On Touhou 20 and ZUN's Opinion on AI
Now that Touhou 20 is out, I thought I'd make an article explaining the usage of AI in the game as well as ZUN's opinion on AI. I hope this article will clarify things for you and you'll be able to decide whether to stick around or to move on without any lingering questions or doubts.
As a preamble, there is one thing that makes discussion of AI really difficult, and it's the fact that some people single out genAI as a specific type of AI, while others lump them all together as just AI. The reason this makes things difficult to discuss is because some people think all AI is unacceptable, while some only have a problem with gen-AI, and this heavily affects not only one's own perception of AI, but also how they perceive other people's perception on AI. In ZUN's case, it appears that he's the type to lump them together.
With that out of the way, I hope reading this article will allow you to form your own conclusion on things.
ZUN's Opinion on AI
Let's start with ZUN's opinion on AI. What's important to note is that ZUN's thoughts on AI has been forming for quite a while, and predates Gen-AI. I believe the first time he mentions AI would be in the afterword of Dr. Latency's Freak Report.
Here he mentions how as time goes on, AI might eventually be able to do the things that humans can, but that relying on it isn't everything, and there are still fun things that can't be found with technology. This line of thought would be continued in WBaWC.
Within the plot of the game itself, the human spirits were at the mercy of the animal spirits, until Keiki and her haniwa came. At first the haniwa helped the humans survive and thrive against the animals, but eventually, the humans began serving the haniwa instead. The allegory is clarified in the SCoOW interview, that being "humans becoming subservient to something that's meant to bring benefit to them".
The above train of thought then leads to UDoALG, where beasts and AI are once again front and center in the game. This time, Zanmu is the AI allegory. She's seeming capable of predicting everyone's actions and come up with the perfect plan to neutralize any situation. Due to her abilities, she's able to walk all over the three yakuza factions and stop their planned invasion of Gensokyo before it even begins.
The game ends with Zanmu defeating Reimu but deciding to leave Gensokyo in Reimu's hands, as she finds newfound appreciation in Reimu's perspective on the situation. This ending reinforces the message of the game : Even if humans do not have the capabilities of AI, there is still value in leaving things in human hands.
The above message is discussed in both the afterword and the SCoOW interview. The impression one gets from reading it is that, in ZUN's opinion, AI has more or less reached the level that he thought it would have way back during Dr. Latency's Freak Report, and while that does pose an economic threat to creators, it does not pose a creative one, as the usage of AI does not satisfy one's creative drive, making it just a useful tool for efficiency.
"AI is just a useful tool" is the opinion he currently holds and is part of the reason why he used it in Touhou 20. To be more specific, his opinion is "No matter how much AI improves, it will never be more than just a useful tool", and that is the message he's trying to send.
Touhou 20 Demo Discourse, Touhou Station Stream
Back when the Touhou 20 demo was first released, and the AI assets were discovered, there was a lot of fervor and discourse on whether or not it had been used intentionally or accidentally. Things only escalated when someone found a summary of the drunk livestream he had in December of 2024.
At the time, I decided to dispatch someone to ask the people who had attended the stream for more information, as there were some parts of the summary that I didn't fully understand, and I felt that I had to verify as much as I could before I could actually comment on the situation. Thus, I decided to not say anything about the situation until I had learned everything I could. Looking back, I probably should have at least tweeted out something like "I'm aware of the situation, I'm currently looking into it, please wait as I verify the information". Perhaps had I done that, the discourse would not have escalated to the level that it did.
It did not help that one of my tweets became a huge point of discussion that ZUN wouldn't use AI because in a drunk stream AI meant "Alcohol Infinity". But this stream was held all the way back in 2018, years before AI being used in games became a huge point of contention. https://www.youtube.com/watch?v=E2JvZR6y75E
He sometimes holds drunk livestream at venues, whenever he does this, the stream is named after topics in the gaming industry, sometimes using puns. An example of this was 3D standing for Dangerous Drunk Dimension in 2012, and of course AI for Alcohol Infinity in 2018. In hindsight, the moment people started QRTing that tweet, I should've immediately clarified the situation.
When the steam page for the demo was finally released, and we had confirmation that AI assets were indeed used, things even further escalated to the point where we started getting articles about it. Still, there was no explanation on the reasoning behind the usage that came directly from ZUN. Only speculation and inherences.
It would not be until 2 weeks later, on the 23rd of May 2025, that ZUN appeared on a Touhou Station livestream, and the topic of AI being used was brought up by BeatMARIO and Kanamai. This is when we finally get to hear his reasons for using AI. While I have made a thread on the stream (https://x.com/richard_effendi/status/1926243695503773947) , I'll once again cover his statements. This time, I'll be translating the relevant parts directly, rather than summarizing them and using MTL for the images.
Do note that I'll skipped over a lot of the banter and in-between conversations, if you'd like to see the full conversation, it's available in the following link :
Touhou Station Stream Summary
Kanamai : By the way, the steam page shows that the game uses AI...
ZUN : Yes, I thought a lot about that. I actually wanted to talk more about it after the full release.
Right now, there's a lot of dispute on AI, in all sorts of production. Personally I think the most important thing is to not lose to AI. That's something I've always said. When it comes to creating things, entrusting everything to AI would be to lose to it. However, rejecting it simply because you don't like it would also be losing to it. Should I be acknowledging AI as something to be feared, or should I accept its existence? I was split between the 2. The reason I used AI here, is because I wanted to reduce AI to a mere tool.
"AI is nothing but a useful tool".
That's why I wanted to express that I was definitely not going to lose in the creative parts. Plus, it's a bit of a spoiler, but for thematic purposes, I absolutely wanted to use AI for this game.
ZUN : I wanted to say this after the full release, but I fully intended to use AI here.
And I think people can also tell, I used it at the place where it didn't matter, the backgrounds that would look the same regardless on who made it, to put it bluntly, the part that has always been just made from photos or stock assets. It would've been easier to not use AI, but as a way of expressing my intentions, I wanted to use it to deliver a message.
I wanted to aggressively use AI and reduce it to a tool. That's my firm intent.
ZUN : I don't want people to get the wrong idea. I didn't use AI for the creative parts. Because if I were to use it, then it wouldn't be fun anymore.
There are people in the world who proclaim themselves AI artists, but people like them are being used by the AI who have surpassed their own abilities. We should not become like that, but to completely reject AI would also be to lose to it.
Nothing but a useful tool, AI is nothing but a useful tool.
So long as it's the kind that doesn't violate copyright.
ZUN : I also have a more personal reason for using AI.
Nowadays, a lot of fields require youngsters to use AI. We're at an era where you have to learn how to use AI for work, and that's being taught everywhere. Because of that, I also need to let my children learn how to use AI.
"I absolutely do not use AI. However, I do teach my kids how to use AI". Doing something like that would be contradictory, so I need to teach them that this would be the correct way to use AI. That's why I can't just close my eyes and go "I don't want to see AI, I don't like it".
ZUN : Collecting other people's art and teaching an AI to generate something similar, like the Ghibli stuff, that sort of thing is no good.
For a lot of people, their image of AI is probably like that, that's why they dislike it, but I think there are better types of AI out there.
That's about it for the parts of the stream relevant to this article. I've said this before, while I understand the perspective, I do find it to be one made from a place of privilege, a place that has only encountered AI issues related to copyright and creativity, rather than the much more pressing social, political and environmental issues that it has brought to countries like the United States.
That said, since only the issues of copyright and creativity were the ones ZUN discussed for his reasons on using it, the rest of this article will also only be limited to discussing those issues. Apologies for being unable to fully address the other issues related to AI.
ZUN, Backgrounds, and Stock Images
After the 23rd May 2025 Touhou Station Stream, there was some controversy regarding what ZUN said about backgrounds. A common takeaway that people had with ZUN using AI for backgrounds was that it meant he thinks that backgrounds are unimportant. This discourse existed in both the English and Japanese fandom.
Side note : if you're curious on how ZUN used backgrounds up until this point, here are some websites that collect exactly where he got some of the assets he's used in his games. Basically all of them are royalty free stuff from books, asset packs, the internet, or pictures he's taken of nature, usually stacking multiple of them together or adding filters to change their appearance.
https://ohgosho.ehoh.net/
https://nrr08.com/background.html
https://kiribumekki50.wixsite.com/th-hsifs
https://kiribumekki50.wixsite.com/wbawc
https://kiribumekki50.wixsite.com/ucn-mkts
ZUN : "I used it at the place where it didn't matter, the backgrounds that would look the same regardless on who made it, to put it bluntly, the part that has always been just made from photos or stock assets."
I believe that last part makes it clear that rather than saying that backgrounds didn't matter/were unimportant, it was more that because his backgrounds have always used stock images, it didn't matter what was being used for it. We can actually see this same sentiment in the SCoOW interview for Touhou 19, where he lists looking for images on the internet as separate from making things with his own hands.
While I don't agree with the sentiment, i also find it really difficult to refute it. Because unfortunately, there have been multiple examples of works and creators being mocked for using stock assets. It also doesn't help that criticism of stock image usage is expressed in the form of statements like "I can make the same exact thing in 30 minutes".
In cases where the usage of stock assets isn't mocked, the best it gets is a backhanded compliment or a compliment that is directed to the skill of the user, rather than the assets themselves. "The usage of stock assets does not detract from the quality of the work" or "The creator was actually able to use stock images in a creative way".
Even my own thread on his usage of stock images gave praise in this exact same nature, especially with "Making stock images look like presentable backgrounds". In saying that, I had inadvertently suggested that stock images are not presentable backgrounds. In general, usage of stock assets is mainly considered a shortcut or a compromise, its presence subtractive to the value of the work at worst and only neutral at best, never an addition.
And this is the case with all sorts of stock assets, be it images, music, animations, 3D models. Regardless of the quality of the assets themselves, the simple fact that they were premade for the purpose of being sold, rather than tailor made for the project, reduces their worth to a shortcut. I think the only asset to escape this fate is probably stock sound effects or sound fonts. Presumably because they're the ones that most people simply can't make themselves, requiring resources and equipment that the layman wouldn't have access to.
Even with the rise of genAI, I can't say that people's appreciation for stock assets/stock images have gone up. When people see a person making stuff with AI, the argument is always to make it themselves, or commission someone to make it. There aren't people batting for stock images and going "look at all this stuff that's already available to use".
There's definitely a discussion to be had regarding the merits of stock assets, but I think that warrants an article of its own, and I've already dedicated too much space rambling about it in this one.
AI in Touhou 20 (Warning : Spoilers)
Touhou 20 makes extensive use of AI in its backgrounds, both in spell card BGs, and in stage BGs, replacing the usual stock images. I believe it is used in every spell card BGs, the BGs in stage 2, 3, 4, the first half of stage 5 (the 2nd half reuses LoLK's BG, since it's the same location), most of stage 6 (It appears that the torii is not), and the extra stage. To explain their purpose in this game, I'll need to first explain the story.
The Plot of the Game
The gist of the story goes as follows : The Lunarians have a facility in the sanctuary that's made for the purpose of collecting information from living creatures, processing that information and sterilizing to remove the kegare, and then filtering the processed information to the Lunar Capital.
This facility was set up by way of the Lunarians forcefully converting an existing structure for their own needs, imprisoning one of the gods, Iwanaga Ariya, who originally created the structure deep underground within the structure, while subjugating another one of the gods, Yuiman・Asama, into doing their bidding, making Yuiman the facility's AI, in charge of processing the information and filtering out the kegare.
Due to events in the Outside World, the amount of information entering the pyramid has increased, surpassing Yuiman's processing ability. Thus, unfiltered information began reaching the Lunar Capital, threatening the Lunarians' eternity with kegare.
In order to rectify the situation, Toyohime decides to unseal the other god of the structure and have her use her ability to reject change on Gensokyo, in order to freeze the torrent of information and kegare currently flowing to the moon. This plan is merely a stopgap solution though, and Toyohime hopes that Eirin would be able to find a more permanent one, as this stopgap also has the side effect of causing the days to start repeating in Gensokyo (Not in a groundhog day way, as time does keep moving, it's just that the same events keep happening). Should this keep up, Gensokyo would become the same as the Lunar Capital, devoid of kegare.
Eirin does indeed find a more permanent solution in the form of the incident stones, stones that contain the same type of magical energies as the ones that have appeared in past incidents. These stones were crafted by Misumaru, and then delivered to the protagonists by Yukari. Upon receiving these stones, Reimu/Marisa become able to perceive the unchanging days, gaining awareness that an incident is occurring, and that the incident has something to do with the sanctuary.
The protagonists head off to the sanctuary and enter the pyramid, where they encounter and defeat Yuiman, seemingly freeing her from the mind control that the Lunarians had placed on her. After defeating Yuiman, the protagonists set off deeper into the pyramid, eventually getting warped to the sea of tranquility, where they meet with Toyohime.
Toyohime explains the situation to the protagonists and tells them that defeating her would not stop the incident in Gensokyo, as the one causing the days to loop was not her, but Iwanaga. Toyohime then sends the protagonists to where they need to go to confront Iwanaga and put a stop to the incident.
Once they arrive at Iwanaga's location, Iwanaga explains the events leading up to the incident, about her nature as a goddess of permanence, her ability to reject change, and that defeating her is necessary to end Gensokyo's incident. Once the protagonists defeat her, Iwanaga also warns that with her power no longer active, a flood of kegare would flow into Gensokyo, and that it was up to Gensokyo to decide how the kegare would change them. Naturally, the Gensokyo simply accepted the kegare with no noticable change to anything.
The Themes of the Game
With the plot explained, we can finally talk about the themes of the game and the usage of AI. Though, as a disclaimer, I would note that what you're about to read is the interpretation I got after reading the plot. It is possible that I misinterpreted ZUN's intentions or misread some lines of dialogue, we will likely find out more the next time ZUN goes on stream. In any case, I'll begin.
I've seen people say that AI in the game represents corruption, impurity or kegare, and that's why the backgrounds look uncanny, to represent the corruption from AI. However, I think that even if the AI art had looked perfect and indistinguishable from human art, it still would've accomplished its role in this game. It was not important whether the art looked like AI, what was important was that it was made by AI. Because AI in this game doesn't represent corruption or kegare. Instead, AI represents the lack of kegare, and stagnation.
The incident instigator in this game is the goddess Iwanaga Ariya, a goddess of permanence and rejecting change, while the principal antagonists of the story are the Lunarians, a society that has decided to remove all traces of kegare from their environment, attaining them an eternal and unchanging existence. Both of Iwanaga and the Lunarians are entities who abhor change, and this aversion to change lead to disastrous consequences for both of them.
For Iwanaga, her nature of rejecting any sort of change led to her being unable to do anything to improve her or Yuiman's situation for eons, and she was forced to resign herself to her fate, accepting that she will simply be left alone, sealed underground beneath her own pyramid.
For the Lunarians, while their form of eternity means that they're do not suffer from aging or disease, they have also completely stagnated as a society, unable to improve. Their situation is likened to death by both Yuyuko and Orin.
The Lunarians and Iwanaga both embody stagnation, and this stagnation, this inability to progress, is represented through AI art, as those who rely solely on AI will find themselves stagnating, unable to progress themselves due to not allowing themselves to experience the creation and learning process in favor of having AI do everything for them.
In opposition to Iwanaga and the Lunarians embodying purity and stagnation, we have the incident stones embodying kegare and change. It is due to the incident stones' ability to allow the protagonists to experience and accept change that the incident was able to be resolved. And the benefits of change is not restricted to Gensokyo either, even Iwanaga and the Lunarians were able to benefit from change.
In Iwanaga's case, her continuous rejection of any sort of change meant that no matter how much time had passed, she was not able to do anything to solve her and Yuiman's predicament. It is only when she finally accepts a change instead of rejecting it, that she was able to do something about the situation.
In the Lunarians case, as a result of their stagnation, they are unable to resolve any of their problems by themselves. In SSiB, LoLK, and this game, they would not have been able to successfully tackle their respective situations without the assistance of Eirin, someone who's now rejected their eternity and is living on the kegare-filled Gensokyo.
The inciting cause of the incident may have been the overwhelming amount of kegare/impurity flowing into Gensokyo, but Touhou does not treat kegare as a bad thing. That is not to say that kegare is a good thing. after all, kegare is the reason why things are able to degrade and decay. However, it's also due to kegare that things are able to improve. Kegare isn't good or bad, it is simply something that comes with being alive, it is part of human nature.
This is why the AI backgrounds in this game does not represent the impurity/corruption brought on by the information coming from the outside world, but rather the stagnation brought on by the removal of kegare, namely due to Iwanaga's power. That said, both Iwanaga and Yuiman are also allegories for AI in this game, as both have the ability to remove kegare/ the human element from their environment, so saying that the AI backgrounds are a sign of AI influence would not be an entirely inaccurate statement.
Where Iwanaga and Yuiman represent AI, the incident stones seem to represent fan-works. Throughout the multiple endings of the games, the various incident instigators the stones are based on comment on how the stones give off the same energies as the youkai they're based on. The stones themselves are made by Misumaru with directions from Eirin, whom Yukari notably chooses to refer to as "people who love Gensokyo". According to the credits, the stones were also made by Ki-san instead of ZUN.
In other words, the incident stones are :
-A creation derivative of characters and events from the series
-Made by a master craftsman who was not the creator
-Made due to their love of Gensokyo
Marisa even notes that the incident stones are copyright infringement in one of her endings.
There's one notable thing about ZUN's comments on AI. For all of his criticisms on its usage, I don't believe he's ever had an opinion on the quality of its output, nearly all of his criticisms are focused on how unsatisfying it is to use, and the dangers of relying on it. It appears that for him, the thing that separates AI art from human-made art has nothing to do with appearance.
What makes a piece of AI art, AI art is not what it looks like. Even if it manages to make something that looks exactly like something a human would make, it would still be AI art. Similarly, even if a human were to make all of the mistakes and habits an AI would make, they would still not be making AI art. Because regardless of what the final product looks like, a human was still involved in the process of its creation. What makes AI art, AI art is that it wasn't made by a human.
I do think that the usage could be more explicitly explained in the afterword, similar to AFiEU, a book that criticizes right wing rhetoric by featuring Aya attempting to make a magazine echoing those rhetoric. However, where AFiEU dedicated a huge chunk of the afterword explaining what the book was about, FW's doesn't clarify anything about the theme at all. Instead, most of the explanations are from the May 23rd 2025 Touhou Station stream. For example, he does specify that we shouldn't use AI that trains itself of data used without permission, however that part is not stated in the game itself, and one would only be aware of it by watching the stream.
https://note.com/yshell_38k/n/nc4ac6209675d?sub_rt=share_pb
ZUN : AI is nothing but a useful too, so long as it's the kind that doesn't violate copyright
ZUN : Collecting other people's art and teaching an AI to generate something similar, like the Ghibli stuff, that sort of thing is no good.
Ultimately, whether or not you think the themes of the game justifies the usage of AI, and whether or not it was handled well, that is up to you to decide.
My Own Thoughts on the Future of AI
My personal opinion is, while the coveted "do everything AI" will probably not bear fruit, I do think that the technology itself will probably continue to permeate our daily lives. because it's now easy enough for a company to create their own specialized AI.
The main issue with AI like ChatGPT, OpenAI, Grok, etc is that these companies are trying to make a tool that's capable of everything, and they also want to make it as fast as possible, because they have to be the first one to achieve it, less the market gets monopolized by the competitor first. So they're ignoring every moral, ethical, legal, and environmental quandaries, trying to achieve something that should take decades to achieve to happen in years. Such a thing is unlikely to succeed, and I think it's more likely than not that they'll eventually run out of money before they can truly achieve that goal.
But if you're making a smaller more specialized AI that only needs to do one thing, you don't need to spend an ungodly amount of money or resources. The most common type of this that I've seen is object recognition AI, where different companies make an AI, teach it to recognize a specific object or parts of an object, then flag that object according to set parameters. I've seen this type of AI being used to sort fruits, to spot and tag defects in objects, to count the number of objects within a given container, and so on.
So while the big "do everything" AI might not last. I think the technology that they're built with will continue to be used to well into the future, in a more responsible (government regulated) manner. I just hope that whatever damage done in the meantime will be reversible.
Closing Statements
Before I end this article I do have a few things to say regarding inter-fan reactions to the entire AI situation. The main thing I'd like to say and emphasize is to avoid antagonizing each other over differing opinions and decisions.
Regardless of what people think about the situation and what they want to do in response, deciding on what to do next is their prerogative. If your opinions differ on something, the most you can do is discuss and share each other's perspectives by providing each other with the information that lead you to your respective conclusion. Whatever the other person does with the information you provided will be up to them. And whatever you do with the information they provided will be up to you.
That does also apply to myself and this article. Ultimately, I can't stop people from choosing to argue if they want to argue, the best I can hope is that the information I provided is enough to convince them to choose a more diplomatic solution.
I understand why people had an immediate and angry reaction to the usage as AI, as for a lot of people, especially those in the US, the impact and consequences of the technology is much closer and more dire compared to others. Whether that consequence is being laid off by their companies in favor of AI, or their environment steadily being destroyed due to the sheer amount of resources required by AI data centers to continue their operations.
For some of us, it might be easy to say "Isn't the problem with the companies, not with the technology itself?", but I feel like that would be doing the same thing as telling someone whose family has been a victim of gun violence that guns aren't the problem. It might not be a problem for us, but it's definitely a problem for them, and they can't just ignore it.
And so I'd like to reiterate, let those who want to swear off from the series leave in peace, and let those who wish to stay remain without a fight. Thank you for reading this article.


コメント
6It changed because I realized that the whole thing revovled around """"AI"""" and I thought it was absolutely genius and original. In this case, it made absolute sense.
But generally, I'm against """"AI"""".
Why all the quotes behind the acronym? Well, I'm sorry, what exists today is not an "artificial intelligence". A real AI would not need to have information ingested to learn, talk, thinnk or have any emotions. What's available today are LLM - large-language models - which are... (3/?)
... computer programs that have huge amount of data ingested and have it "learn" to regurgitate it in a different manner (very rough explanation I know). "LLM" is just not a good buzzword so a misleading definition of "AI" was more buzzworthy.
These extremely-flawed tools are replacing real people's jobs to drive extreme capitalism at the expense of everything else: people's wellbeing, the rest of the economy, the environment... (4/?)
There are also new psychological troubles being introduced by people trusting AI over everything else (which is covered in Touhou 20), people choosing shortcuts over learning (again covered in the past in Touhou albeit indirectly)... but also amplify other problems such as more sophisticated scams, spread of dis/misinformation, poverty, lost of jobs, etc.
(this 500 chars limit is annoying) (5/?)
As such, on top of corporation trying to shove their tool down our throat, I think that the downsides swallows all the good sides it could bring because it's always used for the worst intentions in mind: to replace the human touch on everything and taking advantage of others to pass their own agenda.
In the end, Zun used LLM to actually send the right message; not only in Touhou 20 but in previous games as metaphors.
However, I have ZERO faith for the general populace to use it correctly.(6/6)