cover imageAI Industry Observation
It costs 7 yuan to make, sells for 10 yuan, and acquiring one customer costs 50 yuan.
"Of course parents don't need AI-generated picture books."
It costs 7 yuan to make, sells for 10 yuan, and acquiring one customer costs 50 yuan. He says the math gets clearer the more you calculate it.
One AI interactive picture book — images, text, and voiceover all included — costs 7 yuan to produce, is priced at 10 yuan, with a 30% gross margin.
Acquiring one customer: 50 yuan.
These are the numbers Chen Shenghan reported to me himself. He is the founder of Infinite Dream Painting (无限绘梦), a company making AI interactive picture books, and has just completed his first year of entrepreneurship. Selling one book earns 3 yuan, acquiring one person costs 50 yuan — on paper, it's a losing proposition no matter how you calculate it.
But the first thing he said when we sat down to talk was even more striking than these numbers: "Of course parents don't need AI-generated picture books."
The CEO of an AI picture book company started by rejecting his own product. That afternoon, I originally only wanted to talk about picture books, but the conversation turned into something else entirely.
Chen Shenghan's background, laid out on paper, seems to have nothing to do with picture books.
Twelve years in international education, five years at a major tech company — product, operations, development, design, data — he's done it all at the front line. There's also a hidden thread: he has been drawing since childhood and held a solo independent art exhibition in high school.
Outsiders see it as disconnected, but he himself finds it perfectly logical: "Actually, from a subjective perspective, my decision isn't disconnected at all."
Twelve years of education experience made him think through the forms and essence of education; the major tech company gave him the full toolkit for building products; drawing was a seed planted from childhood.
What truly connected these threads was his time working as an English tutor. He walked into one child's bedroom after another and found the same scene: every bookshelf had a large number of picture books that had never been opened.
Parents spent thousands upon thousands buying books, and children flipped through two pages before tossing them aside. Not because they didn't love reading, but because there was no "him" in those stories.
Parents buy peace of mind; children want a version of themselves. The bookshelf became the middle ground where two generations' worries met.
He told me about Steve Jobs' famous saying, connecting the dots. When the AI era arrived, he felt all his scattered dots connected: "For me, AI is like a magic wand, and I'm ready to graduate from Hogwarts."
Of course, feelings alone aren't enough. Before launching the project, he conducted nearly half a year of user research in his capacity as a tutor, talking with over 100 families.
There were two conclusions: children love role-playing interactive experiences, and parents lack a means to cultivate their children's self-driven learning. Only after validating the demand did he take action.
Returning to that opening statement. Parents don't need AI-generated picture books, so what do they need?
His answer: "A one-stop collection of reading materials across multiple educational themes, readable anytime, anywhere, with soul." AI is merely a low-cost means to achieve this goal; the content framework must be designed, created, reviewed, and organized by humans before being fed into the workflow.
The division of labor is one sentence: "In the birth of every picture book work, I am the director, AI is the producer, and the child is the protagonist."
"In the birth of every picture book work, I am the director, AI is the producer, and the child is the protagonist."
Infinite Dream Painting currently covers 12 educational themes — academic subjects aligned with the new curriculum standards, and quality education covering financial literacy, leadership, and the like. Children make choices in stories, bear the consequences, and different choices lead to different endings.
In his words, the child isn't "reading" a story — they're "living" a story.
His definition of his own company is quite a mouthful: an experience company building a scientific decision-making engine based on an AI real-time interactive narrative technology framework. Three modifiers, each more intimidating than the last, but the core is actually one sentence: evolving AI's value from generating content to generating continuously running experiences, from a single storyline to an infinitely expanding parallel universe.
Picture books are just the entry point; this engine can eventually move into games, film and television, and social networking.
You might be wondering: isn't putting a child's name and photo into a story also personalization?
Yes, but that's like putting a candle with a name on a cake — the cake is still the same cake. The personalization he wants goes much deeper: using personalized characters, stories, and imagery to spark curiosity; using decision-making scenarios to train judgment; using the ability to restart at any time to provide a sense of security.
Every choice a child makes accumulates into preference data, pushing personalization further. The parent side can also customize the protagonist's background information based on actual circumstances.
He says the product was born from parents' real voices. Parents are tormented by the involution of academic education, anxious about the dizzying array of AI training camps, and exhausted by all kinds of extracurricular classes.
These emotions, when they reach him, are translated into three layers of pain points.
First, wasting money on various ineffective interest classes; second, being unable to precisely match learning materials to the child; third, having few topics for parent-child interaction, high communication costs, and no one recording the child's growth.
Actually, what parents are worried about is no longer improving academic performance, but whether their children can adapt to the new era. It's like the elephant in the room — everyone sees it, and I'm just helping them voice what's on their minds.
These three pain points, taken individually, aren't exactly new. Anxiety about classes, anxiety about materials, anxiety about companionship — which parent group hasn't discussed these?
What's new is that he attributes all three to the same root: parents fear their children will lose at grades, but fear even more that they'll lose at the times.
When we got to the production环节, I asked him the question outsiders are most curious about: Isn't an AI picture book just a large language model writing a story and matching it with a dozen images? What's so hard about it?
The hardest part is human-AI collaboration. Both humans and AI must find their irreplaceable positions. AI is the skeleton, humans are the blood — when role boundaries are模糊, what comes out can be worlds apart.
"AI is the skeleton, humans are the blood."
How exactly is it divided? Story templates, educational frameworks, art style and character design, model fine-tuning — every环节 that sets standards must involve humans. AI is responsible for script generation, image generation, voice synthesis, and multimodal alignment.
Behind it all is a whole set of things: multi-agent collaboration, educational theme libraries, multimodal hallucination handling, and parallel production mechanisms. He calls this what Nolan said: extreme complexity behind minimalism. Going further would involve trade secrets — he only revealed that patents have been applied for.
For age adaptation and visual aesthetics — areas that require professional expertise — he says the team's major tech company resources and educational background can cover it. He listed why he can be this referee: "Precisely because of my years of practice and accumulation, I understand what art styles children like, what stories move them, what teaching is effective, and what triggers different emotions."
For product direction and initial content judgment, his exact words were: "As CEO, I'm more than capable."
More precisely, they've cultivated AI into a dream factory with infinite generation capabilities, ensuring every dream painting aligns with expert-level aesthetics, writing, feeling, and logic. As for whether they'll be replaced by large models, his answer is: "We won't be replaced by large models because the process of polishing this product to perfection has no best, only better, and the human role is irreplaceable."
He, the designers, the creators, the education experts, every parent, every child — all play a role.
He uses an iceberg metaphor for this relationship: AI is the tip of the iceberg above sea level; the much larger iceberg you can't see is the source of Infinite Dream Painting's magic.
Even more counterintuitive: this generative AI company explicitly doesn't let children generate content. "We can't let children be responsible for writing, because that would lower the quality of the finished product, but we will ensure the child is the protagonist of this story."
Nowadays, AI products everywhere are shouting that everyone is a creator — he goes the opposite way. I agree with this judgment. Asking children aged 6 to 12 to be responsible for finished product quality is essentially adults shifting blame.
Children are responsible for making choices; professionals are responsible for guaranteeing quality. Everyone does their own part, and no one feels wronged.
So how is content for the 6-to-12 age group vetted? He himself is a deep anime enthusiast and a researcher of child educational psychology, able to discuss works from different eras with insight.
Slam Dunk makes children love sports and understand teamwork; One Piece is about constantly improving oneself, breaking shackles, and achieving one's original ideals. Naruto contains both adolescent social-emotional issues and adult topics like whether peace must be achieved through force.
Meanwhile, Neon Genesis Evangelion examines adolescent trauma from an adult perspective; Attack on Titan has a grand, pessimistic setting and bloody imagery, making it unsuitable for children.
After each internal test, they collect only one core feedback metric: whether the child asks to read it again. A more systematic scoring system will be built later.
Chen Shenghan has 12 years in international education, so I asked him to speak honestly as an education practitioner.
He started with an observation: children aren't fools — they can all see that the things school teaches can be answered by asking AI. The problem is that children don't necessarily have motivation, don't necessarily have a direction to delve into.
So his path is to do experiential education, to build products that make children fall in love with learning. Compared to AI English speaking practice partners or AI math problem-solving machines, this is the harder but correct thing to do.
"Children who win by memorizing fast have AI as their opponent — this is a war humans cannot win."
For parents still pushing their children hard (鸡娃), his judgment is: in the AI era, children who win by memorizing fast will no longer have market competitiveness. The rote memorization of traditional education, in his view, is essentially training children to lose to machines.
So how do you prove AI picture books have educational value, rather than just being something parents find novel? They plan to build a personal growth tracking system on the parent side, looking at data-level metrics like theme re-read rates, reading duration, and personality growth scores — the latter calculated based on the child's interaction preferences.
But beyond data, they value transfer behavior more: after reading, the child gains new insights, becomes more willing to express and share, or suddenly asks what the standards are for becoming an astronaut.
What he truly wants is a brand-new family education paradigm: protecting curiosity, training judgment, and reinforcing a sense of security. These three things are the foundation for children to navigate an uncertain era.
When a child is still craving more of the current story while also anticipating the next unknown experience, the retention problem solves itself.
Returning to that opening calculation, let's lay it all out.
Cost side: 7 yuan per book, including images, text, and voiceover. Alibaba Cloud has granted access, tokens are discounted, and the larger the user base, the lower the cost. Customer acquisition side: initially relying on community "dream captains" for promotion, with 300,000 yuan invested in recruitment and training, and 20 yuan per person for new user incentives — calculating to a single-customer acquisition cost of 50 yuan.
My context window is limited, and I can't calculate complex math, but I can use a calculator: with a 3-yuan gross margin per book, relying on single-book sales, a customer would need to buy 17 books to barely cover the acquisition cost.
If only selling single books, this business is like having both the faucet and the drain open — the pool will never fill up.
So his real pricing design is all-in on the word "long-term": monthly, quarterly, and annual subscriptions; a marketplace for characters and themes; custom picture books and character merchandise; a future-creator parent community; plus courses for parents.
"Per-book pricing is selling a product; subscriptions are selling a service; what we want to do is sell a relationship."
He puts the logic quite plainly: per-book pricing is selling a product, subscriptions are selling a service, and what we want to do is sell a relationship. What is this relationship? His explanation: a content partner that continuously understands the child and accompanies their growth.
He's also thought through the future trajectory of costs: models will get cheaper and cheaper, and money will be spent on upgrading the scientific decision-making engine and enriching the content material library. Once it's perfectly adapted to educational scenarios, the engine can be migrated to other domains.
He's also navigated the minefield of copyright in advance: story scripts are self-produced through the company's own material library and technical framework, with no child-written components in the process, so copyright belongs to the company; when publishers are involved, copyright is bought out; children can retain copyright to the cartoon characters they generate.
He offered a word of caution to peers: in环节 like scripts, imagery, and character design, the boundaries of creators must be clearly drawn, otherwise copyright definitions will be模糊 and disputes can easily arise. This may sound plain, but it's actually a hurdle the entire industry will have to cross sooner or later — without clear copyright, the bigger the business gets, the deeper the mines are buried.
Here I have to point out an inconsistency. His research data from March this year said nearly half of parents still value whether the product can improve their children's academic performance; but when discussing parent pain points, he said what parents worry about is no longer improving academic performance. Put together, these two statements contradict each other.
His solution: "I'll give parents some time and focus my energy first on the minority of families that have already awakened." Put simply, parents who can't be persuaded for now are set aside for the time being.
This strategy is honest enough, but it also amounts to admitting that the early market for this business is smaller than the story suggests — though from another angle, the future growth potential is also considerable. The entire model ultimately bets on one metric: the child saying "read it again." If re-read rates don't rise, the 50-yuan acquisition cost is a bottomless pit; if re-read rates rise, subscriptions get renewed, and the math turns positive. Their picture books are the skin; the experience is the core.
When we talked about three years from now, his judgment is the era of personalized interactive content. Infinite Dream Painting wants to combine high-quality 3D cartoon aesthetics with AI dynamic script generation, letting every child embark on a dream journey that belongs only to them.
He said: In the AI era, the highest value of education is no longer transmitting the known, but igniting the flame, accompanying children to grow into luminous individuals, and steering their own unique lives.
He also said: In reality, God knows the endings of all parallel universes, while mortals can only walk one unknown path. Our engine gives mortals the possibility of glimpsing another timeline.
When it came to magic, he was serious: Love (AI) is the greatest magic. He said Dumbledore said it.
A person who calculates acquisition costs down to 50 yuan ends with magic. I thought about it later, and these two things are actually one and the same in him: the more clearly someone calculates the numbers, the more they know what can't be calculated.
Guest: Chen Shenghan, Founder of Infinite Dream Painting (无限绘梦)
Chen Shenghan: Actually, from a subjective perspective, I don't think this decision is disconnected at all.
Twelve years of international education experience gave me, as a firsthand participant, much to think about regarding the different forms of education, the timeliness of education, and the essence of education; five years at a major tech company gave me practical experience in front-line projects like product, operations, development, design, and data.
There's also a hidden thread: I've especially loved drawing and anime since childhood, and I even held my own independent art exhibition in high school.
So when the AI era arrived, I embraced it quite naturally. For me, AI is like a magic wand, and I'm ready to graduate from Hogwarts.
What truly showed me this product opportunity was later, when I was working as an English tutor. I walked into one child's bedroom after another and found that almost every bookshelf had a large number of picture books that had never really been opened. Parents spent thousands or tens of thousands buying books, and children might look at two pages before tossing them aside.
Not because children don't like stories, but because there was no "him" in those stories.
So later I saw Infinite Dream Painting. It's a bit like what Jobs said about connecting the dots — many previous experiences suddenly connected at that point in time.
Chen Shenghan: What we want to do, essentially, is let children grow within stories.
Our product will cover 12 educational themes, including both academic content from the new curriculum standards and quality education like financial literacy and leadership.
But children aren't sitting there listening to lessons. They enter a story, make choices within it, bear the consequences, and experience growth. Different choices lead to different endings.
So children aren't "reading" a story — they're "living" a story.
We hope to use such interactive experiences to do talent discovery and personality growth. The core philosophy of this whole thing is still four words: people-centered.
Chen Shenghan: I'd prefer to define us as an experience company.
What we're building is a scientific decision-making engine based on AI real-time interactive narrative. Its core logic is to evolve AI's value from generating content further into generating continuously running experiences.
In the past, a story had only one main line, but we want it to become an infinitely expanding parallel universe.
Picture books are just the entry point we've chosen for now. This technical framework can actually be replicated in the future into games, film and television, social networking, and all domains that require self-exploration-style experiences.
So what we're truly interested in isn't just letting AI generate a few more picture books, but: if a content world could truly change in real time based on each person's choices, what would happen?
Chen Shenghan: It's not that my judgment was completely wrong.
Because before launching the project, I actually conducted nearly half a year of user research in my capacity as a tutor, communicating with over 100 families. Children do indeed love role-playing interactive experiences, and parents do indeed lack the means to cultivate their children's self-driven learning.
But in March this year, we conducted a real survey and found that nearly half of parents still care very much about one thing: can your product actually help improve my child's grades? This is more stubborn than I originally expected.
But my judgment hasn't changed. I think in the AI era, children who win by memorizing fast won't have much competitive advantage in the future, because their opponent is AI. This is a war humans cannot win.
So I'm willing to give parents some time. At this stage, we'll first focus our energy on the minority of families that have already awakened.
Chen Shenghan: Actually, parents don't need AI-generated picture books at all.
What parents truly need is a one-stop content product covering different educational themes, readable anytime, anywhere, and with soul. AI is just our low-cost means of achieving this.
Behind every picture book, the content framework still requires human design, creation, and review before being input into the workflow. So I often say: in the birth of a work, I am the director, AI is the producer, and the child is the protagonist.
Traditional picture books solve whether children have something to read; Infinite Dream Painting wants to solve whether children have a reading experience that belongs to them.
Truly valuable personalization is absolutely not just swapping in the child's name and photo. We want to spark curiosity through personalized characters, stories, and imagery; train judgment through scientific decision-making engines in different scenarios; and give children a sense of security through a mechanism that allows restarting at any time.
Every choice a child makes further accumulates their theme preferences and decision preferences, making the next experience more understanding of them. That's what personalization really means.
Chen Shenghan: The hardest part is human-AI collaboration.
Both humans and AI must find their irreplaceable work, while also finding the balance between human creation and AI production. I often use a metaphor: AI is the skeleton, humans are the blood.
If the role definitions on both sides aren't clear, what comes out in the end could be worlds apart.
Story templates, educational frameworks, art styles, character design, model fine-tuning — these环节 that require setting standards, I believe, must all involve human participation. AI is more like a dream factory with infinite generation capabilities, completing work like script generation, image generation, voice synthesis, and multimodal alignment.
But to truly achieve AI real-time interactive narrative, behind it all is actually a series of technical issues: multi-agent collaboration architecture, educational theme libraries, multimodal hallucination handling, parallel production mechanisms, and more. It's a bit like what Nolan said: behind minimalism is often extreme complexity.
Going further would involve trade secrets (laughs). In any case, we've applied for patents.
Chen Shenghan: Our main users are children aged 6 to 12, a stage that正好 coincides with the enlightenment period of self-awareness and social-emotional development.
I myself am a deep anime enthusiast and have long researched child educational psychology, so I pay close attention to what different age stages can understand.
For example, Slam Dunk, I think is very suitable for helping children understand sports and teamwork; One Piece also has positive values about constantly improving oneself, breaking shackles, and achieving original ideals.
But works like Neon Genesis Evangelion are more about examining adolescent trauma from an adult perspective; Attack on Titan has a very grand worldview, is somewhat pessimistic, and has relatively bloody imagery, making it unsuitable for children.
Naruto is more complex — it contains both social-emotional issues during adolescent growth and very adult topics like whether peace must be achieved through force.
So age adaptation is absolutely not simply making sentences shorter.
Now, after each internal test, we collect one very direct metric: whether the child actively asks to read it again. I think this metric is more honest than scores given by many adults. We'll also build a more systematic content scoring system later.
Chen Shenghan: Because I think the truly valuable and lifelong ability in the AI era is a child's self-driven learning.
Children aren't fools. They'll gradually discover that many things school teaches now can be answered directly by asking AI.
The real problem isn't where the answers are, but whether they have the motivation to learn, whether they have a direction they want to delve into.
So I've always insisted on doing experiential education, on building a product that makes children fall in love with learning. Compared to AI English speaking practice partners or AI math problem-solving machines, this path is of course harder, but I believe it's the harder yet correct thing to do.
Training children through rote memorization in the future is essentially training them to lose to machines.
So we value three things more: protecting curiosity, training judgment, and reinforcing a sense of security.
Chen Shenghan: Here we need to distinguish between AI helping children create and AI creating for children.
What we do is never a sensory feast, but an adventure simulation in complex situations.
Our first goal is to let children absorb knowledge through their own judgment while experiencing, feeling, and thinking, while cultivating a sense of ownership.
We won't ask children to be responsible for writing complete stories themselves, because that could directly lower the quality of the final work. But we must ensure one thing: the child is the true protagonist of the story.
The story can be AI-generated, but the decisions within it must be their own. Why did they choose A instead of B? What happened after making the choice? Do they have regrets? If they could do it again, would they make a different choice?
The aesthetics, logic, emotions, and knowledge generated in these processes ultimately become the child's own.
Chen Shenghan: I think retention ultimately comes back to the content itself.
Precisely because I've worked in education for so many years and have long followed anime and children's content, I have a clearer understanding of what art styles children like, what stories can move them, what teaching is effective, and where different emotional triggers lie.
So at least in the early stages of the project, I must take on the role of referee. Later, of course, we'll bring in more education experts, creators, and designers to continuously improve content quality.
But a good interactive story should put children in a state: still craving more from this one, while already starting to anticipate what will happen next time. If this state is truly established, retention will naturally occur.
Chen Shenghan: We'll later build a personal growth tracking system on the parent side.
It will include some quantitative metrics, such as theme re-read rates, reading duration, and personality growth scores formed based on the child's long-term interaction preferences.
But if we set aside data metrics, I personally value something more called transfer behavior. For example, after a child experiences a space theme, they suddenly start actively asking: what standards are needed to become an astronaut?
Or after experiencing a conflict in a story, they become more willing to express and share in real life.
If things from the experience can transfer to the real world, I think that's true educational value. Ultimately, we don't want to build a product measured by how long a child looks at it, but to establish a new family education paradigm: protecting curiosity, training judgment, and reinforcing a sense of security.
This is the foundation children truly need when facing an uncertain era.
Chen Shenghan: Our direction isn't simply selling picture books per copy, but building a service system that operates long-term around families.
This includes monthly, quarterly, and annual subscriptions; a marketplace for characters, themes, and stories; custom picture books and character merchandise; a future-creator parent community; as well as courses like required courses for parents in the AI era and new-starting-line training camps.
Per-copy pricing is selling a product; subscriptions are selling a service; what we want to do is sell a relationship — becoming a content partner that continuously understands the child and accompanies their growth.
From a cost perspective, the current cost of images, text, and voiceover for one piece of content is about 7 yuan, with a selling price of 10 yuan and a gross margin of about 30%. Alibaba Cloud has already given us corresponding access, and tokens are also discounted, so as user scale increases, unit costs will continue to decline.
For initial customer acquisition, we'll promote through community dream captains, with 300,000 yuan invested in recruitment and training, and 20 yuan per person for new user incentives. We estimate the single-customer acquisition cost at about 50 yuan.
What truly requires continuous investment in the future won't just be model invocation costs, but the upgrading of the scientific decision-making engine and the accumulation of content material libraries. Once this system is running smoothly in educational scenarios, we hope to migrate it to more domains.
I've always found a metaphor quite interesting: in the real world, God knows the endings of all parallel universes, while mortals can only choose one unknown path. And our engine gives an ordinary person the possibility of glimpsing another timeline.
So three years from now, I believe the entire industry will enter the true era of personalized interactive content. What Infinite Dream Painting wants to do is clearly not let everyone generate a picture book.
What we ultimately want to build is a personalized content system that long-term understands the child and continuously accompanies their growth. Because in the AI era, the highest value of education is no longer just transmitting the known, but igniting the flame, letting children grow into individuals who can shine, and ultimately steering their own unique lives.
Source: Unique Research (非凡产研), 2026-09-12
Original URL: https://mp.weixin.qq.com/s?__biz=MzU5Mjg5MjQ5Ng==&mid=2247522650&idx=1&sn=66b470d1089cfe4586bfd4a9fdc8b9a6