Entrepreneurs
3 Curated Dialogues
闹闹
Key Views & Dialogues
78. Where Do Video Agents Go After Seedance? | A Conversation with OiiOii’s 闹闹 on the Secrets of Video Models: Data, Ecosystems, and Structuring the Subjective
Seedance validates multi-reference inputs plus language as creator-aligned, but the race is not over, with at least 3 players remaining. The moat is organizational: tens of thousands of annotators, evolving standards, and inherited creative-algorithm talent let incumbents reach 80-90 points versus startups’ 20-30. Rising costs push UGC toward Pro C and small B; OiiO’s routing and black-box layer improve certainty, but Agents have only a 70-80% chance of avoiding absorption.
View summary (8)
- 闹闹’s core product thesis is that multiple reference images plus natural-language description better match how creators think, but video models have not fully converged on that paradigm; Seedance is “basically an upgraded Sora 2.” After Sora 2 launched, she removed every first/last-frame model from her product and now believes that was the right call: all of her earlier prompt-engineering work carried over to Seedance, which supports true multi-reference inputs and natural shot transitions; chasing ever-longer videos “doesn’t make sense,” as long takes can be little more than a flex.
- “Video models are simply not a startup business.” The bottleneck is not just algorithms but the organizational machinery for data labeling: with tens of thousands of annotators and standards that must be set and iterated, incumbents can get to at least 80-90 points while startups manage only 20-30; that is why Seedance leads in China, with Kling next, and why short-video platforms inherited their edge from special-effects AI Lab and Kuaishou Y-tech creative-algorithm teams.
- China’s short-video ecosystem is the world’s leading breeding ground for models, and Sora’s missed opportunity was the ecosystem, not just the technology. “OpenAI had Sora 2 but failed to build and run the ecosystem around it; if it had launched an API first instead of that app, things might have worked out differently.” The video-model race is not over: she expects at least 3 players, and says incumbents have an advantage only when a project is not the first to be abandoned and continues to receive attention.
- The case for the Agent layer is straightforward: models deliver only 30-40 points of the finished capability, while an Agent can bring the application and product close to passing—and the layer itself is a black box. Even coding, as powerful as it is, is not something that can truly be delivered as-is; Midjourney’s survival against larger models shows that multimodal product layers need not be winner-take-all. The Agent’s job is to “understand how to get the best out of a model,” using routing and stage-by-stage benchmarks, with model-product, algorithm, and quality-design roles working together.
- Seedance has changed the cost curve, driving UGC away and pushing the market toward Pro C and small B, but she does not think foundation models can keep raising prices in a video ecosystem. “If raising prices means more people can no longer make money, it is unsustainable”; short-drama margins are already thin, and ByteDance cares more about the vitality of UGC creation on Douyin—cutting prices may actually maximize value. Consumer access cannot remain free against inference costs: “Even ByteDance-sized, Doubao has started charging,” because AI generation is an “unlimited cost” and may cost more than buying equipment to shoot video.
- Her verdict on the current video-Agent boom is blunt: “I think it’s a delusion.” Only 2 or 3 products may be genuinely solid; many are wrappers or routing stations cashing in on the Agent buzzword before disappearing. Every product markets itself as a 100, even though some are at 30-40 and others 70-80—“none is a 100, including us”—and users have no shortcut: they have to try them.
- The hard multimodal problem is “structuring the subjective,” not porting the language-model playbook directly into vision. Language is explicit and standardized, which is why coding got traction first; visual expression is implicit. Subjective quality “cannot be quantified, but it leaves traces, like a color band,” so benchmarks vary by person and product layers will necessarily “bloom,” rather than collapse into winner-take-all.
- Her restraint under investor questioning is notable: she assigns only a 70-80% probability that the Agent layer will avoid being absorbed by the model, and calls that objective rather than conservative. “People speak with far too much certainty; ‘the wolf is coming’ has been cried too many times.” Seedance 2.0 taught her less about technology than posture: “not flashy at all, not sci-fi at all—just honestly and solidly doing what needed to be done.” What AI startups lack most are people willing to sweat the details and do the dirty work.
2026, the Year of AI Video Is Coming | A Conversation with OiiOii Founder 闹闹: After WeChat and ByteDance, How Do You Capture the Next Wave of Opportunity?
No verified TL;DR or digest evidence was supplied for this edition. An investor-oriented description cannot be written without adding unsupported facts to the supplied record.
View summary (7)
- OiiOii is still only about 4 months old, with 1 month of beta testing, 18–19 full-time employees, and zero revenue or profit—but its product bet is already highly concentrated. The team is raising a Pre-A round that is now nearing completion; after Sora 2 emerged, it paused the original first-and-last-frame pipeline and decided to “stop building the old pipeline for now and switch everything to Sora 2.” It is a deliberate choice to trade consistency risk for stronger finished-video performance.
- 闹闹 does not believe future models such as Sora 4 and Sora 5 will simply swallow video Agents, because foundation models and vertical applications are more like a “market” and a “restaurant.” Models provide ingredients with different strengths, while Agents build knowledge bases, visual language and orchestration strategies around specific audiences; roughly “60–70% of the work” at OiiOii goes into seasoning and controlling the heat, not user-facing features.
- OiiOii’s initial wedge is neither the heavily distribution-driven manhua-drama market nor a UGC community that requires massive users and inference costs, but individuals and small self-media, animation and science-education studios. 闹闹 says these 2–3-person teams used to release roughly 1 episode a week; with OiiOii, they could theoretically release “10 episodes a day,” or 1–2 carefully selected episodes a day. The first value to validate is a step-change in production capacity.
- The beta revealed demand broader than the original target market, but the team plans to take verticals one at a time rather than serve everyone at once. Manhua-drama teams download only the efficiently generated storyboards and edit them themselves; non-professional users use animation to express mental states they do not want to show on camera; parents, couples, pet owners and adult children also use it for private relationships. The strategy is to turn the structure of each category’s best videos into a knowledge base and “攻 one by one.”
- One central challenge for Agent products is the structural conflict between freedom and reliability. OiiOii has rebuilt its architecture 4 times in 2 months: from free but disobedient System Prompt-based multi-Agent workflows, to stable but uneditable Workflows, and then to a system where Agents could jump out, modify the process and return. It still sometimes “can’t get out,” “stops moving again,” or forgets to return after jumping out.
- Sora 2 has changed the value distribution across editing tools: complex editing may be replaced by models first, while the simplest cutting, sequencing and TTS assembly remain. 闹闹 once thought transitions, effects and other heavy editing would be difficult to replace; she now believes models absorbed those techniques while training on video. The future is more likely to compress a 3-hour CapCut job into 30 minutes than to make CapCut disappear.
- 闹闹’s view of 2026 remains restrained: quality, editability and real-time performance will probably continue improving, but the technology leap that truly expands the audience may not be the grandest one. Stronger editing could push products toward professional users, while real-time interaction may be constrained by passive consumption habits shaped by “swiping up and down”; Sora 2 has already shown that simply making cuts more natural can create a much larger audience.
147: OiiOii 闹闹: “Douyin First, CapCut Second” No Longer Holds—How to Build Creative Tools and Content Communities in the AI Era
OiiOii is reversing the usual sequence by using creative Agents to unlock PUGC supply before building a community, targeting China’s 1.8M–2M ACG accounts and their small teams. Its moat lies in editable workflows and model orchestration, but freedom creates significant complexity; retention, consumption and the risk of becoming a bespoke To B provider remain unresolved.
View summary (7)
- 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.”