An infinite world creation experiment of an AI otome game
A promotional video for an AI otome game has recently drawn widespread attention on social platforms such as Xiaohongshu and Bilibili.
Within the less-than-two-minute footage, a pink-haired idol, a Republican-era dignitary, a white-furred orc, a royal advisor, and an emperor appear one after another. Players can meet them across different worlds, experiencing hand-holding, ambiguous interactions, conflicts, and even dramatic "love triangle" scenarios.
Related videos have received over ten thousand likes across all platforms, with a flood of comments asking, "When will this game be released?"
Otome games are typically designed for female players, centered on romantic narratives and character relationship development. Traditional otome games are limited by high production costs, so most of them focus on a small number of fixed male protagonists.
This game, however, attempts to explore an alternative possibility: players continuously enter new worlds and meet new male leads with different identities.
The team spent three weeks completing the promotional video and an early gameplay demo. They hope to verify players' interest in quick world traversal, multi-world settings, multiple male leads, and AI-driven interactive experiences during the early development stage.
The project, named to reflect the concept of "countless worlds," is tentatively codenamed "Myriad."
Aishang Technology, the technical supporter behind "Myriad," is not a traditional game company. Its core product, PixVerse, mainly provides AI video generation services for overseas markets and has amassed over 100 million global users.
In July 2026, Aishang Technology announced that it had completed a total of 2.98 billion yuan in Series C financing, and began extending its video generation capabilities to real-time world models and interactive entertainment.
Qiushi, head of PixVerse Interactive Entertainment, told AllWeather Tech that the changes brought by AI are not just about cost reduction and efficiency improvement, but also make it possible to produce projects that would have been completely impossible in the past.
"With the advent of AI technology, we feel that the fantasy of multi-character narratives has for the first time become truly accessible. It can fundamentally solve the supply problem in terms of volume, allowing excellent characters, stories, and IPs to be incubated from this process," Qiushi said.
Working on projects like "Myriad," a team of more than ten people is simultaneously verifying two things:
Whether AI can transform the production methods of games, and whether it can create a kind of player experience that did not exist before.
Recently, AllWeather Tech had a conversation with Beibei, producer of "Myriad," and Qiushi, head of PixVerse Interactive Entertainment, trying to gain a deeper understanding of how this team develops AI games, as well as their definition of AI-native games and their vision for its future.
I
Quick World Traversal and the "Harem Dream"
The idea for "Myriad" originally came from producer Beibei's personal interest in quick traversal novels.
The so-called quick traversal means that players continuously enter worlds with completely different settings, complete tasks with new identities, and meet new characters. She wanted to create an otome game that truly allows players to travel to different scenarios and meet different characters.
In one chapter, the player might be a rookie actor in the entertainment industry; in the next, they enter a palace-intrigue world to survive; later, they could travel to a xianxia fantasy setting, an apocalyptic wasteland, or a wealthy family drama, encountering male leads and relationship dynamics that are all distinct from one another.
What connects all these worlds is that the player always exists as the same "self." The choices they make, the relationships they build, and the important events they experience in different scenarios will accumulate into a continuous "journey memory."
Whether in terms of content scale or interaction freedom, this is a product form that could only be truly considered in the AI era.
In the traditional 3D otome game production process, modeling a single character alone might take four to five months, with rigging and animation taking another two months. This is not a project that a small startup team could even greenlight.
But "Myriad" initially had only three female creators. In Beibei's words, "We didn't use AI video just to cut costs and improve efficiency — we used it because this quick traversal game concept simply couldn't be made without AI video."
At present, the team uses AI for image and video generation to help explore character appearances, scene atmospheres, and key visuals. The results still require manual screening and hand-drawn adjustments, while the plot, copywriting, and lines are all fully created by humans.
Beibei believes that generation capability itself does not equal gameplay. The real change lies in whether AI can penetrate into the core of the game, participating in shaping players' choices, world responses, and the evolution of character relationships.
This brings about far more complex problems than just controlling image quality.
Free input does not mean the system will obey every command from the player.
For example, if the player has just met a reserved male lead, even if they directly type "make him confess his love to me," the system cannot immediately generate an intimate scene. It must first determine whether this request aligns with the character's personality and the current stage of their relationship.
According to the team's settings, a reserved character can make an exception for the player after their relationship progresses, but cannot suddenly become overly affectionate and outspoken. A dominant character can retain their sense of oppression, but their behavior should never cross the line into being offensive.
The system will also extract important experiences from the player's historical choices.
For instance, if the player once protected the male lead, concealed a secret from him, kept a promise, or missed a confession, these events will be recorded as relationship facts that continue to influence subsequent side stories, dates, and companion content.
If the player often acts tsundere, teases the character proactively, or is very protective of them, the system will mark these traits as their personality labels, so that the male lead can respond in a way that is consistent with his own character design in subsequent interactions.
Only when all these boundaries are clearly defined will players believe that "he really remembers me," rather than assuming the model is just generating random lines based on the current context.
However, "Myriad" is still in the demo and capability verification stage. The team has not yet found a stable balance between product requirements, player freedom, and the current capabilities of existing models.
Beibei's long-term vision is to allow players to create their own worlds, characters, and storylines. They can not only create new male leads, but also bring official characters into a completely different world to restart their relationship.
Right now, the game team at Aishang Technology is trying to implement this process through an AI game editor.
Players can describe the worldview they want in natural language. The system will then gradually confirm details such as the era and background of the story, as well as the number and personality labels of the male leads, and finally generate a playable game demo.
If this path continues to move forward, the full version of "Myriad" will hopefully realize its multi-world relationship vision:
Players will no longer be limited to a set of pre-designed male leads created by producers, but will enter a game space that can continuously generate new worlds, new characters, and new relationships.
II
Creation Itself Becomes Gameplay
The market is full of imagination for AI-native games, not just because AI may spark another efficiency revolution in game production, but more importantly, because it has the potential to create levels of freedom and content scale that traditional development methods can hardly achieve, thereby redefining the boundaries of interactive entertainment experiences.
In early 2026, Aishang Technology released its real-time world model PixVerse R1, which can continuously generate audio and video streams and modify subsequent content in real time based on user input.
In June of the same year, based on PixVerse R1, Aishang Technology launched its self-developed AI game engine PixVerse Game, aiming to verify whether real-time generation can evolve from a content production tool into an integral part of gameplay.
PixVerse Game consists of three core components: a real-time world model, an AI Agent orchestration layer, and a game mechanics engine.
The real-time world model is responsible for continuously generating visuals, sounds, character animations, and environmental changes. The game mechanics engine manages numerical values, resources, mission conditions, state transitions, and win/loss outcomes.
The AI Agent sits between the two, responsible for understanding player intent, scheduling corresponding capabilities, and ensuring that generated content stays synchronized with the established game rules.
When a player says "summon an ice wall to block the pursuers," the system not only generates the corresponding visual, but also determines whether this action conforms to the current ability level and gameplay rules, updates the position and movement of the pursuers, and carries the result over to subsequent storylines.
Qiushi describes this game experience as "speech that immediately manifests as reality."
He told AllWeather Tech that the indie game *Mage Arena* has already preliminarily demonstrated the appeal of natural language interaction. In this multiplayer competitive game, players take on the role of wizards and cast spells by speaking into the microphone.
*Mage Arena* is not a fully realized AI-native game. But in Qiushi's view, it at least proves that when language becomes a direct operation method, even with simple gameplay, it can bring a fresh experience that is completely different from traditional keyboard and controller inputs.
AI has the potential to take this "speech becomes reality" experience a step further. Players will no longer be limited to uttering a few pre-set spells defined by the system, but can freely describe the form, target, and effect of their magic. The system will then understand their intent, check against the game rules, and generate the corresponding performance.
Looking further, the decoupling and recombination of character assets, world settings, and gameplay mechanics could even allow the same set of characters to appear in different types of games.
Qiushi used *Honor of Kings* as an example to illustrate this point. The game has hundreds of heroes, each with their own unique background stories and skill systems. With traditional development methods, porting all these characters into auto-battler, card, otome, or racing games would almost mean remaking several separate games from scratch.
But if a company can use engineering tools, multimodal models, and large language models to split character designs, visual assets, and gameplay rules, and then recombine them, the same set of characters could potentially be adapted into different game genres in the future.
In that scenario, players will no longer just purchase a fixed game product, but gain access to a shared cast of characters and a connected world that can span different gameplay styles.
However, this is still a theoretical vision that PixVerse Game hopes to verify.
In the currently available demo, the team has tested three gameplay types: turn-based, RPG, and roguelike. One user once spent about 20 hours continuously generating game content set against the background of the Three Kingdoms period.
But broader market feedback has remained relatively muted so far.
"People usually expect world models to power MMOs, open-world games, and real-time combat games. The three gameplay types we tested currently don't deliver enough of that 'exciting feeling'," Qiushi admitted. The generation stability, spatial consistency, and response speed of PixVerse Game still lag behind mature commercial games.
The team plans to add character pathfinding and spatial stability capabilities in future model iterations. For example, when a player walks from point A to point B and then returns, objects in the scene will remain exactly where they were placed.
Qiushi told AllWeather Tech that Aishang Technology hopes to attract developers from various industries to participate in exploration by opening access to the demo, and work together to find feasible ways to deeply integrate generation technology with games.
III
The Real Gap to Commercialization
Even if technical problems are fully resolved, the core question that AI games still need to answer is: What exactly are players willing to pay for?
One of the most straightforward approaches is the token economy model. Players pay for exclusive storylines, custom characters, or world-modifying capabilities based on the number of generation requests.
In May this year, the domestically developed AI historical strategy game *Historical Simulator: Chongzhen* sparked controversy over this payment model. The game allows players to issue imperial edicts in natural language, and then the AI simulates changes in the national treasury, natural disasters, wars, and the political situation in the imperial court.
Theoretically, players can make choices that traditional strategy games cannot pre-program, such as lifting maritime bans, appointing unconventional officials, or even distributing modern supplies to ancient armies.
But this freedom also means that every command and world simulation requires calling the model. The game was initially priced at 48 yuan. After purchasing the base game, players needed to consume additional paid tokens to continue advancing their turns.
After the controversy, the developer announced that they would open up custom APIs and a creative workshop, allowing players to connect their own models to control costs. On June 25, the game went free-to-play for the base game, retained the point consumption model, and provided compensation to players who had previously purchased the game.
In Qiushi's view, the token consumption model widely adopted in the short drama industry cannot be directly applied to the gaming sector. In short drama production, the production team still calls the model to generate all content before users watch it. AI in this scenario mainly reduces the production cost of hit content.
But if games charge by tokens, that means players will have to keep paying for every story generation and every interactive operation.
Qiushi said: "Ordinary users may not have strong enough creativity or the desire to create. Your product must have an extremely low entry barrier, while also delivering stable and high-quality output."
Another path is to return to the traditional commercial logic of the gaming industry. Players pay not because the product uses AI, but because AI can provide personalized experiences that traditional games cannot achieve at scale.
"If the content we provide is unavailable in other games and only we can deliver it, the product will naturally have pricing power," Qiushi said.
From Minecraft mods and custom maps, to UGC levels in Roblox and *Eggy Party*, the new generation of players is already very comfortable with the idea of creating while playing. But there is still a long way to go to turn this demand into a large-scale profitable business.
What is restricting the development of the industry is not just technology and content supply, but also the immature commercialization and distribution systems.
Behind the boom of AI short dramas, a complete ecosystem covering content production, platform procurement, traffic boosting, recommendation, and payment conversion has already taken shape.
The gaming market is not short of distribution channels, but platforms like Steam, mobile app stores, TapTap, and various mini-game platforms are fragmented and disconnected from each other.
Even if AI can generate games in batches, there is still no clear answer to who will curate this content, where it will be distributed, and how ordinary users will discover these games.
"If you ask consumers now whether they like AI games, they won't really know, because they haven't experienced one yet. People can only evaluate things they have actually tried," Qiushi said.
This has trapped AI games in a cycle: due to the lack of mature monetization channels, the track has not yet been occupied by large tech companies. But without platform resource investment, content supply and distribution mechanisms can hardly mature quickly.
Qiushi believes that in this context, even if the production side is fully AI-native, if products still follow the traditional "develop first, then publish" path, AI games will hardly see large-scale explosive growth.
"There are far too few AI games on the market right now — far fewer than there should be," Qiushi said.
At the moment, "Myriad" has only completed the first step from concept to demo.
It features constantly changing worlds, a large number of male leads with diverse background settings, and the vision of making storylines respond to players' personal expressions. But it still relies on manual work to build the core framework of the content, and is constrained by model stability, operating costs, and content review requirements.
AI has not yet created a game world where players can act with unlimited freedom, but it has already made the quick-traversal otome game with "a hundred male leads" no longer just a producer's fantasy.
The next step is to prove that this possibility is not only technically achievable, but also engaging enough to make players want to stay and explore.