---
title: "AI Can Generate Dozens of Episodes a Day, But the Most Expensive Thing Just Got Pricier"
author: "Unique Research"
sourcePublication: "Unique Research Substack"
originalPublishedAt: "2026-09-18T17:07:30+00:00"
canonical: "https://ffcap.cn/en/research/ai-can-generate-dozens-of-episodes"
source: "https://uniqueresearch.substack.com/p/ai-can-generate-dozens-of-episodes"
language: "en"
---

# AI Can Generate Dozens of Episodes a Day, But the Most Expensive Thing Just Got Pricier

AI didn’t make content cheaper. It just moved where the cost lives.

“One sentence to generate an entire animated episode.”

When I first heard that claim, my reaction was the same as most people’s: the entire economics of animation are about to collapse. An episode of motion-comic drama used to take weeks. Now it’s one sentence. Isn’t content about to become dirt cheap?

That wind is definitely blowing. AI-generated motion-comic dramas are hot right now. Content platforms are hungry for supply, and more people are piling in with money, as if whoever generates fastest will corner the whole business.

After sitting down with Wang Jun, CEO of Anigo.ai, I realized the math is backwards.

Anigo has been in the content industry for years — animation, motion comics, IP work. Their representative titles have racked up over 150 million views across platforms. Public filings mention roughly 10 million RMB in 2025 revenue. The team is small — the kind of content veterans who were forced by AI to rebuild their own production line from scratch.

The veteran’s verdict: **AI didn’t make content cheaper. It just moved where “expensive” lives.**

I asked when he first realized that AI wasn’t just making the production team more efficient — it might rebuild the entire animation production line.

His answer was precise to a timeframe: the end of 2025, when image-to-video and character consistency actually became usable. “Before, the question was ‘can it be drawn?’ Now it’s ‘decide what to draw.’”

“When the most expensive step on a production line shifts from execution to judgment, you can’t treat AI as a faster brush. You have to redefine every stage’s output as something machines can read and verify. That’s the moment I knew the line had to be rebuilt.”

For the past hundred years, the bottleneck of animation industry has been hands: how fast, how detailed, how stable. The entire industry’s talent structure, production workflow, and pricing system were all built around the word “execution.” Fast hands, steady hands — those were the people worth paying.

Now drawing belongs to machines. The most expensive step in a production has shifted from “how well it’s drawn” to “knowing what should be drawn.”

Every handoff on the line must be in a form machines can read and verify, or the next step simply can’t pick it up.

This is why they ultimately chose to build Anigo themselves instead of just buying off-the-shelf image and video generation tools.

“Tools are purchasable. Production lines are not,” Wang said.

There’s a popular formula in the industry right now: **Agent = Model + Harness**. Model capability is the same for everyone — it comes from the same few providers. The harness is the entire set of workflows, standards, checks, and memory you build around the model. That layer is different for every company.

Translated into plain language: the model is a rented engine. The harness is the car you build yourself. Everyone can rent the same engine, but how the car drives depends entirely on your own craftsmanship.

“The interesting thing about the vibe-coding era is exactly this: every company has different production workflows and methods, and the agents that grow out of them are unique,” Wang said. “I keep telling our team: we’re not using AI to make content. We’re turning content capability into AI. Anigo isn’t a model we built — it’s our own production methodology turned into a system that does the work.”

“One sentence, one episode” sounds simple enough from the outside, but in between sit world-building, characters, scripts, settings, storyboards, video, voiceover, and editing — a long chain.

I rephrased the question: without using words like “Agent” or “workflow,” how would you explain to a completely non-technical comic creator what Anigo actually does for them?

His answer stripped out all the technical vocabulary:

> “Simple. Just ask them: you know how to be a boss, right? You know how to spot problems? You know how to say ‘figure this out’ or ‘this doesn’t feel right’? Everyone’s done this. Everyone has experience with it. Just treat Anigo as your intern. If you’re not satisfied, just say so. You don’t need to know how to draw. You don’t need to know how to edit. You just need to assign tasks and critique the output.”

That’s the most unorthodox — and most accurate — explanation of an AI creation tool I’ve ever heard.

Anyone who’s ever assigned a task to a subordinate at work and said “this doesn’t feel right” already has all the skills needed to use Anigo.

Of course, there needs to be an honest measure of what “one sentence” actually achieves.

Wang’s metric is “number of human-AI interactions.” Theoretically, once an idea is settled, every remaining decision can be handed to AI, all the way through to the final cut — zero interventions in between, just that first interaction. The cost is high uncertainty.

If you want certainty, you increase the number of interactions and participate in the creative and visualization stages.

In plain terms: one interaction and a finished episode is buying a lottery ticket. Multiple interactions is running a business. Tools always advertise the former. Clients always pay for the latter.

I asked a question the industry has debated for a long time: why not build a completely free-form chat Agent, instead of structuring source text, settings, script, storyboard, and final cut into a defined production pipeline? Which matters more — freedom or certainty?

“I want freedom in content, because that’s the soil for good work. But I want certainty in process, because industrial workflows haven’t fundamentally changed in a hundred years.”

He said Anigo does have a chat Agent. The difference is whether there’s a locked-in pipeline behind the conversation.

“Behind our dialogue is a locked production line. When you say ‘redo this section,’ it knows exactly which steps after the storyboard need redoing and what doesn’t need to be touched. Chat provides freedom. The pipeline provides certainty.”

This design solves a real pain point. Production in the AI era is so fast that “it used to take weeks to make one episode, and people could remember where the project was. Now you might produce dozens of projects in a single day.”

That’s why Anigo includes a resident producer Agent that tracks every project’s stage, what assets are missing, and which shots need rework. You just talk to it and tell it to get things done.

I pushed harder: video models are updating so fast. If a much stronger model appears tomorrow, does Anigo’s value get erased?

“It gets stronger,” he said. “Model capability is model capability. Delivery capability is delivery capability. The widely accepted view is Agent = Model + Harness. The Model is raw intelligence. The Harness is what turns intelligence into work-doing capability. No matter how strong the video model gets, it only answers ‘can this segment be generated better?’ It doesn’t answer ‘how should this episode be made?’”

I agree with this judgment. Model vendors are in an arms race. The stronger the models get, the more leverage the layer above — process and memory — has.

What’s truly scarce was never the brush. It’s the person who knows how to stage the entire episode.

When the conversation turned to money, Wang’s words got noticeably denser.

“’How many people, how many days per episode’ calculates the cost of commodity work. But nobody buys commodity work anymore — everyone wants premium content. The cost per episode is no longer ‘how much to make it,’ but ‘how much to make it right.’ Two years ago, the math was a production ledger. Now it’s a judgment ledger.”

Then came the line I most wanted to bold in this entire interview:

**“On judgment, AI hasn’t saved us a single cent. In fact, because the density of decisions has increased, it’s become more expensive.”**

This is perhaps the most direct dismantling of the “AI reduces costs and boosts efficiency” narrative.

AI cut execution costs, but the number of decisions on every production line has exploded. Before, one episode a week meant a few dozen decisions. Now, dozens of projects a day, each requiring human sign-off.

More decisions, and every decision still needs a human. The production savings get handed right back to more expensive judgment.

This also explains a phenomenon that puzzles many people: generation tools keep getting cheaper, but good content hasn’t gotten cheaper. What got cheap is the “making it” part. What got expensive is the “making it right” part.

Commodity work pricing will keep collapsing to the bottom. Premium work pricing, by contrast, now has a reason to hold.

Failed shots are another ledger outsiders can’t see. I asked how many out of ten generated shots actually survive. He wouldn’t give a number, and his reasoning was practical: “With the same model, someone who understands it deeply has a very low failure rate. Someone who doesn’t understands, nine out of ten are wasted. Both types really exist, so the number is meaningless.”

But he did share failure patterns: ensemble scenes with multiple characters, seams between two video segments, and fine contact points in fight scenes — these are still the most likely to break.

So claims like “AI reduces content costs to one-tenth” miss two things: the wasted portion, and the judgment portion. The former is money. The latter is more expensive money.

When it comes to hits, his attitude is equally calm. Anigo has made content with 150 million views, but he said: “AI still can’t judge for you what’s worth making. But it has changed how judgment works: before, an idea took weeks to produce before you knew if it worked. Now you can produce three openings in a few days and test them. So hit-making capability hasn’t been replaced — it’s been accelerated.”

Notice the shift: before, testing an idea meant betting weeks of production budget. Now, you can put three openings in front of the market in days.

The cost of trial and error has dropped. The responsibility of the final call has gotten heavier, because there are more options to choose from, and choosing wrong still hurts just as much.

I planted a hypothetical question in the interview: a founder with zero animation experience, holding 500,000 RMB, ready to start an AI motion-comic company. Should the money go to models, team, IP, or distribution?

Wang’s answer was one sentence: “I’d probably tell him not to enter the market yet.”

That’s the last thing anyone in the hype narrative wants to hear. Everyone asks “where should the money go?” He dismantled the question: don’t spend it yet.

But stringing together the earlier ledgers, this answer follows logically: models don’t buy production lines, money doesn’t buy judgment, and judgment is something you only develop after years immersed in content.

500K buys tool subscriptions. It doesn’t buy “knowing what’s worth making.” When the most expensive cost in the industry becomes judgment, an inexperienced person walking in with cash is like showing up to a road-reading contest with a tank of gas.

I also asked him the classic question: what’s the cognitive difference between people who actually make content versus people who only build AI tools?

His categorization was “content people who learned tech” versus “tech people who learned content.” The former start with technology and learn content; they’re most sensitive to what models can do. The latter spend years in content first, then learn technology; they’re most sensitive to what audiences want.

“Both types are valuable. But I think the one difference is: the technology side is being leveled. Content judgment is hard to level, so it becomes more scarce,” he added. “Of course, I came up through the content side, so this view might be biased.”

Bias or not, the ledger is self-consistent: technology is being commoditized. Judgment is appreciating.

Following the money down led to the people question. Anigo itself is a small team, yet it’s reached eight-figure revenue and nine-figure view counts. As AI goes further, how many people does an efficient motion-comic company need?

“Fifteen to twenty people is enough. And you only need one kind: the polymath,” he said. “Before, a production group had many people each handling their specialty. Now this person needs to understand story, visuals, pacing, audience — and command AI. That’s the person getting more expensive. They’re rare in the market.”

The jobs of screenwriter, director, storyboard artist, art director, and producer are being encapsulated into a single system. The person left behind is more like an individual managing an AI film crew.

This isn’t good news for the job market, but it’s a clear signal for people building composite skills: single-skill roles are being eaten by the production line. People who can span story, visuals, and pacing while directing AI — those are the ones whose salaries are going up.

My last question was broad: if in two or three years, “one sentence, one episode” becomes baseline capability, and hundreds of thousands of new episodes appear every day, what becomes the new scarce resource? IP, aesthetics, distribution, or user attention?

“A non-consensus view: when everything can be done fast, that’s exactly when you shouldn’t be fast.”

“Hundreds of thousands of new episodes every day. Not a single one will be remembered. Being remembered is the scarce commodity in this business. So what we should build isn’t more content. It’s turning content into IP, into series — making audiences remember a name, follow a character, instead of scrolling and forgetting. A name that’s already been remembered is worth more than a hundred mediocre new concepts.”

In an era of production inflation, attention isn’t the scarce resource. **Being remembered is.** When everyone else is pressing the gas, what’s truly valuable is knowing where it’s worth hitting the brakes.

This is about motion comics, but it holds for every industry being swept into the AI production race.

At the end, I asked him: if Anigo truly succeeds in three years, what do you hope it ultimately becomes — an AI motion-comic tool, an AI content company, or an AI studio anyone can use?

“I think none of those. Those words grew up in the pre-AI era. In the AI era, anything is possible — at least those three aren’t what I want.”

“What I want doesn’t have a name yet today. If I had to put it: what Anigo brings isn’t just making animation faster. What I hope more is that **stories can, for the first time, not end.**“

_Stories can, for the first time, not end._

After someone who’s spent over a decade in content rebuilt his entire production line, that’s what he wanted.

Production capacity can be infinitely replicated. Judgment can’t. Neither can a name that’s been remembered.

* * *

**Guest: Wang Jun, CEO, Anigo.ai**

**Q1: Anigo wasn’t founded in 2025 or 2026. You’ve been doing animation, motion comics, and IP for years. When did you first realize that AI wasn’t just boosting efficiency but might rebuild the entire animation production line?**

Wang Jun: I’d say it was around the end of 2025, when image-to-video and character consistency became usable. I realized the bottleneck had moved — before it was “can it be drawn?” now it’s “decide what to draw.” When the most expensive step on a production line shifts from execution to judgment, you can’t treat AI as a faster brush. You have to redefine every stage’s output as something machines can read and verify. That’s the moment I knew the line had to be rebuilt.

**Q2: Why did you ultimately choose to build Anigo yourselves rather than just using off-the-shelf image and video generation tools internally?**

Wang Jun: Because tools are purchasable. Production lines are not. Everyone’s been saying Agent = Model + Harness: model capability is the same for everyone, from the same few providers. The harness is the entire workflow, standards, checks, and memory you build around the model — that layer is different for every company. The most interesting thing about the vibe-coding era is that every company has different production methods, and the agents that grow out of them are unique. I keep telling our team: we’re not using AI to make content. We’re turning content capability into AI. Anigo isn’t a model we built — it’s our production methodology turned into a working system.

**Q3: Anigo’s pitch is direct: “one sentence generates an animated episode.” Without using words like Agent, workflow, or AIGC, how would you explain to a completely non-technical comic or novel creator what Anigo actually does for them?**

Wang Jun: Simple. Just ask them: you know how to be a boss? You know how to spot problems? You know how to say “figure this out” or “this doesn’t feel right”? Everyone’s done this. Everyone has experience. Just treat Anigo as your intern. If you’re not satisfied, just say so. You don’t need to know how to draw. You don’t need to know how to edit. You just need to assign tasks and critique.

**Q4: You’ve done both content and tools. From your current vantage point, what’s the biggest cognitive difference between people who make content versus people who only build AI tools?**

Wang Jun: We’ve discussed this before. You’re asking about tech people who learned content versus content people who learned tech. The former start with technology and learn content — they’re most sensitive to what models can do. The latter spend years in content first, then learn technology — they’re most sensitive to what audiences want. Both are valuable. But the one difference is: the technology side is being leveled. Content judgment is hard to level, so it becomes more scarce. Of course, I came up through the content side, so this view might be biased.

**Q5: From a story or novel to a watchable episode, there are world-building, characters, script, settings, storyboard, video, voiceover, editing — many steps. How much can Anigo actually automate today?**

Wang Jun: The degree of automation is abstract. I’d reframe it as “number of human-AI interactions.” Theoretically, once an idea is settled, every remaining decision can go to AI, all the way to the final cut — zero interventions, just that first interaction. The uncertainty is high. If you want certainty, you increase interactions and participate in the creative and visualization stages.

**Q6: Why did you choose not to build a completely free-form chat Agent, but instead structure “source text → settings → script → storyboard → final cut” as a defined production process? In real content production, which matters more: freedom or certainty?**

Wang Jun: Both freedom and certainty are important. I want freedom in content — that’s the soil for good work. But I want certainty in process, because industrial workflows haven’t fundamentally changed in a hundred years. Actually, Anigo does have a chat Agent. The difference is whether there’s a locked pipeline behind it. Behind our dialogue is a locked production line. When you say “redo this section,” it knows exactly which steps after the storyboard need redoing and what doesn’t need to be touched. Chat provides freedom. The pipeline provides certainty.

**Q7: Anigo has a resident producer Agent that tracks project stage, missing assets, and shots needing rework. Why does the next phase of AI video need a “producer” who truly understands the whole project, rather than more generation buttons?**

Wang Jun: Production in the AI era is too fast. Before, one episode took weeks and people could remember where the project was. Now you might produce dozens of projects in a single day. So we want a producer that monitors all projects, helps move them forward — you just talk to it and tell it to get things done.

**Q8: Video models are updating extremely fast. If a much stronger model appears tomorrow, does Anigo’s core value get replaced faster, or does it get stronger? What moat do you truly want to build?**

Wang Jun: It gets stronger. Model capability is model capability. Delivery capability is delivery capability. The widely accepted view is Agent = Model + Harness. The Model is raw intelligence. The Harness is what turns intelligence into work-doing capability. No matter how strong the video model gets, it only answers “can this segment be generated better?” It doesn’t answer “how should this episode be made?”

**Q9: Compared to traditional production two years ago, how much has the economics of delivering a commercial-grade motion-comic episode changed today?**

Wang Jun: “How many people, how many days per episode” calculates the cost of commodity work. But nobody buys commodity work anymore — everyone wants premium content. The cost per episode is no longer “how much to make it” but “how much to make it right.” Two years ago, the math was a production ledger. Now it’s a judgment ledger. On judgment, AI hasn’t saved us a single cent. In fact, because the density of decisions has increased, it’s become more expensive.

**Q10: There’s an easily overlooked cost in AI video: wasted shots. What types of shots are still most likely to fail today?**

Wang Jun: With the same model, someone who understands it deeply has a very low failure rate. Someone who doesn’t understands, nine out of ten are wasted. Both types really exist, so the number is meaningless. As for which shots fail most often — we have plenty of experience. Ensemble scenes with multiple characters, seams between two video segments, fine contact points in fight scenes — those are the usual culprits.

**Q11: Your representative work has exceeded 150 million views across platforms. After making content with real reach, how do you see the relationship between “AI generation capability” and “hit-making capability”? Can AI judge what content is worth making?**

Wang Jun: AI still can’t judge for you what’s worth making. But it has changed how judgment works: before, an idea took weeks to produce before you knew if it worked. Now you can produce three openings in a few days and test them. So hit-making capability hasn’t been replaced — it’s been accelerated.

**Q12: Suppose a founder with zero animation experience walks in with 500,000 RMB to start an AI motion-comic company. What’s their biggest risk? Where should the money go?**

Wang Jun: I’d probably tell them not to enter the market yet.

**Q13: Public info shows Anigo itself is a small team, yet you’ve achieved eight-figure revenue and nine-figure views. As AI further enters production, how many people might an efficient motion-comic company need in the future? Which roles will clearly shrink, and which will become more expensive?**

Wang Jun: Fifteen to twenty people is enough. And you only need one kind: the polymath. Before, a production group had many people each handling their specialty. Now this person needs to understand story, visuals, pacing, audience — and command AI. That’s the person getting more expensive. They’re rare in the market.

**Q14: If in two or three years, “one sentence, one episode” becomes baseline capability, and hundreds of thousands of new episodes appear daily, what becomes the new scarce resource? IP, aesthetics, distribution, user attention, or a repeatable hit-making method?**

Wang Jun: A non-consensus view: when everything can be done fast, that’s exactly when you shouldn’t be fast. Hundreds of thousands of new episodes every day. Not a single one will be remembered. Being remembered is the scarce commodity. So we shouldn’t build more content. We should turn content into IP, into series — make audiences remember a name, follow a character, instead of scrolling and forgetting. A name that’s already been remembered is worth more than a hundred mediocre new concepts.

**Q15: If Anigo truly succeeds in three years, what do you hope it ultimately becomes — an AI motion-comic tool, an AI content company, or an AI studio anyone can use to manage a complete virtual crew?**

Wang Jun: I think none of those. Those words grew up in the pre-AI era. In the AI era, anything is possible — at least those three aren’t what I want. What I want doesn’t have a name yet today. If I had to put it: what Anigo brings isn’t just making animation faster. What I hope more is that **stories can, for the first time, not end.**

[![cover](https://substackcdn.com/image/fetch/$s_!nrap!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fb0e59e50-525b-44de-a53d-43cd7a25d526_2048x1152.jpeg)](https://substackcdn.com/image/fetch/$s_!nrap!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fb0e59e50-525b-44de-a53d-43cd7a25d526_2048x1152.jpeg)

[![cover](https://substackcdn.com/image/fetch/$s_!nrap!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fb0e59e50-525b-44de-a53d-43cd7a25d526_2048x1152.jpeg)](https://substackcdn.com/image/fetch/$s_!nrap!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fb0e59e50-525b-44de-a53d-43cd7a25d526_2048x1152.jpeg)

---

Original publication: https://uniqueresearch.substack.com/p/ai-can-generate-dozens-of-episodes
On-site reading page: https://ffcap.cn/en/research/ai-can-generate-dozens-of-episodes