, I found that the bones were attached in an abnormal way, making the character unable to move. I thought it might work with a little more effort, so I tried it for a couple of months before giving up.”
And any such ‘novel’ mechanics, with Kim raising the example of voice recognition:
“We incorporated voice recognition for users into Focus on You [a 2019 VR photography title he worked on during his time with his previous employer Smilegate], and to do so, we introduced an AI service. However, maintaining this service incurs costs, and it was difficult to include those costs in the game’s price. Even though it is a packaged game, the structure requires users to pay each time they play, so the voice recognition feature has been removed.”
In summarizing his AI caution, the Blue Archive EP posited to his audience that, when all was said and done, the technology was not ready for the widespread handling of the video game mediums’ mechanical and personal intricacies.
“When people think about using AI for development, they might imagine that AI can draw images, write dialogue, and code, and that the game will just ‘pop out’ with a click. In fact, many such games are available on Steam. However, these games don’t sell well, receive poor reviews, and aren’t fun to make. I don’t think this is the right approach to game development.”
However, as noted above, Kim’s attempted tempering of AI expectations was not based in the sentiment of ‘All AI bad’, but instead ‘Playing to its strength’, with the Blue Archive team making use of Large Language and Large Image Models for clerical and ‘randomization’ work:
“We also provide services in Japan and China. Currently, there’s a feature that translates comics when you input them. It reduces the communication burden and repetitive tasks that arise when multiple people work together.”
“[The animation that plays when you unlock a character] is a composition that zooms in on parts of the character, which is a necessary task every time a character is created. Since it’s somewhat repetitive, we thought about training it to automate the process and tested it. When we actually tried it, we found that it could be used as a tool to create cutscenes by highlighting notable parts of the character. This allowed us to complete a task that previously took two days in just 30 minutes.”
“We also considered providing users with unique experiences through AI, rather than just reducing repetitive tasks. Currently, this is applied in Blue Archive, where Arona calls the user’s name. Originally, we considered automating ‘lip-sync animation,’ which synchronizes character mouth movements with pre-recorded voice lines. However, feedback indicated that in scenes where the character’s face is prominently displayed, the results did not match the intended outcome and required significant adjustment. Instead, we thought that synchronizing the mouth movements when Arona calls the teacher’s name would be acceptable.”
Drawing his thoughts to a close, Kim ultimately told the crowd that when it came to AI and video games, “We need to approach this from the perspective of how we can improve the convenience of those who are currently developing.”
“I think that a position that provides focused support for such work is extremely important. As such a position, I believe that the roles of machine learning engineers and technical artists will become increasingly important in game development.”
