Kaiying’s SoonPlay Opens Public Beta: Deconstructing the ‘TikTok of Games’ — a Chinese AI UGC Platform Where Videos Become Playable Games#

By Wubian (NTGame Editorial · AI + Gaming Watch Column)

Hello friends, I’m Wubian.

On October 20, SoonPlay — the AI casual-game platform from Jiyi Artificial Intelligence, a subsidiary of Kaiying Network — officially opened its public beta. The official press positioning calls it “the world’s first AI casual-game generation interactive platform” — to be clear, this is Kaiying’s own phrasing, and no third party has endorsed the “world’s first” claim. Per the official description, users can both play games on their phones and make games with SOON AI; for now, the public beta entry point is downloading the experience from the Jiyi Artificial Intelligence website.

In media coverage, the product has an even more widespread nickname: “the TikTok of games.” That analogy comes from a Zhongguancun Online headline and has since been adopted by many industry write-ups — it is a media analogy, not an official self-description. But the analogy does land: per the official framing, SoonPlay connects the three scenarios of “watching videos, playing games, making games,” and every video on the platform corresponds to a game generated by SOON AI — effectively mapping the short-video feed directly into a game feed.

Over the past week, we have covered this track from two angles: a few days ago, my Google Playground breakdown covered the impact of the overseas ecosystem launch, with a companion technical deep dive by Zixing; and yesterday’s Steam Next Fest AI-backlash observation covered the creator-ecology controversy and player pushback. Today’s piece turns the lens back to China: SoonPlay is a domestic single product already in open beta and genuinely downloadable — together with the two above, it completes a three-perspective set of “domestic open-beta product / overseas ecosystem launch / creator-ecology controversy.”

First, on sourcing: every specific fact in this article (dates, features, figures, statements) treats this round’s verified fact card as the sole authoritative source; single-source data is flagged in place, undisclosed official information is labeled as such, and no speculation is offered.

1. Video-to-Game: the Creation Starting Point Shifts from “Writing Prompts” to “Picking a Video”#

1.1 The Official Generation Path#

Per Kaiying’s official release, SoonPlay’s video-to-game path breaks into four steps:

1. The creator uploads a video;
2. SOON AI identifies the gameplay elements in the video;
3. The platform automatically generates a game — no programming background required, and no fiddling with prompts;
4. The generated result can be edited, play-tested, published, and shared on mobile at any time.

The most notable line in the official wording: every video on the platform corresponds to a game generated by SOON AI, instantly converting “curiosity while watching a video” into a game experience. That sentence is the product’s crux — it compresses the distance between “seeing something interesting” and “playing the corresponding game” into a single click.

1.2 Why Video, and Not Prompts#

Tan Kai, CEO of Jiyi Artificial Intelligence, previously made a remark that serves as the official footnote to this product logic: traditional AI game generation requires precisely written prompts to describe gameplay logic, but game logic architectures are complex and ever-changing, and the prompt barrier discourages many creative people. SoonPlay’s answer is to add “pick a video” as a creation starting point alongside “write a prompt.”

The difference between these two starting points becomes clearer in the recent industry context. I compared SoonPlay’s creation flow with Google Playground, which launched a few days earlier:

From “one sentence” to “one video” looks like a single step, but the barrier lands in a completely different place: a natural-language description still requires you to “think the gameplay through before saying it,” whereas a video is itself an already-established gameplay demonstration — the creator doesn’t need to organize words; the material speaks for itself. Following the industry-observation thread, the creation starting point has roughly climbed three rungs: first writing prompts, demanding gameplay logic be spelled out in text; then providing reference material for the model to imitate; now SoonPlay places the starting point directly on a complete video, effectively skipping the most brain-intensive step — “describing the gameplay” — entirely.

1.3 Mobile Is the Premise of This Design#

One easily overlooked detail in the official framing: editing, play-testing, publishing, and sharing all happen on mobile. This choice is bolted to the “watching videos” scenario — video-watching already happens on the phone, so the game-creation experience lives on the same screen; no device migration is needed between “curiosity while watching” and “the act of playing,” minimizing funnel attrition. Desktop AI game tools think “move development into the browser”; SoonPlay thinks “move creation into the feed” — the two target fundamentally different usage postures.

2. “The TikTok of Games”: the Interaction-Conversion Logic of a Three-Scenario Loop#

2.1 How the Three Scenarios Interlock#

The official framing summarizes SoonPlay’s use into three scenarios — “watching videos, playing games, making games” — with four “light” pillars at its core: light creation, light experience, light cost, and light presentation, which the company calls an exploration of “content and games generating each other.” Unpacked, the three scenarios carry different roles:

The interlock is a loop: you watch a video, play its derived game, tweak your own version and publish it, and become content someone else “watches.” The classic short-video loop is “watch—like—shoot”; SoonPlay swaps the middle “like” for “play,” and the “shoot” for “mod the game.” That is the real structural support for the “TikTok of games” analogy — not surface resemblance, but loop isomorphism.

2.2 Where the Analogy Holds and Where It Hangs in the Air#

The holding part has been stated: feed and game flow map one to one, and the three scenarios connect end to end. The hanging part must also be said: a short-video platform’s real moat is its recommendation algorithm and creator incentives, and in SoonPlay’s currently public information, neither the recommendation mechanism’s details nor a creator-incentive scheme has been disclosed. On the loop diagram, the three spots — “what gets fed to you,” “why you keep scrolling,” and “why creators stay” — are blank today. Undisclosed official information cannot be invented on its behalf, so my conclusion: the analogy holds at the product-structure level and still lacks evidence at the ecosystem level. That half-sentence of reservation is worth far more than mindlessly hyping “the next TikTok.”

3. Full-Stack Layout: SoonPlay Is Only the Tip Above Water#

3.1 A Three-Layer Structure of Platform, Large Model, and Dev Tools#

Per the official release, Jiyi Artificial Intelligence has completed full-stack integration of game platform, large model, and development tools. Within that stack, SoonPlay is not everything — it is the mass-market layer. Here is the three-layer structure as a table:

This division of labor deserves a second look: SoonPlay turns “making games” into a mass-market action, SOON keeps “making games” at professional depth, and the same large-model foundation strings them together. The mass end runs volume, the professional end runs quality; volume products sediment methodology upward, and professional assets lower generation costs downward — in the ideal state, a self-sustaining dual product line. Of course, the words “ideal state” must be kept: the actual effect of dual-line synergy awaits open-beta data.

3.2 SOON’s Workflow System and the AP Tier System#

Two details from the Zhongguancun Online report deserve separate treatment.

The first is the workflow system: SOON has consolidated over a hundred development workflows — character animation, game scenes, primitives, game UI, adaptation optimization, render acceleration, and more — into a single AI tool stream unified with AI capabilities. The significance: “making games” as a verb is decomposed into hundreds of workflows that AI can take over one by one, rather than a black box where you press a button and get a result. Workflow-ization means errors can be located, components swapped, and progress accumulated — fix whichever step breaks. That is the dividing line between a toolchain and a toy.

The second is the AP tier system: AP stands for “AI+People,” presenting a growth path for platform-defined works from AP1 to AP6, conveying the message that “generation is not the endpoint — human and AI collaborate continuously”; SoonPlay carries this system over. In my view, the AP system is setting expectations for the UGC ecosystem — it attempts to answer “does a generated work count as a work?”: generation is only the starting point, and the tiers are the yardstick of ongoing human-AI collaboration. What UGC platforms fear most is “one-shot generation” — works posted once and dead. Offering a climb from AP1 toward AP6 is effectively coaxing creators to come back and keep editing, keep building. Whether this narrative holds depends on how many works actually climb the tiers.

3.3 Filing First, Beta Second: the Compliance Sequence#

The timeline, organized:

The filing details, per the screenshot of Zhejiang Province’s newly-added generative AI service filing list embedded in the official release:

Filing first, public beta second — this sequencing is worth domestic peers noting: for a generative AI product to serve the public, filing is a precondition; during the SOON platform’s second closed beta, 3,000+ games had already emerged, showing the generation pipeline ran one controlled lap first, and the public beta scales what has been validated. For Chinese AI game products, this is a directly referenceable compliance sequence — model filing first, platform beta second; reversing the order invites trouble.

4. No Ads, No IAP: the Acquisition Hook vs. the Long-Term Monetization Paradox#

4.1 The Current Business-Model Wording#

Per currently available public information, SoonPlay at this stage touts no ads, no in-app purchases, and unlimited play time, with a newcomer pack and launch benefits for first-batch users. The sourcing must be stated precisely: “no ads, no IAP” appears in SoonPlay’s promotional materials and third-party coverage — the official release’s body text does not spell it out verbatim; “unlimited play time, newcomer pack” comes from a single-source report by Yaotu Network. In other words, this bundle of selling points most likely reflects real product strategy, but strictly speaking it is not officially confirmed word-for-word — carry that caveat when citing it.

4.2 Where the Paradox Sits#

Place this bundle of selling points inside an AI platform’s cost structure, and the paradox surfaces:

• Acquisition view: no ads and no IAP is the sharpest acquisition hook right now. Casual-game users’ tolerance for ads and IAP is already low, and a “pure play” posture builds goodwill fast and lowers the trial barrier;
• Cost view: AI-generated games carry real inference costs — more users, more generation, longer play sessions mean bigger compute bills, and “unlimited play time” means the cost exposure has no cap;
• Monetization view: the company has disclosed no follow-up monetization arrangement — whether IAP, ads, or revenue shares will be introduced later is undisclosed. What subsidizes the experience today, and what recoups the cost tomorrow — neither question has an answer.

The author of the gaming-industry briefing has an industry reading of this that I think is worth relaying (note: this is that author’s opinion, not fact): Kaiying’s move is seen as shifting from “lowering costs for developers” to “handing tools to ordinary people,” aiming at the distribution gateway — “video as advertising, advertising as game.” If that judgment holds, “no ads, no IAP” is not selling paid conversion but user time: migrate the attention habit of video-scrolling onto the game flow first, and monetization can be deferred to the distribution stage. Of course, all of this remains an open question for now — monetization details are undisclosed, and the long-term revenue path is an open problem, not established fact.

4.3 Whether the Parent Can Bear It#

Judging how long this model can last requires looking at Kaiying Network behind it. Per same-day disclosed fundamentals (single source: the gaming-industry briefing): ticker 002517.SZ, H1 2026 revenue of RMB 4.797 billion, up 86.06% year over year. At that scale, sustaining an AI platform that does not directly earn money early on is not hard; the real test is not money but ecosystem — which is the next section’s topic.

5. 3,000+ Games from the Second Beta: the Scissors Gap Between Quantity and Quality#

5.1 The Data and Its Caveats#

Data first, weight second:

Put those four rows together and the conclusion is clear: the 3,000+ figure has reference value, but it is single-source, and not one user-side official number has been given. Use it to sense the order of magnitude, not to draw conclusions.

5.2 The Scissors Gap Is Every UGC Platform’s Destined Problem#

What does 3,000+ games mean? That was still the closed beta, aimed at the advanced creation tool SOON; once SoonPlay lowers the creation starting point to “upload a video,” supply will almost certainly only grow. The question follows: quantity is up — where is the quality median?

This is not SoonPlay’s problem alone; it is the destined problem of every platform that pushes generation cost toward zero. Yesterday’s Steam Next Fest observation already showed the isomorphic scene: AI-content demos flooding the storefront and players petitioning for a “no AI” filter. SoonPlay’s casual-game form is lighter than storefront demos, so the “quantity-vs-quality scissors gap” should, in theory, only be steeper.

5.3 A Ceiling That Arrives Earlier#

The gaming-industry briefing’s stated risk is more direct: content moderation and homogenization will become the ceiling earlier. Both cut especially hard at SoonPlay, for two reasons:

1. Double moderation pressure on input and output: the input to video-to-game is a video — which itself requires content moderation; the generated game is a second layer of content, also requiring moderation. Double the supply, double the moderation bandwidth cost;
2. The casual-game form amplifies homogenization: casual games are shallow and short-lived; once one viral video is mass-derivative into countless variants, the platform’s content diversity dilutes rapidly, and the “watching” experience collapses before the “playing” one.

For “the TikTok of games” to hold, recommendation-and-distribution is the heart — and on the recommendation mechanism precisely, the company has disclosed nothing. Whether the scissors gap can be caught by algorithms and moderation is a more essential question than “can it generate games.”

6. Same-Track Competition: Kaiying’s Position in the China-vs-World Picture#

6.1 Competitor List (Mind the Source Caveat)#

Yaotu Network’s report placed SoonPlay in a list of domestic and overseas same-track products; I reproduce it as-is. The list is single-source; this article does not expand on or evaluate each product’s specific progress:

6.2 Where Kaiying Stands#

Against this table, Kaiying’s differentiation is not “can also generate games” — nobody on this track lacks generation ability — but the combination of three things:

1. Full-stack self-ownership: the game platform, the large model, and the dev tools are all in-house, with no dependence on external model vendors;
2. Compliance first: the Jiyi SOON large model’s generative AI filing is already on the books, giving it a head start on beta sequencing;
3. Gateway positioning: building the consumer-side “video → game” entry point directly, rather than building developer tools first and trickling down later.

Google Playground takes the “ecosystem launch” route — laying down creation tools and professional capability first, pulling in Unity for professional depth; SoonPlay takes the “single-product open beta” route, aimed straight at mass-market consumption scenarios, trading a three-scenario loop for user time. Which route closes the loop first depends on whether distribution or monetization snaps shut first — which is, fittingly, the shared unknown for every player domestic and overseas.

7. Wubian’s Closing: the Temptation of the Gateway and the Homework of the Ecosystem#

Time to gather the threads.

The most interesting thing about SoonPlay is not “AI can generate games again,” but the cut it chose: setting the creation starting point at a video, the consumption scenario inside a video feed, and business model step one at “we won’t take your money.” Put those three together and the ambition points at the distribution gateway — whoever lets users switch frictionlessly between watching and playing holds the door handle of the next content platform. Kaiying Network’s half-year revenue base of RMB 4.797 billion (single-source figure) can also afford this burn-for-the-gateway window.

But as someone who has long watched the long-term operations of mobile and online games, I must put it coolly: the gateway’s temptation is large, and the ecosystem’s homework is heavier. How recommendation distribution works, what keeps creators around, how moderation keeps pace with supply, and where money comes from after no-ads-no-IAP — on all four questions, the company has given no answers today. The 3,000+ games from the second beta prove generation-side capability; the platform’s final quality will be judged by retention-side data and ecosystem-side governance — that is the evidence on which the name “TikTok of games” stands or falls.

For ordinary players, the best posture right now may be simple: download it from the Jiyi Artificial Intelligence website and play, upload a video you shot yourself and try generating, claim the newcomer pack and glance at what signals hide in the launch perks — then come back with those four questions and watch its next move. A platform’s story is never finished in a launch event; it is finished by the people who stay.

— Wubian, NTGame Editorial