- Sep 25, 2026
- 17
- 21
I know this topic has been beaten to death. Every few days someone starts a fight about it. But with recent events—WIT Studio's apology, Inio Asano pausing his serialization, Pixiv's crackdown—I think it's worth revisiting. I'm not here to start a flame war. I want to lay out the facts, the fault lines, and my own position.
Let me state my bias upfront: I draw for commissions. I have a clear problem with models trained on artists' work without consent. But I don't think "AI bad" is a useful conclusion. The reality is messier.
1. First, the facts: AI is already everywhere in ACG spaces
This isn't just about a few people generating fan art.
AI-generated motion comics / anime-style short dramas are already a huge market. According to Chinese industry reports, the "AI motion comic" sector hit around 16.8 billion RMB in 2025 and is projected to exceed 24 billion RMB in 2026. In January 2026 alone, over 14,600 titles launched—roughly 470 new titles per day. Production costs have been pushed down to about 500 RMB per minute. A small team of ten people can produce 50 episodes in 20 days. For comparison, even low-budget traditional animation costs thousands per minute.
The tools are improving fast. Platforms like PixAI are ranking higher and higher. Tencent is testing an AI product aimed at ACG fan works. Japan has even backed an open-source anime AI model, AnimeGen. Local deployment is mature enough that people say, "If your GPU can handle it, you can run whatever you want."
But the other side is brutal. In Q1 2026, active producers in the AI motion comic space dropped from 1,216 to 698—a 42% exit rate. Most creators earn less than 10 RMB per month. Less than 1% of the top creators take almost all the profit. That's not "democratization." That's a traffic oligopoly.
2. The real fight isn't "can AI be used?" It's "how?"
This forum often collapses "AI-assisted" and "AI-replaced" into one thing. I think we need at least three layers.
Layer 1: AI as a creation tool.
Drafts, references, modeling, background assistance. Many professionals already use it quietly. Inio Asano said in an interview that "as things stand, manga artists can't use AI. But in reality, if you use AI, the quality improves a lot." He used Unreal Engine and Blender for over 90% of his backgrounds in his new serialization. He thinks AI could speed up modeling. But he paused the series because the industry's attitude toward AI is so ambiguous. A creator stopping because nobody can say clearly whether AI is allowed—that's the cost of having no consensus.
Layer 2: AI-generated works published or sold.
This is what artists hate most. Pixiv's crackdown came after "AI faces" flooded search results. You search for a character and get pages of near-identical AI images, burying actual hand-drawn artists. Worse, some people use AI but label it "original." Some abuse R-18 tags for traffic. Pixiv's new rules are harsh: violating works aren't deleted, but they're hidden from search results. A soft shadowban. You can post hundreds of images, but only you can see them. And the platform won't publicly explain the criteria, so people can't game it.
Layer 3: Training on others' works, especially character LoRAs.
This is the clearest legal red line. In Shanghai's first AI model copyright infringement case, a user trained a LoRA on images of Medusa from Battle Through the Heavens and published it. The court found it infringed both reproduction rights and the right of communication through information networks. The logic: the essence of reproduction is "re-presentation." If a LoRA model generates content substantially similar to the input work under user commands, it infringes. The platform wasn't held liable because it provided a service with "substantial non-infringing uses" and couldn't foresee or avoid the infringement. But that's case-by-case. If a platform "knew or should have known," the result could differ.
3. The industry is more split than the forum
Here's the ironic part: people inside the anime industry disagree even more fiercely than random posters.
WIT Studio messed up badly. In Ascendance of a Bookworm, viewers noticed AI-generated backgrounds in the OP. After backlash, the studio admitted it, apologized, promised to remake the segment, and said it would "in principle" not use generative AI in production. But they also admitted they used it because of "insufficient production management and checking." So even a studio that had rules against it couldn't resist the pressure of deadlines and cost.
Some Jump+ authors are openly positive. One said that if you fed previous works into AI, "the manga could take off directly." A commenter summed it up perfectly: "Publicly scold AI, privately use it."
Illustrators are splitting too. Veteran illustrator Aoi Nanase was attacked as "AI fallen" for saying that Comiket shouldn't exclude AI-generated works. Her argument: if we use "copyright infringement" to suppress AI too aggressively, doujin creation itself could become a target. That logic is worth thinking about. The legal boundary between fan works and AI training overlaps in uncomfortable ways.
And the data is ugly. An NBER study found that after AI art tools launched, illustrators' uploads and attention dropped significantly. Creators began avoiding tags that AI preferred and expanding the IPs they worked with to avoid direct competition. This isn't simple replacement. It's forced niche migration.
A Japanese survey found 22% of creators said their income had fallen because of generative AI, and 49% expected further declines over the next three years. That's not just an economic problem. It's a "why am I even drawing?" problem.
4. My position: I'm not anti-tech. I'm anti-power-imbalance.
I don't oppose AI as a tool. If AI lets creators focus on what actually needs a human—narrative, storyboarding, emotional expression—then it has value. Using AI to generate background assets and using Blender to model backgrounds isn't fundamentally different in the sense of "using a tool to create."
What I oppose is two things.
One: unauthorized mass scraping and training. Artists' work is silently eaten by models, and then the model generates similar images to compete with the original artists. That's not "neutral technology." That's an unfair distribution of benefits.
Two: AI content flooding the ecosystem. When one person can generate hundreds of images a day and a hand-drawn piece takes hours, competing through natural traffic is impossible. Pixiv's new rules are essentially trying to repair that imbalance—not banning AI, but requiring honest labeling and limiting how much AI content disrupts normal search order.
But I also worry about overcorrection. Some artist circles treat any use of AI as "theft" and attack newcomers who use AI assistance. That purity spiral won't stop the technology, and it drowns out the real questions: copyright, compensation, and platform governance.
Aoi Nanase's point still nags at me: if "protecting copyright" is used to ban AI-generated content today, the same logic could be used to ban doujin tomorrow. Once copyright law tightens because of a new technology, it may not only tighten against the side you want to target.
5. Questions I think we should keep discussing
1. Can platform labeling actually work? Pixiv's soft shadowban seems effective for now, but AI images are getting harder to detect. What about false positives against hand-drawn artists?
2. Where exactly is the legal line for LoRA training? The Medusa case made commercial use a key factor. What about personal, non-commercial training? We need more cases.
3. When will the AI motion comic bubble burst? 470 new titles a day, 42% of producers exiting, most creators earning under 10 RMB a month. Is this "democratization" or a meat grinder?
4. How do we solve the artist identity crisis? Some professionals say their daily job is "running AI and retouching AI." They no longer feel the joy of drawing. If 22% already lost income and 49% expect worse, this isn't just about money. It's about why anyone would keep drawing.
Final thought:
Every AI thread eventually turns into team sports. But the real questions aren't "is AI good or bad?" They're: How are the rules set? How are the benefits distributed? How are creators protected?
The technology is already here. You can't curse it away. What we can do is make sure it doesn't become a tool for a few to harvest the many.
Rational discussion welcome. No personal attacks.
Let me state my bias upfront: I draw for commissions. I have a clear problem with models trained on artists' work without consent. But I don't think "AI bad" is a useful conclusion. The reality is messier.
1. First, the facts: AI is already everywhere in ACG spaces
This isn't just about a few people generating fan art.
AI-generated motion comics / anime-style short dramas are already a huge market. According to Chinese industry reports, the "AI motion comic" sector hit around 16.8 billion RMB in 2025 and is projected to exceed 24 billion RMB in 2026. In January 2026 alone, over 14,600 titles launched—roughly 470 new titles per day. Production costs have been pushed down to about 500 RMB per minute. A small team of ten people can produce 50 episodes in 20 days. For comparison, even low-budget traditional animation costs thousands per minute.
The tools are improving fast. Platforms like PixAI are ranking higher and higher. Tencent is testing an AI product aimed at ACG fan works. Japan has even backed an open-source anime AI model, AnimeGen. Local deployment is mature enough that people say, "If your GPU can handle it, you can run whatever you want."
But the other side is brutal. In Q1 2026, active producers in the AI motion comic space dropped from 1,216 to 698—a 42% exit rate. Most creators earn less than 10 RMB per month. Less than 1% of the top creators take almost all the profit. That's not "democratization." That's a traffic oligopoly.
2. The real fight isn't "can AI be used?" It's "how?"
This forum often collapses "AI-assisted" and "AI-replaced" into one thing. I think we need at least three layers.
Layer 1: AI as a creation tool.
Drafts, references, modeling, background assistance. Many professionals already use it quietly. Inio Asano said in an interview that "as things stand, manga artists can't use AI. But in reality, if you use AI, the quality improves a lot." He used Unreal Engine and Blender for over 90% of his backgrounds in his new serialization. He thinks AI could speed up modeling. But he paused the series because the industry's attitude toward AI is so ambiguous. A creator stopping because nobody can say clearly whether AI is allowed—that's the cost of having no consensus.
Layer 2: AI-generated works published or sold.
This is what artists hate most. Pixiv's crackdown came after "AI faces" flooded search results. You search for a character and get pages of near-identical AI images, burying actual hand-drawn artists. Worse, some people use AI but label it "original." Some abuse R-18 tags for traffic. Pixiv's new rules are harsh: violating works aren't deleted, but they're hidden from search results. A soft shadowban. You can post hundreds of images, but only you can see them. And the platform won't publicly explain the criteria, so people can't game it.
Layer 3: Training on others' works, especially character LoRAs.
This is the clearest legal red line. In Shanghai's first AI model copyright infringement case, a user trained a LoRA on images of Medusa from Battle Through the Heavens and published it. The court found it infringed both reproduction rights and the right of communication through information networks. The logic: the essence of reproduction is "re-presentation." If a LoRA model generates content substantially similar to the input work under user commands, it infringes. The platform wasn't held liable because it provided a service with "substantial non-infringing uses" and couldn't foresee or avoid the infringement. But that's case-by-case. If a platform "knew or should have known," the result could differ.
3. The industry is more split than the forum
Here's the ironic part: people inside the anime industry disagree even more fiercely than random posters.
WIT Studio messed up badly. In Ascendance of a Bookworm, viewers noticed AI-generated backgrounds in the OP. After backlash, the studio admitted it, apologized, promised to remake the segment, and said it would "in principle" not use generative AI in production. But they also admitted they used it because of "insufficient production management and checking." So even a studio that had rules against it couldn't resist the pressure of deadlines and cost.
Some Jump+ authors are openly positive. One said that if you fed previous works into AI, "the manga could take off directly." A commenter summed it up perfectly: "Publicly scold AI, privately use it."
Illustrators are splitting too. Veteran illustrator Aoi Nanase was attacked as "AI fallen" for saying that Comiket shouldn't exclude AI-generated works. Her argument: if we use "copyright infringement" to suppress AI too aggressively, doujin creation itself could become a target. That logic is worth thinking about. The legal boundary between fan works and AI training overlaps in uncomfortable ways.
And the data is ugly. An NBER study found that after AI art tools launched, illustrators' uploads and attention dropped significantly. Creators began avoiding tags that AI preferred and expanding the IPs they worked with to avoid direct competition. This isn't simple replacement. It's forced niche migration.
A Japanese survey found 22% of creators said their income had fallen because of generative AI, and 49% expected further declines over the next three years. That's not just an economic problem. It's a "why am I even drawing?" problem.
4. My position: I'm not anti-tech. I'm anti-power-imbalance.
I don't oppose AI as a tool. If AI lets creators focus on what actually needs a human—narrative, storyboarding, emotional expression—then it has value. Using AI to generate background assets and using Blender to model backgrounds isn't fundamentally different in the sense of "using a tool to create."
What I oppose is two things.
One: unauthorized mass scraping and training. Artists' work is silently eaten by models, and then the model generates similar images to compete with the original artists. That's not "neutral technology." That's an unfair distribution of benefits.
Two: AI content flooding the ecosystem. When one person can generate hundreds of images a day and a hand-drawn piece takes hours, competing through natural traffic is impossible. Pixiv's new rules are essentially trying to repair that imbalance—not banning AI, but requiring honest labeling and limiting how much AI content disrupts normal search order.
But I also worry about overcorrection. Some artist circles treat any use of AI as "theft" and attack newcomers who use AI assistance. That purity spiral won't stop the technology, and it drowns out the real questions: copyright, compensation, and platform governance.
Aoi Nanase's point still nags at me: if "protecting copyright" is used to ban AI-generated content today, the same logic could be used to ban doujin tomorrow. Once copyright law tightens because of a new technology, it may not only tighten against the side you want to target.
5. Questions I think we should keep discussing
1. Can platform labeling actually work? Pixiv's soft shadowban seems effective for now, but AI images are getting harder to detect. What about false positives against hand-drawn artists?
2. Where exactly is the legal line for LoRA training? The Medusa case made commercial use a key factor. What about personal, non-commercial training? We need more cases.
3. When will the AI motion comic bubble burst? 470 new titles a day, 42% of producers exiting, most creators earning under 10 RMB a month. Is this "democratization" or a meat grinder?
4. How do we solve the artist identity crisis? Some professionals say their daily job is "running AI and retouching AI." They no longer feel the joy of drawing. If 22% already lost income and 49% expect worse, this isn't just about money. It's about why anyone would keep drawing.
Final thought:
Every AI thread eventually turns into team sports. But the real questions aren't "is AI good or bad?" They're: How are the rules set? How are the benefits distributed? How are creators protected?
The technology is already here. You can't curse it away. What we can do is make sure it doesn't become a tool for a few to harvest the many.
Rational discussion welcome. No personal attacks.
