The barrier to creation has been lowered by AI; Bilibili now seeks to magnify the resonance that follows.

Business Sep 6, 2026

Bilibili has used the AI Creation Open Competition and its Toy platform to reduce the barrier to AI-driven product development. Relieved of server costs and distribution demands, amateur creators are able to obtain genuine user feedback, community engagement and iteration support — outcomes by which the rewards of creation are substantially amplified.

This article examines the "build in bilibili · AI Creation Open Competition" and the successes of the non-professional developers it surfaced. Diankao Huangdaiyu, a Bilibili user, built World Gate, an AI-generated open-world game, without assistance; his video has been viewed 4.96 million times. Zhake Ji, operating alone across three tracks, took awards for a personality test, a 3D shooting game and an abstract sculpting experience. Fan Ge of "AI Research Lab" produced AI glasses for the blind and drew large-scale user participation in subsequent iterations. Event data: 13,400 submissions; roughly two-thirds entered by developers without professional backgrounds; more than 80 percent the work of one-person teams; nearly 90 percent from accounts with fewer than 10,000 followers; half of all products completed in under 30 hours at costs below RMB 500. Consumption of AI knowledge content on Bilibili has grown 72 percent year over year, and monthly active users now stand at 371 million. By shouldering hosting and discovery through Toy and related tools, the platform has leveled the barrier to building and enlarged the returns to creating.

One all-nighter, 200 million tokens consumed, a 3D printer deployed, a RMB 4,000 bill paid, a 512GB Mac Studio acquired — and, after being reduced to a daze by relentless vibe coding, zero users to show for it.

With CodeX, many hopeful builders find WorkBuddy transformed into a wish-granting machine for product ambitions. Building the product, it turns out, is the least demanding element of the dream; attracting users, refining the offering and distributing it consistently present the greater difficulty.

If building a product with AI yields a baseline satisfaction of 100 for the average creator, the calculus changes once the product is used, played and anticipated by tens or hundreds of thousands; refined through the contributions of engaged and discerning users; and ultimately recognized with awards, revenue or commercial traction. Under those conditions, the original satisfaction is amplified by orders of magnitude.

The winners of the "build in bilibili · AI Creation Open Competition" now occupy precisely that position. [IMAGE:0]

An AI creation that astonishes others must first astonish its maker.

For 14 years, the Bilibili user known as Diankao Huangdaiyu consumed content without publishing any of his own. That posture has now changed: two uploaded videos, five minutes and ten seconds in combined length, have attracted 4.96 million views. The reversal owes nothing to algorithmic favoritism or to any innate gift for absurdist comedy or video-making. It is attributable to a single factor — a video introducing World Gate, an AI-generated open-world game built by his own hand.

The game has recently been recognized as an award winner in the "build in bilibili · AI Creation Open Competition."

Where open-world titles such as No Man's Sky and The Legend of Zelda: Breath of the Wild demand large teams and multi-year development cycles, World Gate has been built by a single developer, Diankao Huangdaiyu, in under six months. Its principal expense is token consumption, estimated at roughly RMB 1,500 per month.

What World Gate supplies, however, is the defining attribute of the genre: freedom. [IMAGE:1] [IMAGE:2]

In World Gate, any location may be selected by the player, whereupon new entrances and scene maps can be created and randomly generated NPCs engaged through AI-defined interactions.

It is the feature of which Diankao Huangdaiyu is most proud, as he explained to APPSO:

"World Gate differs from titles such as No Man's Sky in a fundamental respect: AI allows each individual's vision to be realized. As a Touhou Project player — a franchise encompassing games, music, manga, novels and compendia of settings, with Fields medalist Yu Deng among its admirers — I set out at once to construct a Touhou world. Stepping into it was overwhelming; after all these years, I was at last standing on that ground. It is an entirely different sensation."

The most consequential difference is that World Gate enables players to materialize the world they most genuinely crave.

World Gate remains in limited testing, unavailable to the general player base, though its momentum has already outpaced expectations. Upon first entering a world realized through vibe coding, Diankao Huangdaiyu reports having experienced what he describes as a chill down the spine; that same sensation, he reasoned, would appeal to kindred spirits. Views in the hundreds of thousands were the stated expectation for the demonstration video.

In the event, the target was surpassed by an order of magnitude. [IMAGE:3]

This is the archetypal Bilibili AI-creation narrative: tools such as CodeX have become near-wish-fulfillment engines, and Bilibili developers have used them to produce new engines answering both their own desires and those of their audiences. The threshold to realizing an idea has been lowered by AI, as has the cost — psychological and practical — of publishing it.

Video production is no longer the only route to creator status on Bilibili. Once Diankao Huangdaiyu judged that he had something sufficiently compelling to publish, the platform had a creator. That inflection point is attributable to AI.

AI has established itself among Bilibili's most resonant subjects. Consumption time for AI knowledge content has risen 72 percent year over year, and more than 190 million users now view AI-related content monthly. The platform serves not merely as a tracker of model and tool releases; it has also enabled innumerable former "invisibles," Diankao Huangdaiyu among them, to discover a source of visibility — inventing, building and publishing novel uses of AI.

One individual, three awards; ten thousand one-person teams.

On the evening of April 9, social feeds were saturated with the "SBTI Test." Where the MBTI presents an appearance of methodological rigor, with exhaustive questionnaires and, at best, marginal insight, the SBTI Test makes no such pretense: it is self-evidently unscientific, arbitrary and absurd.

Little matter. The working and school day is draining enough; the hours that follow are one's own, and one may be as irreverent as one pleases.

The "SBTI Test" ranks among the most successful products on Toy, Bilibili's interactive-content platform. A cluster of personality-test mini-games has since followed, among them the "Gaming Personality Test" by Zhake Ji, a Bilibili creator whose entry was likewise named a winner in the "build in bilibili · AI Creation Open Competition." [IMAGE:4]

For personality-test products, the Toy platform offers distinct advantages in development and distribution. Chief among them is scale: a viral product need no longer be exposed to the risk of server overload, or of capacity acquired at prohibitive cost.

The initial SBTI Test URL collapsed under a sudden influx of traffic after the author's Bilibili video went viral. The platform's own infrastructure has been built to absorb considerably greater loads.

Two further Zhake Ji entries were also recognized at the event. Soul Knight 3D, a fan-made reinterpretation of the 2D roguelike pixel shooter Soul Knight, is distinguished by the introduction of a vertical axis, which converts the game to 3D and liberates the camera.

The game, moreover, remains accessible via a single link on Toy.

The third winning entry, by Zhake Ji, is Meme Squish. [IMAGE:5]

The product has no utilitarian purpose whatever; it exists purely for amusement, which on Bilibili constitutes utility of the highest order. The mechanics of Meme Squish are uncomplicated: internet memes — including the mooing cow Niulai, the dragon Nailong, the barking dog Da Gou Jiao and the sword-and-shield dog Dao Dun Gou — are rendered as malleable clay figurines, to be reshaped into grotesque forms by the user and shared as original creations.

The product is also layered with memetic references. Excessive pressure applied to Niulai, for instance, elicits a plaintive "Mama—mama—."

The demonstration video for Meme Squish, with more than three million views, stands as Zhake Ji's most-watched upload. The appeal, on analysis, lies in escalation: an already abstract meme is rendered still more abstract and, critically, interactive. Users have been converted from spectators into contributors — petitioning Zhake Ji for new characters, circulating their own misshapen creations and, in numerous cases, adopting them as avatars.

Where Diankao Huangdaiyu's trajectory ran from spectator to creator, Zhake Ji arrived at the competition with an established following and standing in Bilibili's gaming community, having previously released numerous interactive works and videos.

The distinction this time lay in the competition and Toy, which relieved him of the burdens of sourcing infrastructure and audiences. Works acquired players; the account acquired new material; and his identity as a creator was, on that basis, renewed. [IMAGE:6]

A single creator waging three parallel campaigns and taking awards in all three is a statistically improbable outcome rendered near-inevitable by the format. By August 20, the competition had drawn 13,400 entrants. Approximately two-thirds were submitted by developers without professional backgrounds; over 80 percent by one-person teams; nearly 90 percent by accounts with fewer than 10,000 followers; and more than half of all products were completed within 30 hours at costs below RMB 500.

When the threshold to production is sufficiently reduced, the weight borne by the creator is correspondingly diminished.

The platform is capacious enough to accommodate three creations from a single author: a personality test of borderline seriousness, an engaging fan-made 3D shooter and a thoroughly absurd sculpting game. The significance is not that a single idea of a single person may be realized, observed and engaged; it is that all of that person's ideas may be.

Users, peers, would-be product managers: everything a product requires is present here.

In the age of AI, the act of creation is best understood as two distinct undertakings. The mechanical half — production — has had its barriers largely eliminated by tools; the generative half — origination — remains in short supply.

The competition, together with Toy, has served to activate creativity and engagement among a segment of Bilibili's user base, many of whom are not conventional video creators. Within the context of AI product development and the event, video has been adopted by these users as an instrument for demonstration, user acquisition and subsequent iteration.

Diankao Huangdaiyu, the author of World Gate, described to APPSO the circumstances under which a former Bilibili nonentity became a Bilibili creator:

"I had previously been something of a nonentity on Bilibili. In my view, a video was not worth publishing unless it contained something genuinely interesting."

"That changed on the night I found myself able to step into the world of my imagination within World Gate. The world itself, and the form of interaction it permitted, struck me as genuinely remarkable. Instinct counseled a video; the matter had to be shared."

It was only subsequently that World Gate drew public attention and, on the strength of it, entered the competition.

For Diankao Huangdaiyu, Bilibili's value resides in the unmediated market feedback it provides. Its users constitute the general public in miniature, and their verdicts determine whether an idea merits further development.

Among this broad audience, moreover, a substantial number possess genuine expertise; sustained engagement with them yields concrete avenues for product improvement. The platform's users demonstrate willingness to act on ideas, contributing effort and, in some cases, time and financial resources to their realization.

"AI Glasses for the Blind," submitted by Fan Ge of "AI Research Lab," ranks among the event's most humane and technically demanding projects. By pairing eyewear with multimodal AI recognition, the project seeks to confer a functional substitute for sight through audio and haptic feedback: spoken description of the immediate environment, supplemented by navigational cues and object-location alerts.

For a team of one or two, a functional demonstration of concept and features already represents a considerable achievement. Far more demanding is the production of a device that visually impaired users can genuinely adopt.

The obstacles to be surmounted were numerous: the prohibitive cost of continuously running cloud-based AI; the competing demands of high-performance components and compact eyewear on space and battery; recognition models initially limited to categorical scenes — fruit, beverages — and unequal to the complexity of real environments; and a reliance on tactile paving for navigation, paving that in practice is frequently obstructed by electric bicycles and shared bikes.

The project, fortunately, attracted growing assistance. Fan Ge's initial conception bore traces of assumptions made about blindness by the sighted, and its logic was not unassailable. Yet an expanding circle of contributors — their ideas, counsel and practical aid — has progressively refined the product.

The dependence on tactile paving — paving routinely rendered unusable — was addressed by a revision of the navigation rules: passage no longer requires the characteristic textured path, only terrain that is level and free of evident obstruction. That, in turn, demanded that the vision model grasp the concepts of "level" and "obstruction." To this end, Fan Ge, together with five faculty members and students from Anhui University of Technology, manually annotated more than ten thousand images to instruct the model in the recognition of passable terrain. [IMAGE:7]

A wider circle, including many blind users, has contributed suggestions and aspirations for the product. Numerous Bilibili users employed in AI and hardware have, in addition, volunteered their technical output to be open-sourced in service of further improvement. [IMAGE:8] [IMAGE:9]

Zhake Ji, purveyor of the outwardly bizarre and the deliberately absurd, is a doctoral student in materials engineering at the National University of Singapore: research by day, AI and absurdity by night. Fan Ge, by training a designer, holds no professional credentials in AI. The comment section beneath the eyewear video, meanwhile, reveals a visually impaired audience far larger than expected.

Bilibili — long styled by its users as the "little broken site" — now reports 371 million monthly active users, a scale bearing no relation to "little." The diminutive, in the platform's usage, may instead signify the opposite end of the scale: that the smallest idea will find its community here, and the most obscure subculture its counterparts.

Few communities make the discovery of kindred spirits so effortless.

It is precisely this interplay of the massive and the minute — and the friction between them — that generates possibility and novelty. Ideas, counsel, time, energy, technical output and, at the least, goodwill are accordingly offered up with willingness.

For Diankao Huangdaiyu, the first public product and demonstration video delivered an audience of millions. Zhake Ji's solo, three-front campaign was enabled by platform-managed distribution and by users who disseminated, parodied and engaged voluntarily. Fan Ge's selfless endeavor has been repaid in abundant encouragement and extensive pro bono assistance.

The satisfaction derived from AI creation does, at root, originate in the act of creation itself. Once a creation elicits joy, emotion, approval, applause, tears, whistles and gasps — a repertoire of "wow moments" — that satisfaction is, repeatedly and cumulatively, magnified.