147: OiiOii 闹闹: Why “Douyin First, CapCut Second” No Longer Holds

Summary

• 闹闹’s core correction to AI content products is that the “Douyin first, CapCut second” sequence can no longer be copied; today, the right move is to use creative Agents to build supply first, then build the community. Early Douyin users could tolerate content that was repetitive and monotonous but entertaining; consumers today are accustomed to abundant content, so the monotony and novelty deficit of pure UGC can no longer support retention. OiiOii is therefore first packaging professional creative capabilities into an “end-state first” tool, then using consumable content to support the future community.
• OiiOii is targeting PUGC supply suppressed by high production barriers—not generic UGC cultivated from scratch or a handful of leading animation-drama studios. 闹闹 estimates that China has roughly 1.8M–2M ACG accounts, typically run by teams of 2 or 3 people posting once a week; the product aims to push output to “10 posts a day.” Even reaching 2 posts a day would be transformative for animation. 闹闹 acknowledges that he is not certain consumption will rise, but believes that once supply is genuinely released, consumption could expand dramatically.
• The value of multiple Agents is not just automation; it lets ordinary creators act as “directors” who control character design, art, storyboarding and other stages. 闹闹 believes end-to-end automatic production “is not there yet” and would weaken creators’ sense of ownership, so OiiOii exposes both the workflow and the editable canvas. The cost is an exponential combination of task assignment, handoffs, arbitrary entry points and two-way synchronization, making the technology far more complex than an encapsulated workflow.
• Each model has its own boundary, and the product’s accumulated edge lies in understanding content and engineering model orchestration for different shots. MJ is better at preserving style but weaker at preserving characters; 4o and Nano are better at character consistency and editing accuracy. On video, a Task Agent once selected intelligently among roughly 6 models, but Sora 2’s advantages in immediacy, naturalism and audio-video synchronization have led the team to create a separate primary path for now. “Only when none of the models is perfect do we have the possibility of combining them.”
• The team deliberately avoided the AI animation-drama battlefield centered on scripts, traffic buying and bespoke delivery, because it would pull the company toward heavy To B work. 闹闹 acknowledges that the team is neither good at writing hit scripts nor capable of buying traffic; leading studios also have radically different workflows, making it easy for a tool provider to become a custom-adaptation shop. After the beta, animation-drama companies began approaching the team on their own, with one major need being to efficiently turn “scripts into large volumes of storyboards,” which studios could then download and edit themselves.
• Monetization and community-building are sequenced: first improve music and story experiences, then expand into comedy, MV, science education and other verticals while serving small B customers such as game and toy companies, and try the community in roughly another 6 months. When the community launches, hits will drive acquisition while the rich supply accumulated through the tool will drive retention; otherwise, “it’s just a peak that disappears, and everyone leaves in one wave.” At the time of the interview, OiiOii had roughly 150 user groups across WeChat and Discord, including about 124 test groups and more than 20 partnership groups. Invite codes had shifted from official distribution to 5 per existing user, conditional on the user creating real work and staying active for 2 or 3 days.
• 闹闹’s founder-market fit comes from the overlap of content, technology and organizational experience—not simply from an interest in animation. From 2014 to 2019, he spent roughly 6 years building an extreme-sports content business and experienced an “ivory tower” collapse and rebound; he later helped take CapCut from roughly 5M–6M DAU to 36M DAU at ByteDance, led multiple breakout Douyin effects, and then validated the supply constraints in animation at Bilibili. “Many things are determined by mental strength,” which is why he describes the setbacks of his first startup as “antibodies.”

Deep dive

1. “Douyin First, CapCut Second” Only Works Halfway in the AI Era

• 闹闹 initially built the anime community “离谱” with obvious path dependence: Douyin built the community first, then CapCut captured the creative activity, as if content products naturally had to move from UGC to PGC. At the time, multimodal models were only good enough for users to “play around,” while communities could tolerate unstable output better than productivity tools.
• His reflection after building 离谱 was that the variable that no longer held was timing. In Douyin’s early days, consumable content was scarce, so users accepted repetition and monotony; today, users are accustomed to abundant content and will return to mature platforms after briefly playing around in a new community. “Its monotony becomes unbearable.”
• Model capability has also changed the basis of judgment: the market is no longer discussing only “how good is this model?” but also the content inside the generated output. AI output is gradually becoming consumable, turning creative tools from technology-demonstration toys into potentially viable productivity products.
• OiiOii therefore reverses the sequence: produce sufficiently rich, consumable content first, then build the community. 闹闹 still wants to build a content platform; only the path has changed. “Building the tool itself is also part of the process at this stage.”

2. An Agent’s Irreplaceable Value Comes from Working Backward from the End State

• 闹闹 breaks CapCut into two parts. Traditional editing provides general-purpose operations and can therefore be replaced; templates, by contrast, reverse-engineer the production process from the beat-synced music content that ultimately becomes popular on Douyin. “Only by using the templates inside can ordinary users make the corresponding finished videos at low cost.”
• He believes many AI tools are still operating at the GUI-engineering layer, while a vertical Agent is more like a template ecosystem: first determine what a given type of person wants to make, then hide the extensive tuning between the model and the finished product, packaging a professional creator’s capabilities into “a prompt template.”
• AI templates are richer than old templates. Even with exactly the same prompt, the output still varies; user inputs also change how templates are combined. The tool can therefore preserve the structure of the target content without producing a batch of completely identical works.

3. The Tool Is Building the Future Community’s Content Shelf in Advance

• If the future community needs anime MVs, beat-synced music, mashups or other vertical formats, OiiOii can first embed those modes into its Agents and let creators produce them directly. 闹闹 puts it plainly: “Whatever type of content I want you to create, you can create with this tool.”
• This path lets the platform decide in advance how many content categories it wants to cover, then fill them one by one. When the community launches, it will not have to wait for a few good pieces to emerge by chance from a sea of monotonous UGC; the content will have stronger consumption value from day one.
• 祝颖丽 therefore concluded that OiiOii is not moving from a creator tool toward a community; it is using the tool as the intermediate route for building the community. 闹闹 agreed that Agents make “tool first, community later” more workable than mechanically copying Douyin’s path.

4. Animation’s Core Contradiction: Consumption Has Expanded, but Production Capacity Has Not Been Democratized

• 闹闹’s starting point is a supply-demand imbalance. Animation audiences have expanded from children to adults, while animated films, Chinese animation and overseas content have all broadened their reach. Supply, however, remains highly concentrated among leading players, and tools have become increasingly specialized instead of releasing more creative capacity in line with demand.
• Live-action production moved from PR, DaVinci and Final Cut to CapCut as barriers fell, and smartphones democratized creation. Animation remained constrained by drawing and 3D modeling, continuing to evolve toward greater specialization. “AI is effectively breaking into this field all at once.”
• On Bilibili, creators continue enthusiastically learning and producing with “extremely difficult-to-use” Live2D. To 闹闹, this proves that the supply is not absent but “held back,” waiting only for a sufficiently low-barrier release mechanism.

5. Expanding Supply Can Create Consumption That Was Previously Hard to Imagine

• 闹闹 admits he is not certain consumption will rise, but his directional judgment is that once supply is genuinely released, consumption could naturally expand.
• His analogy is that before Douyin, many people would not have believed ordinary people’s daily lives were worth consuming. Once democratized production tools appeared, users became engrossed in watching the lives of strangers. Animation may likewise need sufficient supply before demand becomes visible.
• Animation also works for history, finance, relationships, the workplace, education and science communication. These subjects often lack real-world scenes that can be filmed and can only be narrated; if animation becomes cheaper than live-action production, verticals outside ACG could become new markets.

6. OiiOii Found Its PUGC Entry Point in Two and a Half Months

• The team began pushing forward around August and formally launched its beta on November 10, amounting to roughly two and a half months after excluding the National Day holiday. It knew from the start that it would continue working on animation, but the specific form was not fully clear on day one.
• The team first considered an overseas version of 离谱, then concluded that UGC was still constrained by user scale, paid acquisition, inference costs and community retention. It also researched AI animation dramas but ultimately did not make them the initial market.
• 闹闹 divides content production into UGC, PUGC, PGC and OGV, with animation dramas sitting roughly between PGC and OGV. After ruling out both ends, the team identified a major gap in the middle: PUGC creators can run media accounts, are motivated to produce video and typically have some willingness to pay.

7. 1.8M–2M ACG Accounts Form the First Quantifiable Market Layer

• 闹闹 estimates that China has roughly 1.8M–2M ACG accounts. A typical account is operated by only 2 or 3 people and posts about once a week; the efficiency of the existing workflow has not been fully improved.
• OiiOii’s internal target is to help these accounts reach “10 posts a day.” He also deliberately lowers the assumption: even 2 posts a day, compared with weekly posting, would be “extremely transformative” for the animation industry.
• This group offers clear initial demand while also providing a path into non-ACG verticals. 闹闹 therefore defines it as an entry point with ecosystem potential, rather than merely a way to remove a few steps for existing animation studios.
• PUGC balances quality and scale. Creators already have the desire to express themselves and the judgment to evaluate content, so the platform does not need to educate them from scratch; the tool lowers the cost of professional processes that once required a team, making higher-frequency, richer supply possible.

8. Multiple Agents Are Not a Universal Answer for Video Generation

• 闹闹 emphasizes that not every video needs multiple Agents. Animation and film have been industrialized over a long period and naturally involve sequential handoffs among screenwriters, character designers, artists, storyboard artists and other specialists. Even higher-quality UP creator content inherits this way of thinking.
• Short, fast-turnaround ads may only require stacking assets and may not need characters, complex storyboards or narrative. Whether to preserve a multi-specialist structure depends on the target video and whether users need a smooth transition from traditional creative logic—not on the number of Agents available to showcase.
• A stepwise structure also requires edits to propagate through the workflow. Users can revise during character design and then hand the work to the storyboard artist, or start from different entry points such as a character or an illustration. The workflow must therefore preserve stable handoffs while supporting different creative sequences.

9. A Visible Workflow Turns the User into a Director, Not Someone Who Presses a Button Once

• 祝颖丽 uses MovieFlow as a comparison: it encapsulates the director, script and storyboard Agents in the background, allowing users to enter a prompt and receive a finished video. OiiOii instead exposes workflow progress and lets users edit every node.
• 闹闹’s first judgment is that AI is “not there yet” when it comes to producing a good video completely automatically from beginning to end; every critical node still requires human participation. He therefore did not fully encapsulate work that already required user involvement.
• His second judgment concerns creative psychology: creators need to “enjoy a bit of control over the creative process.” OiiOii makes the user the director. The actual work is done by a team of Agents, but because the user triggers, controls and participates in the work, the result can feel like “mine.”
• Full automation would make participation and ownership “extremely weak.” The purpose of multiple Agents is therefore not to reproduce a company org chart, but to preserve the creator’s participation in and control over the work.

10. Open Control Turns the Technical Challenge into an Exponential Combination Problem

• Each Agent may have multiple capabilities: a storyboard artist may create images and video, while the art director and character designer may also generate images. Once a user submits a task, the system must understand who should receive it, who should assign it and whether the request concerns the character, the art or the shot.
• A fixed workflow must also hand work off reliably, like “the seven dwarfs,” without allowing users to skip steps arbitrarily. Yet users may start with a character rather than a story, create only an illustration or revise a completed character before sending it to storyboarding; the order is not fixed.
• The real difficulty is knowing when the system should exit the workflow, when it should re-enter and which flow it should take after re-entry. Stability and flexibility constrain each other. 闹闹 says the team is “solving combination problems every day,” and many existing bugs are concentrated here.
• The canvas and chat module must also synchronize in both directions. Edits in the visual area must update relevant upstream and downstream states, while progress in the chat module must be reflected on the canvas. One-way generation has become a system linking multiple roles and interfaces.

11. The Team Rejected an AI Animation-Drama Starting Point Built Around “Making a Hit”

• Animation dramas have a direct revenue path, but the first barrier is the script. 闹闹 admits the team has almost no human scriptwriting capability. A model may be able to generate scripts, but that is not the team’s core strength; spending time acquiring those resources at the outset would mean entering the battlefield where it had the least confidence.
• The second barrier is paid traffic. Like short dramas, animation dramas are “a business that relies entirely on buying traffic to make money.” If the goal is a hit, the content must contain a hook from the first step. The team would need strong scriptwriting and traffic-buying capabilities, and OiiOii had neither at the time.
• The third barrier is that customers often do not actually need a single Agent. They need point solutions, such as faster image production within a mature human workflow. Human dependence remains strong, which conflicts with the team’s goal of reducing it as much as possible.
• 闹闹 is not ruling out entering the market in the future; he insists that a startup must begin from the capabilities it knows best. “Coming into contact from the beginning with things we are not good at and do not like most” would create too large an obstacle for the team.

12. Leading Animation-Drama Customers Would Turn the Product into Heavy To B

• If the company serves leading studios and pursues showcases, the potential customer base is limited to a few companies, each with a different workflow. The tool would have to adapt to each one, becoming “very much like a small To B company.” That fits neither the ecosystem goal nor 闹闹’s strengths.
• If the company moves down to mid- and lower-tier studios, the difference between them and PUGC is already small. OiiOii therefore chose to build general platform capabilities first rather than design around the bespoke workflows of leading customers.
• After the beta, many animation-drama companies still approached the team. They did not necessarily want complete synthesis; they wanted to put in a script, quickly receive a large volume of storyboards, download the assets and edit them themselves. OiiOii unexpectedly met strong demand for the “script to storyboard” segment.
• If revenue signals become strong enough in the future, the company could form a separate small team to serve mid-sized or relatively large To B customers, but it will not shift its main forces immediately.

13. The Animation Ecosystem and the Animation-Drama Business Differ at the Level of First Principles

• 闹闹 does not deny that making money is legitimate, but believes the original intent determines every subsequent action. Animation dramas and short dramas typically aim for short-term profitability and hits; his own starting point is that he “really wants to make animation”—to express imagination through content and help users find what they genuinely want to make.
• OiiOii therefore wants to cover a range of formats: absurdist animation, original stories, imaginative remixes, time travel, warm emotional expression and “sudden reversals” lasting only a few dozen seconds. The latter may generate strong traffic without necessarily constituting a “drama.”
• The platform is pursuing richness of style and expression rather than steering creators toward a few formulas for rapidly triggering and releasing emotion. “We want to build an ecosystem, not just a few hits.” Whether something goes viral will depend more on the user’s own content ability.

14. The Initial Aesthetic Covers Both Familiar IP and Unfamiliar Illustration Languages

• The first content direction is familiar IP styles, including Lego, The Simpsons, Family Guy, American comics and several major Chinese IP styles. Familiar styles make it easier for users to create remixes.
• The second direction is highly watchable illustration styles from individual artists. They look good as static images but are rarely turned into animation; limited video data also makes them difficult for models to learn. As a result, “the image can be generated beautifully, but the animation may not be.”
• The team still wants to cover a wide range of styles, including cute small-scale 3D, pixel art, hand-drawing, ink wash and watercolor. 闹闹 breaks content richness into 3 dimensions: style, subject and duration. People who genuinely like animation can continue experimenting across all 3.

15. Model Uncertainty Is Treated as “Gacha,” and Surprise Can Be a Product Feature

• 闹闹 says current models are “to some extent like a gacha game.” When the team discovers a strong result, it studies the conditions under which it appeared and reinforces them; if a style consistently fails to produce good audiovisual results, it is temporarily removed from the product.
• A result that diverges from expectations is not necessarily a failure. It may fail to tell a complete story but still express emotion or form an abstract visual language. As long as the image contains a surprise, the team may keep the style rather than force every output to obey real-world physics.
• Animation audiences are inherently tolerant of departures from common sense and physical laws. A model-generated image that “moves differently from how reality would suggest” can become a uniquely animated form of expression. The key question is whether the audiovisual result works, not merely whether it reproduces reality accurately.

16. Image Models Face a Hard Trade-Off Between Preserving Style and Preserving Characters

• 闹闹 believes that “there is no perfect model,” and that this is precisely the opportunity for the tool layer. MJ excels at animation and hand-drawn styles, and its SREF capability is strong at preserving style; its CREF capability is weaker at preserving characters, making it difficult to keep the key elements of the same person stable across images.
• GPT-4o and Nano Banana are natural-language input models that emphasize editing accuracy and consistency. They are stronger at preserving characters but less capable of retaining styles that require imaginative divergence. “The stronger the consistency, the weaker the imagination of the style”—the two are naturally in trade-off.
• If a user uploads 2 reference images, one specifying the character and the other the style, the system may favor MJ because the style requirement is stronger. If the user uploads only 1 character image, it is more likely to use Nano Banana or GPT-4o, prioritizing character preservation while inheriting part of the style.
• On the image side, the system currently does not let an Agent choose the model autonomously. The team understands each model and writes explicit engineering branches for known cases. Character consistency is close to deterministic; an Agent can make the decision, but “it will definitely make mistakes,” so stable scenarios should not be handed to probabilistic decisions.

17. Video Model Selection Was Once Handled by a Task Agent, Then Sora 2 Got Its Own Primary Path

• The first product version integrated roughly 6 video models, with a Task Agent understanding each storyboard and selecting automatically. A fight or conflict between 2 people might go to Hailuo 2, while a delicate crying scene might go to Kling or Jimeng. Video performance is “likely to be this way,” making it suitable for intelligent judgment.
• The automatic routing was later temporarily shut down because “Sora 2 is simply too good.” Its generation principles also differ from those of the other models, so the team had to create a separate path for focused adaptation. 闹闹 describes this as a shift in resource priorities, not as a perfect replacement.
• Sora 2 is not especially strong at character consistency, but it performs very well on immediacy, naturalism and audio-video synchronization; older models may be better on some consistency tasks. Once resources allow, the team plans to restart the older models and make them work better alongside Sora 2.
• 闹闹 divides video models into 2 camps. One uses reference images plus text and is text-heavy, extracting only key elements from the image; the other uses first and last frames and is image-heavy, stably extrapolating from the full image. Sora 2 belongs to the first camp, and the 2 methods are difficult to combine directly at the current stage.

18. The First Moat Is a Scarce Human Operating Model

• Before orchestrating models, the team must understand which elements make up the content, break the intuitive notion of “good” into testable logic and standards, then test and combine every step to find the best solution under the conditions of the moment.
• 闹闹 believes people who understand aesthetics, technology and content are “as rare as phoenix feathers and unicorn horns.” Designers are usually strong in intuition and aesthetics, while technical staff are usually strong in rationality and structure; multimodal work requires “both halves of the brain to be highly developed.”
• He developed a method for finding such people through Douyin’s effects business. The current product team has roughly 3 people, with about 1.5 focused on model orchestration; there are roughly 4 or 5 designers, 3 of whom work on understanding models and controlling output quality, in roles similar to technical artists in the game industry.
• Designers conduct much of the output testing, while product handles the logic of model combinations. The founder goes deep into product details and architecture and makes the final call on combinations others cannot resolve. “Which combination is optimal” still requires 闹闹 himself to take responsibility for both the principles and the results.

19. The First Startup Showed That an “Enthusiast Market” May Just Be an Echo Chamber

• From 2014 to 2019, 闹闹 ran an extreme-sports content company, filming stories about Chinese climbing and downhill-skateboarding creators while also taking on automotive and 3C advertising. The business resembled what would later be called an MCN, though at the time it was simply a self-media and production team.
• The starting point was GoPro’s early entry into China, before it even had Chinese-language translation. The team first built a forum and accumulated users, then discovered that they clustered around skiing, paddleboarding and other circles. 闹闹 was also inspired by the Banff Mountain Film Festival and thought, “There are many people like this around China too. I want to film them.”
• The company ran early paddleboard events, was already working on ultimate frisbee in 2014 and 2015, and partnered with offline venues such as karting tracks to drive traffic. The response from enthusiasts led the team to believe the market was no longer small, only for it to realize later: “We thought it was no longer niche, but it was still niche.”

20. The App, Offline Events and E-Commerce Exposed the Wrong Scaling Paths

• The team initially wanted to build a community app, but smartphone cameras were still weak. Creators had to shoot with GoPro, edit on a computer, then transfer the video to a phone for upload; the short-video culture had not yet formed, making the production path fundamentally unhealthy.
• Offline events and venue activities were popular but labor-intensive and difficult to replicate. E-commerce also failed because equipment knowledge differed deeply across sports, while content shifted with the seasons; the team could not go deep enough in every category at the same time.
• The company eventually focused on content production and was able to support itself normally, but there was no reason for it to grow. 闹闹 chose to leave because his goal had shifted toward “serving creators,” rather than continuing to operate a sports-media business that could survive but could not scale.
• The experience also confirmed that he was merely “using sports to make content.” Extreme sports were his entry point into the content industry; what he was genuinely good at and interested in over the long term was the relationship among content, tools and creators.

21. The Cost of the Animation Industry Forced His Animation Dream Back into His Mind for 10 Years

• Before that, 闹闹 had worked as a product manager at WeChat for roughly 3 years, then spent 6 months learning animation. He had drawn since childhood, came from a family connected to music and was sensitive to audiovisual language. He had always wanted to make animation that was “pure and focused.”
• Animation production had a very high barrier to entry. He recalls that his salary already comfortably exceeded RMB10,000 and may have approached RMB20,000, yet an animation-company internship still required him to pay. Even a modeling director with 5 years of experience seemed to earn only RMB7,000–8,000, in a poor work environment.
• He admits that he “could not sacrifice to that extent for what I loved.” He still had material needs, so he put animation aside and began learning filming, camera settings and editing through extreme-sports content. Because he “knew nothing,” he still went ahead and did it.

22. The Most Valuable Lesson from the First Startup Was Not Zero to One, but Rebounding from Negative One

• The company grew well from 2014 to 2016, but while outsiders were full of praise, 闹闹 already sensed that the underlying capabilities in e-commerce and offline events had not kept up. It felt like an “ivory tower,” and the business subsequently declined, bringing anxiety with it.
• He observed that many entrepreneurs are good at repeatedly going from zero to one but leave when the curve falls. His good fortune was staying with the company long enough to find “the strength to land and rebound.” His analogy is that when a pathogen invades, the body develops antibodies; from then on, when problems arise, he does not first scare himself.
• The company eventually recovered from “negative one” to stable upward growth, and only then did he choose to leave. 闹闹 believes the experience of rebuilding order after hitting bottom was “far more valuable than zero to one,” and became a stable foundation for his later work at a large company and his second startup.

23. Meditation Turned Startup Anxiety from Fear of Outcomes into Concrete Action

• At the end of 2016, 闹闹 began to feel “a very calm but very expansive force.” He then encountered Buddhism, began meditating and found a teacher. He describes it as discovering a new continent that “completely changed the person.”
• His secular explanation is that people become anxious because they cannot accept impermanence. When they overprotect the image of “me,” praise lifts them up while criticism breaks them down. Once he saw that this self was only a shell, neither praise nor abuse directly threatened his real sense of security.
• As fear of death diminished, a large part of his startup anxiety also disappeared. He could judge objectively whether a bad outcome was acceptable and whether it could be improved, instead of repeatedly magnifying imagined disasters. He could then focus on small actions: “Do it step by step, and things will slowly get better.”
• 祝颖丽 summarized this as why investors value experience with setbacks. 闹闹 agreed that “antibodies” keep entrepreneurs from being frightened by themselves first. His conclusion: “Many things in this world are determined by mental strength. It sounds mystical, but it really is that way.”

24. ByteDance Turned His Content Intuition into Organizational and Product Capability

• After leaving his first startup, 闹闹 concluded that mobile creative tools would become a major opportunity for large content platforms and joined ByteDance. CapCut already had roughly 5M–6M DAU when he arrived; by the time it reached 36M DAU, he believed his main contribution was in organization, hiring and business logic rather than any individual feature.
• He considers his biggest contribution to CapCut not a particular function but “screening the people best suited to build this product for the organization.” Another contribution was the business logic linking CapCut templates, Douyin effects, submissions and social growth, which played a major role in the Douyin ecosystem.
• On the effects side, the team created major hits including real-time transformations into cartoons, old people, children, different genders and Pixar styles. 闹闹 calls the real-time cartoon transformation the world’s first. The team also established a breakout mechanism spanning new-technology research, several months of evaluation, launch and amplification.
• This business required imagining new technology as a product users would actually want to play with. 闹闹 says he likes physics and technology and can sense early what a technology is suited to become as an effect. “This business was a very good fit for me.”

25. A Strong Organization Once Created Flow; Organizational Fragmentation Eventually Ended It

• 闹闹 remembers the walls between ByteDance’s early departments as relatively low. Breakout effects required support from Douyin operations and marketing; even when people were in different departments, shared objectives quickly aligned them. “Everyone felt like they were winning a battle.”
• Platform infrastructure also allowed the team to scientifically calculate imitation behavior: users first saw who was using an effect and at what probability, then decided whether to follow. Product could intervene directly in traffic strategy and algorithms, rather than only owning production-side features.
• As Douyin grew, “creators” split into merchants, livestream users, advertisers, local-services merchants and other identities, with related functions and people distributed across businesses. 闹闹 was left on the product side and could no longer coordinate 2 large organizations to reunify creator services.
• The new work involved e-commerce individuals, e-commerce merchants and their respective growth logic. It was a product-architecture problem, not simply a content problem. He believed he could achieve 80 points but not 100, and did not want to rely indefinitely on responsibility rather than “flow and passion,” so he eventually left.

26. Bilibili Confirmed the Animation Capacity Gap but Could Not Support the AI Investment

• 闹闹 spent less than 1 year at Bilibili. Its core value was giving him systematic exposure to UP creators, studios and China’s animation-production capacity, filling in his market understanding of the animation industry he had wanted to enter for years.
• The experience further confirmed that animation supply was constrained by production difficulty, not by a lack of creator desire. He saw the enormous potential of AI animation and became more convinced that lowering the barrier would release both existing creators and new participants.
• He discussed AI animation internally at Bilibili, but the technology and funding resources at the time could not support it. 闹闹 still considers Bilibili’s creators and consumers high quality and says it was “quite a pity” that he could not advance the project there.

27. The Near-Term Roadmap: Fill the Content Shelf, Then Test Small B, Overseas and Community

• User inflows have made the iteration cycle faster than expected. The team first divided content into music and story categories, planning to spend roughly 1 month improving current generation quality and experience, then 2 or 3 months filling out comedy, MV, science communication and other verticals, expanding category by category like operating content channels.
• The third line is serving small B customers, including game and toy companies. They are willing to license their characters and assets, let creators produce memes, remixes and distribution at scale, and purchase tools for ecosystem creators. The team is explicitly not prioritizing large-B customization for now.
• The overseas launch is planned for after Christmas, around mid-to-late January. The domestic path is relatively clear, while the team still needs to explore overseas target users and content differences. 闹闹 expects exploration to be slower, but believes content richness could be greater once the pattern is found.
• The move from tool to community is expected to take “more than half a year”—roughly the second half of next year as understood in the interview. The exact timing will depend on the community’s condition.

28. In Community Growth, Retention Comes Before a Single Wave of Hits

• 闹闹 divides future community content into 2 types: breakout content or UGC hits that drive acquisition, and the diverse work accumulated through the creative tool that “catches” users after they arrive. Growth can enter a positive feedback loop only if users find a large supply of content to consume after joining.
• If the platform only creates breakout content without follow-on consumption, “it is just a peak; once it passes, it is gone, and everyone leaves in one wave.” This is also the weakness of many tool products: they occasionally generate distribution but lack a content shelf that can turn acquisition into a lasting relationship.
• Retention is therefore more important than hits in the early stage. The immediate task is to expand the creator base and the diversity of supply. OiiOii’s current tool-layer value lies not only in one-off generation efficiency, but also in prebuilding content and a creator network for the community that has not yet launched.

29. Roughly 150 User Groups Let the Founder See “Living, Breathing People” Again

• 闹闹 currently spends substantial time reviewing and responding to user feedback, and recently also devoted some time to investors. He especially values organic word of mouth and constructive suggestions; a small amount of abuse or trolling has not become a focus.
• The team has roughly 150 groups in total. The exact number of WeChat groups is unclear and may be between 120 and 150; the website lists 124 test groups, along with more than 20 partnership groups, author groups and others.
• Unlike the vast abstract data he faced at ByteDance, early-stage entrepreneurship has brought him back into contact with “living, breathing people.” He describes the direct interaction as real and enjoyable, and says urgent user needs have repeatedly moved product priorities forward.
• The invitation mechanism moved from the first wave of official codes to 5 invitations per existing user. To prevent viral distribution from diluting user quality, existing users must actually create work and stay active for 2 or 3 days before receiving the 5 invitations.