From Character.AI to OOC: How the New King Breaks Out as the Old One Falls

十字路口Crossing·October 18, 2026·0·3

Why is it OOC?

Last night (October 17), at the afterparty for a Crossing × Table for Six co-branded event, we held a roundtable discussion at Shanghai AI Hacker House.

From the old king Character.AI's bankruptcy restructuring[1] to the new king OOC[2] with $70 million ARR: behind this changing of the guard, new opportunities in AI companionship are emerging.

Roundtable Guests:

Vito Zhong｜Founder, MobAI | Product: Lunaverse Stories[3], exploring AI interactive narrative by combining text, images, video, mini-games, and character relationships into a single experience. His question: how do you let people who love stories easily enter one?

Cheese[4]｜Producer, 1001 Nights[5] | Indie game creator. In 1001 Nights, players tell stories to an AI king, then extract weapons from those stories to challenge him.

Xiaoning｜Host, 405 Game Bureau[6] | Continuously following gaming and AI interactive entertainment. In this discussion, she breaks down this wave of change from the angles of product experience, business model, and game design.

Moderator:

Koji Yang | Founder of Crossing, Investment Partner at ZhenFund

The following is edited from the live discussion. Some long answers have been split into Q&A format with added transitional questions; speakers' views are preserved in their original intent.

Chatting is also exhausting

👦🏻 Koji

Character.AI recently announced bankruptcy restructuring proceedings[1]. As practitioners in this field, how do you feel?

🧑🏻‍💻 Vito

I'm actually not that surprised. It's already a very old product for us, not even on our competitor list.

I tried Character.AI[7] in 2024 and 2025, and even then felt its product was falling behind.

Making a companionship product looks simple: open ChatGPT[8], input a prompt, and you have the simplest companionship product. But whether you can actually get content and emotional fulfillment from it is very difficult. That's also a key differentiator between good and bad products.

🧙🏻‍♀️ Cheese

I'm not as shocked as I might have been, but I do find it regrettable. Not long ago, I gave a remote presentation to their team about interactive narrative. I have friends working there, and from the quality of on-site questions and everyone's energy, it was a team atmosphere I quite admired.

This makes me revisit a question: many companionship, semi-companionship, and semi-story-generation platforms look quite similar. Of course, their cores and emphases may differ.

But what are they actually competing for? Why do people tell stories, and why do they come to these products? Beyond existing forms, are there other approaches? As an indie game developer, I constantly think about these questions.

👦🏻 Koji

So, have you found answers?

🧙🏻‍♀️ Cheese

No. If there were answers, there wouldn't be so many failed games and products.

👩🏻‍💻 Xiaoning

What concerns me more is whether this will make the already chilly AI application funding market even colder.

As for the product, I think AI application entrepreneurship today may require teams to produce new things faster. What brought you past success can also become baggage.

The new king with $70 million ARR: OOC

👦🏻 Koji

OOC[2] has rapidly grown to $70 million ARR, attracting considerable attention. What did it do right?

👩🏻‍💻 Xiaoning

Let me start with the experience. Many past products centered on a single character: you could create a virtual character or a real person, and the core experience was chatting with them.

Chatting is actually quite difficult, don't you think?

Suppose I really want to chat with Koji but can't reach him, so I create a Koji. But then, I have to initiate every conversation. I have to think about what to talk about. Even with the most charismatic person, after hundreds of rounds, you might get tired.

OOC adds world-building and background. It tells you why Koji is here, what's happening in this world, and that you're part of the story too. With background, the story gains momentum to move forward.

👦🏻 Koji

So users don't always have to come up with topics?

👩🏻‍💻 Xiaoning

Right. And its interaction isn't just back-and-forth chatting; it incorporates game-like design such as nurturing and numerical growth. Under the same world-building, many different stories can happen.

This is somewhat like reading a novel. In a good novel, world-building matters because it opens your imagination, making you wonder what else exists in this world.

👦🏻 Koji

What kind of content do people like most in OOC now?

👩🏻‍💻 Xiaoning

Haven't you tried it yourself? How did it feel?

👦🏻 Koji

I entered a magic academy: a group of superpowered boys and girls falling in love on campus, with me playing one of them, and I even created a guardian cat for myself.

What I found interesting was that due to improved long-context agent capabilities, it could remember quite a bit of context, with fast enough responses. I said the cat would give me some superpowers, and later when meeting a new character, his eyes would flash with a light only recognizable between cats.

However, magic academy is still a genre that matches my expectations. Have you observed any new content themes emerging?

👩🏻‍💻 Xiaoning

It's quite diverse now; I personally prefer sci-fi. What the most popular world is doesn't matter that much to me. The key is having enough choices — you can enter a magic academy or a completely different world.

🧙🏻‍♀️ Cheese

I'd describe the experience OOC provides as "text-based short dramas." One particularly noteworthy aspect is how quickly you can enter the situation.

It's not as simple as placing a character and giving an opening line; through good opening design, users know within ten seconds where they are and what their initial motivation is. If the story doesn't attract you, you swipe away and try another.

From a design perspective, I really appreciate this fluid sense of entering situations.

Pay first for a good experience?

👦🏻 Koji

Beyond world-building and interaction, why does this generation of products like OOC look different commercially?

👩🏻‍💻 Xiaoning

Model intelligence has improved rapidly this year, but we need to look at it together with business models.

Previous-generation products tended to encounter two types of users. One type already knew how to play with AI, even using "tavern" (SillyTavern)[9] to build their own experiences. With good models and world books, they could accomplish much on their own, so the additional value provided by platforms wasn't large enough.

The other type found AI chatting novel and were willing to try it for free. Many apps used this funnel: attract users for free, then gradually convert some to paying.

But if you're a product manager who needs to provide chat for large numbers of free users, what model do you choose? Usually only cheaper ones, otherwise model call costs become hard to bear. Now with images and video, the cost pressure is even greater.

This easily creates a cycle: to lose less money, you limit model capabilities; with subpar experience, users are even less willing to pay.

👦🏻 Koji

So OOC's choice was to charge earlier?

👩🏻‍💻 Xiaoning

Right. From the start, it aimed to capture people who already understood this type of experience and were willing to pay. You might need to pay after just a few messages.

👦🏻 Koji

That was my experience too. It's like reverse-filtering users: if you only want to play for free, it identifies that from the start and doesn't serve you; it filters for people willing to pay, then continues.

👩🏻‍💻 Xiaoning

The benefit of this approach is being able to use better models and provide better experiences. It's a different commercial choice.

For products that have already accumulated large free user bases, pivoting to this path would be very difficult. Past user scale can also become a burden for transformation.

👦🏻 Koji

Does OOC's model have opportunity in China?

👩🏻‍💻 Xiaoning

If copied directly, I think it's relatively difficult now. Content compliance, the capabilities of domestically available models — these are factors to consider. China may need a different model.

But I know many domestic teams are going overseas. Chinese teams have considerable experience in refined operations for user acquisition and user operations, so more products may emerge overseas.

👦🏻 Koji

Is it that products like OOC may not have very high DAU, but have relatively high per-user payment?

👩🏻‍💻 Xiaoning

Public data is limited. From some situations I've seen, if judged by traditional internet product standards, their next-day retention may not look great, but long-term retention can be decent, and monthly revenue per user may be higher than expected.

However, they're also facing new questions: have they already reached most users who like this type of content, easily get hooked, and are willing to pay?

To expand further, there may need to be recognizable IPs or breakout hits, like with short dramas. We've talked for so long, but people who haven't played may still think: it's just text chat, how fun can it be? There need to be some truly cross-over hits to get more people willing to enter.

👦🏻 Koji

Cheese, would you play such products long-term?

🧙🏻‍♀️ Cheese

I quite like some of its design approaches, but I personally find it hard to play long-term. This may just be personal preference.

Our team, including part-timers and interns, has about a dozen people. Maybe one or two play otome games, but no one heavily uses companionship software. This doesn't mean there's no demand for companionship, but rather that many people aren't covered by this category.

Everyone knows companionship can create strong stickiness, but it's not everything. Indie games value strong differentiation highly, which has different logic from products that try to serve as many people as possible from the start.

Hiding real-time generation

👦🏻 Koji

Vito, what did you see when analyzing OOC?

🧑🏻‍💻 Vito

Its marketing is very well done. OOC was already an in-community term (Out of Character), referring to when a character's behavior exceeds their original setting. Using it as a name makes people familiar with OC and character creation immediately feel this product was made for them.

Its user filtering, design style, and marketing are connected. Products like Tipsy[10], Emochi[11], Korea's zeta[12], and OOC[2] are actually quite similar; achieving differentiation and then finding a user base isn't easy.

But I think the biggest problem in this track is: not that many people have that much energy for input.

👦🏻 Koji

By "input," you mean creating world-building and scripts?

🧑🏻‍💻 Vito

Whatever kind of input, it's quite troublesome. After work, I might have energy to watch a short drama or a variety show, but not necessarily to open a competitive game for an intense match. They demand different levels of attention and energy.

And the users targeted by OOC-type products are people who already like stories and content. I can read novels, watch TV, watch anime — I don't necessarily need to play a game that requires me to create. From this angle, asking users to create themselves is quite "anti-human."

👦🏻 Koji

So what different choices have you made?

🧑🏻‍💻 Vito

What we're doing at MobAI now has similarities with OOC, but the biggest difference is we don't encourage users to generate in real-time; we even hide that entry point.

We prefer users to first experience prepared content and storylines. The content format is quite mixed: text-image presentation, video, mini-games, voice — all combined.

Beyond that, we also want users to build relationships within it. Whether to add chat functionality — we argued about this internally for a long time. My co-founder felt that making an on-call AI virtual companion is just too boring.

👦🏻 Koji

Then what does an AI character who isn't on-call look like?

🧑🏻‍💻 Vito

We referenced Bubble[13], used by K-pop fans. It gives fans an experience of being able to chat with idols.

We're making a romance game, where romanceable male characters have their own in-game social media accounts. You have the chance to add them as friends, but they have their own daily schedules.

Sometimes you open the product and find he's sleeping. No matter how many messages you send, he won't reply. He'll also proactively share his life and manage his social feed, more like an idol with his own life.

👦🏻 Koji

Is this content all self-produced by you, or will you open it to creators?

🧑🏻‍💻 Vito

In our product, most stories already launched are self-produced. We've also started finding overseas creators to use our content creation tools. They've submitted many bugs, and their time zone happens to be opposite ours, haha, so why I look so tired now — it's about when they wake up and start submitting bugs.

Building the entire content production pipeline is very important to us. In this process, we've genuinely felt that AI capabilities have changed greatly this year.

👦🏻 Koji

Since AI's generation capabilities are getting stronger, why反而 hide real-time generation?

🧑🏻‍💻 Vito

We previously made real-time generation products and saw their limitations there. Very refined content requires design; we currently don't believe this can be entirely handed over to AI for real-time generation.

We place more emphasis on "generative recommendation": after users play for a while, the system knows what they might like, spends an hour or two in the background generating higher-quality content, then pushes it to them.

Is creation anti-human, or human instinct?

👦🏻 Koji

Cheese, your 1001 Nights happens to require players to tell stories themselves. First, introduce how it works?

🧙🏻‍♀️ Cheese

In One Thousand and One Nights, the heroine tells stories to the king. If they're good, the king keeps her until the next day.

In our game, you also tell stories to an AI king, but you have a special ability: you can pull things out of stories. You and the king do story continuation; if his response contains knives, swords, shields — things you need — you can extract them and later use them to beat the king.

What we want to explore more is: is there a narrative framework that's easier to enter, that lets people naturally participate?

So when Vito just said creation is anti-human, my view is exactly the opposite: creation is the most human, most belonging-to-human part of us.

👦🏻 Koji

How do you understand this?

🧙🏻‍♀️ Cheese

People have always had the desire to express and leave their mark. As children, you might have told stories to friends; even if not stories, you've surely gossiped with colleagues? Much of what people listen to and watch is fundamentally story.

When we say we can provide more content, as platform or tool providers, we may of course benefit. But as a player, I also ask: if everyone becomes consumers, who tells the stories?

Some might say, whoever has more compute power, or whoever can hire more writers. What we want to offer is another approach: can we let anyone naturally become a storyteller in the process of playing? Without making them first distinguish: am I playing now, or creating?

👦🏻 Koji

Actually real life is like this. We live our lives while also creating stories that others can hear and see.

🧙🏻‍♀️ Cheese

Right, it's just that you usually wouldn't call that "creation." If I directly say "come, tell me a story," you might say, I can't, I'm not a writer.

We're using games to let people rediscover they have this ability, or give them a situation where they can start.

👦🏻 Koji

Is this a vision, or have you already seen this in players?

🧙🏻‍♀️ Cheese

If we isolate our game's requirement — "you need to write a story" — doesn't that sound intimidating?

👦🏻 Koji

Yes. My immediate thought would be: I have to tell stories to the king to avoid death, and I have to pay to do this?

🧙🏻‍♀️ Cheese

We have a free demo now, so it's fine (laughs). But people who actually play often don't think this way.

They tell me they initially just wanted to tease the king, or see if they could make him "escape." After getting scolded by the king, they started telling stories. In the process of playing, they didn't feel so much pressure.

👦🏻 Koji

So the key is transforming "I need to write a story" into "I need to complete an objective in the game"?

🧙🏻‍♀️ Cheese

Games can establish a very special value framework. Socializing with people in real life has pressure; writing essays has pressure, but put it in a game, and it's a different situation.

Most people don't come for "telling stories." They're thinking: I'm the heroine, I'm going to fight to the death with the king, this game looks unique, let me try.

👦🏻 Koji

What specific changes have recent AI model advances brought to your experience?

🧙🏻‍♀️ Cheese

This project started as my PhD topic, later becoming a startup. We've always had many ideas, but implementation used to be relatively costly.

For example, previously it was mainly player speaks, king responds. We wanted characters in the story to jump on stage and perform like actors. Now Jev[14] can make judgments faster; right after the player finishes, characters on stage can immediately show fear, terror, or joy.

Adding this extra layer in the past might have meant another API call and cost; now the conditions are different.

👦🏻 Koji

Has your internal development efficiency improved?

🧙🏻‍♀️ Cheese

There's also been big changes. I think Claude Opus 5.5[15] and Astra[16] have contributed here. Our iteration rhythm changed from once every few weeks to once a day, or even half a day.

I come from an art background with no development experience. Previously, after designers made plans and designed gameplay, there was a long distance to actually implementing them in a game engine.

Games aren't just writing code; they involve 3D, collision, and various engine things, which in the past were hard to complete through Vibe Coding alone.

Now using AI coding in our game, the experience is increasingly approaching making web demos. Roughly estimated, with just my evening time for R&D, two weeks of results might equal what a programmer could do full-time for two months in the past.

Don't be real-time for real-time's sake

👦🏻 Koji

Finally, please share what new signals in the AI and interactive content direction are worth paying attention to.

👩🏻‍💻 Xiaoning

I actually feel these past few months are when everyone's been heads-down working. After new models came out, many creators found methods and possibilities changing rapidly, so they all went back to work.

I quite look forward to answering this question again in two or three months.

👦🏻 Koji

Are people discussing Vidu S1/S2[17], Vivix[18], and fal's post-trained MiniMax H3 Max[19] — these real-time or high-speed video generation models?

👩🏻‍💻 Xiaoning

I think "real-time" is being demystified. I've made games before, also sandbox games, and always had many imaginings about real-time generation and real-time interaction.

But first, the "real-time" being discussed today still has distance from the reaction speeds truly needed in games.

Second, technology is already changing so fast — isn't it more important to first produce some content? Being real-time for real-time's sake sounds like technical language, not game language.

👦🏻 Koji

Compared to pursuing real-time, what specific changes concern you more now?

👩🏻‍💻 Xiaoning

One is 3D. Not just looking at whether AI can generate a model, but whether its capabilities for decimation, retopology, model cleanup — can these enter a mass-producible pipeline.

In game companies, 3D used to be an important threshold between small and large companies. Changes in this capability are very worth watching.

Another is interaction design, how to reduce user pressure. Chatting is actually quite taxing: you put in a lot of effort, so you expect higher rewards. Watching short dramas and playing games, because the investment differs, the stimulation you expect differs too.

I think there will be many experiments in how to let people participate more easily.

🧙🏻‍♀️ Cheese

I want to follow up on that. More investment, higher expectations — that sounds like a difficulty. But from another angle, if the returns after investment are high enough, surprising enough, it could become a new form of play.

For example, there's recently a dress-making game Needle and Thread, Shadow and Dream[20]. You actually cut fabric and make dresses; what you do isn't easy itself.

It doesn't directly involve AI, but I pay attention to such games because I don't just look at what people play, but what a game lets people create.

Before that there was also a hide-and-seek game MECCHA CHAMELEON[21], letting players draw on characters. Drawing and making dresses look completely different, but they both transform a form of creation that originally seemed difficult into an experience players are willing to participate in.

👦🏻 Koji

So creation itself can also be something we're willing to consume?

🧙🏻‍♀️ Cheese

What I'm asking is: when content becomes cheap and easy to produce, is there value in getting people to create?

I don't have a definite answer now, but these games make me feel this question is worth continuing to ask.

👦🏻 Koji

I think creation itself can be consumed. Like people paying to do pottery — no one questions: I'm creating with my own hands, so why pay?

🧙🏻‍♀️ Cheese

Right, isn't it interesting?

👦🏻 Koji

Vito, what change has most moved you recently?

🧑🏻‍💻 Vito

What has most moved me recently is Claude Opus 5.5[15]. We have many video production needs and are building our own marketing video production pipeline. After the new model came out, some previous limitations could be discarded.

We initially thought the threshold for video generation was video models. Later we discovered many problems still need to be solved by the workflow layer outside.

Previously when looking at video models, people looked at how many seconds per generation, whether reference images were supported; now they look at whether it can be real-time, whether it can stream. But more important than just these metrics is its understanding of content.

Can it understand editing, understand what expression is smooth? When models have stronger capabilities in this, the final results are noticeably different.