AI Tries to Take Over Scanlation Groups' Jobs, Only to Find They Don't Rely on Translation to Make a Living

36Kr Auto Digital Kr Future Consumption Intelligent Emergence Future City Launch Power on 36Kr Global 36Kr Research Institute Tide 36Kr Enterprise Service Review 36Kr Finance Workplace bonus 36 Carbon Post-Wave Research Institute Dark Waves Hard Kr Kr Rui Research Institute Media Brand

Enterprise Accounts Enterprise Service Review 36Kr Research Institute 36Kr Innovation Consulting Enterprise Services

Core Services City Window Government Services

Venture Capital Release LP Source Plan VClub VClub Investment Institution Database Investment Institution Job Referral Investor Certification Investor Services

Event Search Project Recommendation I Want to Join City Cooperation AI Grabs Jobs, Halfway Through Grabbing, Discovers They Don't Make a Living From This At All ACGx · October 21, 2026 18:05 AI participates in fan translation, still requires human completion and handling; fan translation groups still have a reason to exist.

We've heard enough stories about AI grabbing jobs, but I never expected that in the field of fan translation, translation groups could still keep their jobs by translating 18+ content.

Not long ago, a fan-translated manga spread on the internet. In the place where the translation group's name should have been, it had instead become a series of AIs like DeepSeek, GPT, Qwen, and so on. More interestingly, at the very bottom of the credits list there was even a step called "human feedback," which really made people gasp in admiration.

From acquiring the source images to final typesetting, models such as DeepSeek, Qwen, and GPT respectively took on key tasks including collection, upscaling, translation, and typesetting. This means translation groups will from now on be replaced by AI. But in fact, things are not that simple.

Because in a corner no one noticed, AI happened to leave translation groups one job: translating content like "Sorry, I cannot translate explicit X behavior descriptions."

For a long time, translation groups have been an important bridge connecting overseas markets with Chinese players. Especially for works without official Chinese versions, players who want to understand the story often can only wait for fan translations. And most of these works are niche, or just released, or partly 18+ works.

Behind a complete fan-translated work is often a relay of a group of people: some translate, some proofread, some handle image editing and lettering. When encountering works with large amounts of text, even after working for a long time they may not necessarily deliver the finished product in time. Most importantly, many translation groups are originally powered by love. Members gather together perhaps simply because they like a certain work and want more people to understand it. There is no stable pay, nor necessarily a fixed update schedule; everyone works when they have time and releases it when done.

Until machine translation began to become widespread. Content that previously needed sentence-by-sentence translation can now first be handed to machines for processing, and then translation group members proofread and correct it. Although early machine translation often carried a heavy foreign-language word order, and the Chinese it produced read awkwardly and could even produce quite a few jokes—for example, last year in Mingmo: Yuanyu zhi Yu, a plaque read "Mingxin Jianxing," and foreigners, in order to understand the plaque's meaning, had Google translate it, and in the end the translation became "human reproductive behavior makes people understand each other," directly turning philosophical Zen meaning into late-night fare. Although machine translation has big problems, it at least solved one practical issue: first make the work understandable.

Now, AI large models have pushed this one step further. AI can not only translate, but also understand character relationships in context and try to restore characters' tones. Combined with tools such as OCR, some work that originally required collaboration among multiple people has also begun to have the chance to be integrated into automated workflows. In the past, players had to wait for translation groups to complete a whole set of work; now, some people can already try to handle foreign-language manga themselves, and may not even need to wait for a complete fan translation patch to be released.

If the biggest trouble with early machine translation was translation quality, then the new problem brought by AI large models is the boundaries of AI use. Especially in some indescribable scenes, it appears particularly dramatic. When AI encounters content it is unwilling to handle, things become awkward.

It clearly has a certain level of language understanding ability, and it can already recognize the original text, yet it may directly refuse to continue translating because of content restrictions. Besides some indescribable content, there is also some content with obvious copyright issues. Thus, a topic originally discussing technological progress suddenly becomes black humor about employment.

In the past, translation groups lacked people, translation progress was delayed, and players could only urge updates; now, AI seems already prepared to take over the work, but may actively stop work in some places. At this time, the translation group's advantage has unexpectedly become: at least I am willing to finish this job. Of course, refusing to translate is only one joke. Even if AI is willing to continue working, it does not mean it can directly deliver a qualified fan translation.

In fact, even for works that have been fan-translated through AI, if another "human feedback" step is not added, many problems will still appear. Because fan translation itself is not translation; especially for content such as anime and games that highly depends on context, character relationships, tone, puns, and cultural references can all become disaster scenes.

For example, what should originally be teasing between characters comes out like a serious quarrel, and anyone reading that would be baffled. Not to mention the image processing problems that exist in manga. Because Japanese typesetting and Chinese typesetting are not completely the same, some sentences that originally fit may become crammed together after being changed to Chinese, and disaster strikes.

But what matters more is “meme play”: it is hard for AI to grasp memes that circulate only among humans or among fans of a particular work, as well as those extremely localized “in-jokes.” Take Call of Duty 19: Modern Warfare 2 from a few years ago: the game’s Chinese localization was about three or four kilometers away from what a normal person would do. Vulgar transliterations like “motherfucker” and internet slang such as “I really will thank you” made many players feel pulled out of the game’s story.

Although it seemed to give players a “little surprise,” in fact this kind of translation instead made many players very dissatisfied. They said one after another that such inappropriate internet memes not only failed to match the original text in word meaning and context, but also broke the immersion the game brought.

However, more often than not, what fan translation groups have to do is not just translate the text. Skilled fan translation groups also need to use their own brushes to put clothes on some scantily clad characters in manga, solely so that the translated manga can pass platform review...

For example, in the Malaysian edition of the Attack on Titan manga in recent years, the originally naked giants were “put into” underwear and briefs.

So in today’s AI era, for fan translation groups, it is instead work methods and efficiency that have been greatly improved. What fan translation groups need to do more is make users read naturally and understand clearly: when memes should be used to fill in, and what kind of emotion should be highlighted when. Only in this way can AI be used well. After all, AI is responsible for translation, and fan translation groups are responsible for cleaning up the mess. This can also be considered a new form of human-machine collaboration.

Final thoughts

Of course, in the end, fan translation groups are fundamentally a group of people running on love, so the vast majority of fan translation groups have no income. The Supreme People’s Court, the Supreme People’s Procuratorate, and the Ministry of Public Security state very plainly in their opinions on intellectual property infringement: directly or indirectly charging fees by placing paid advertisements in others’ works or bundling third-party works, among other methods, may be deemed “for the purpose of profit.” Therefore, once they accept paid advertisements, fan translation groups cross from “sharing among fellow fans” to “infringement for profit.”

As for fan translation groups that truly run on love and hope to find fellow fans, what they want is not money or traffic, and they certainly do not need efficiency; they can even accept being replaced. Because everything they do is simply share the content they like, hoping more people will see it and be drawn to join as fellow fans. What fan translation groups truly want to translate may never have been only the text in the works, but also that code phrase: “So you like this too.” As long as there are still people willing to stay up late proofreading and adjusting the layout for a work, just so another total stranger can understand it, then even if AI really takes over the whole translation job, fan translation groups may not necessarily lose their reason to exist.

This article comes from the WeChat public account “allan203” (ID: acgxclub), author: ACGx, published by 36Kr with authorization.

The views in this article represent only the author himself, and the 36Kr platform only provides information storage space services.+15

Good article, we need your encouragementACGxGuest author0Favorite+10CommentOpen WeChat “Scan,” open the webpage and click the share button in the upper right cornerWeiboImmersive readingBack to topReportJoin the commentsComments are countless; friendliness comes firstLog in to join the discussionSubmit comment0/1000You may also like these articlesAfter the new economy takes over, when will the AI productivity dividend be delivered?Snatching FDE for 10,000 yuan a day hides the cruel truth of AI implementationAre you sliding into AI’s “hidden trap”?There has never been any such thing as a Personal AgentTranssion invests in Today AI to explore the combination of personal AI assistants and intelligent terminalsTo sell products, Doubao spoke firstThe first wave of AI-generated Jensen Huang concerts, and UP hosts have already made enoughIn 3 months, 5 “outsiders,” a $200,000 budget, and with Claude’s support, “internet-famous research institute” Citrini designed a “scorpion venom-resistant protein”A new high-net-worth group has emergedACGxGuest author

TA has not written a bio, but being reserved is also a form of expressionHas published 333 articles

Recent contentAI grabbed jobs halfway, only to find that they don’t rely on this to make a living at all2 hours agoAfter years of serialization ending, Naruto has held back another big move?2026-10-12Non-game IP-adapted games earn 6 billion a year, and the first place is actually not Spider-Man2026-09-24Read more content, poke here hardView AI reviewsDeepSeekChoose reliable AI and look at real reviewsView 36Kr AI review official communication communityJoinConsultationProject review and onboardingContact 36Kr project recommendationsSubscription accountFollow

Next articleSpaceX profitable for the first time, net profit $4.4 billion; Musk talks about merging with Tesla: a good idea

SpaceX profitable in Q3, AI business props up half the sky, massively expanding computing power

36Kr Europe36Kr Europe36Kr EuropeAI Product DailyOnline rumor information report entranceHot recommendations

Push and interpret cutting-edge, substantive tech venture capital news

Primary market financial information and system service provider

Focus on outstanding global entrepreneurs, project financing rate close to 97%, leading the industry