AI game creation has moved past the novelty of asking a model to produce a character or a simple playable demo. The harder problem is everything that comes after the first successful generation: organizing assets, maintaining game logic, balancing systems, iterating on scenes, testing the experience, and eventually turning a prototype into something players actually want.
That is the problem JIYI is trying to attack with SOON.
Kingnet describes Hangzhou Jiyi Artificial Intelligence Technology as a company built around an “AI + Game” strategy. Its independently developed SOON platform is positioned as a one-stop AI game creation environment that combines AI technology with gaming-industry workflows. Kingnet’s corporate history says its JIYI subsidiary launched SOON as its self-developed AI game creation engine in 2025. Kingnet
The interesting part for indie developers is not simply that SOON can generate game content. It is that the product direction points toward a broader category: software that manages the production system around AI generation.
The real bottleneck is not generation
A generative model can create an image in seconds. It can write code in seconds. It can describe a level, invent characters, or propose a game mechanic.
But a game is not a collection of disconnected generations.
A usable project needs consistency.
The same character needs to look and behave consistently across scenes. Items need relationships with progression. Quests need to connect to actual gameplay. Numbers need to stay balanced. Generated code needs to remain compatible with the rest of the project.
This is why AI game creation has a workflow problem.
JIYI’s SOON is interesting because its public positioning covers more than one generation task. The company describes it as a one-stop platform, while reporting on the platform shows interfaces for managing game assets and production structures rather than only prompting a chatbot. Game Daily
For an indie founder, that distinction matters.
A tool that generates a beautiful demo can create excitement. A tool that keeps the project coherent can create retention.
SOON’s bigger idea: AI as production infrastructure
The traditional game-development workflow separates tools.
One application handles art. Another handles animation. Another handles level design. Code lives somewhere else. Designers use spreadsheets for game numbers. Producers use project-management software to coordinate everything.
AI makes generation cheaper, but that fragmentation can become worse.
If every AI capability creates another isolated output, the developer becomes the integration layer.
SOON’s direction suggests a different approach: bring more of those production activities into one environment.
That can be valuable for small teams because indie studios have the opposite problem from large publishers. They rarely have enough specialists.
A two-person team might need an artist, technical artist, designer, programmer, producer, QA tester, and marketer. AI can help with parts of each job, but only if the tools reduce coordination rather than create more files to manage.
The product opportunity is therefore not “AI makes games.”
It is “AI reduces the number of specialist handoffs required to make a game.”
Why this is especially relevant to indie developers
A solo developer usually has a limited iteration budget.
Suppose the first playable prototype takes two weeks. If changing the combat system requires rebuilding several assets, rewriting code, adjusting numbers, and manually checking every level, the cost of experimentation remains high.
Now imagine the production environment understands those relationships.
A developer changes a weapon.
The system identifies affected stats.
The AI proposes updated values.
Related effects and UI elements are surfaced.
The developer tests the new version.
That is much closer to an AI-native production workflow than a collection of content generators.
Earlier JIYI research with 01.AI also described a goal of connecting game art, animation, code, and numerical data through a vertical game model and SOON engine. The project was designed around game-specific consistency rather than treating the game as a generic software-generation task. AIBase
The lesson: build around the artifact, not the model
There is a useful product lesson here for indie founders building AI software.
Many AI startups begin with the model.
They ask:
“What can the model generate?”
A stronger question is:
“What valuable artifact does the customer need to maintain?”
For SOON, the artifact is the game.
That changes the product architecture.
Instead of generating a character and stopping, the system can care about where the character belongs.
Instead of generating code and stopping, it can care about whether the code is connected to the game.
Instead of producing an image and stopping, it can care about how the asset is used.
This principle applies outside gaming.
For a legal product, the artifact might be a case file.
For a sales product, it might be an account.
For a SaaS product, it might be a customer workspace.
For a developer product, it might be a repository.
The AI becomes more valuable when it operates inside the lifecycle of the artifact.
The hard part will be quality
There is also an important warning.
AI game creation does not automatically solve game design.
Public discussion around AI-generated game platforms shows a recurring problem: lowering the cost of making a game can increase the number of mediocre games even faster than it increases the number of excellent ones. A recent industry analysis noted that platforms such as SOON and other AI game-creation services are still experimenting with how to help creators move beyond low-quality output and build sustainable creator ecosystems. GameTeahouse
That means the winning platform cannot optimize only for generation speed.
It needs feedback loops.
Creators need to know:
- whether players finish the game;
- where players quit;
- which mechanics are confusing;
- which generated assets are actually useful;
- whether a game can be shipped outside the platform;
- and whether the economics work after model costs.
The platform that connects generation with those signals has a stronger chance of becoming infrastructure.
What an indie founder can copy
You do not need to build an AI game engine to use the same product strategy.
Take one workflow you already have.
Then map the artifact through its lifecycle.
For a SaaS founder, that could be a feature:
- Customer request enters the system.
- AI turns it into a product requirement.
- AI proposes implementation tasks.
- Code changes are generated.
- Tests run.
- Documentation is updated.
- Release notes are created.
- Customer-facing pages are checked.
That is more defensible than a standalone “write my release notes” button.
The same principle applies to marketing, finance, customer support, and research.
The closer AI gets to the entire workflow, the harder it becomes to replace the product with a simple prompt.
Where SOON has to prove itself
SOON still faces the same challenge as every AI creation platform: quality must become repeatable.
Generation speed is easy to demonstrate in a launch video.
Long-term creator value is harder.
The platform needs creators who return, publish projects, improve them, and ideally make money from them. It also needs enough control for experienced developers while remaining accessible to people who do not know traditional game-development pipelines.
That balance is difficult.
Too much automation makes the tool feel shallow.
Too much manual control makes it look like another professional editor with an AI assistant.
The sweet spot is an environment where AI handles repetitive production work while the creator retains control over the decisions that define the game.
Bottom line
SOON is worth watching not because “AI can make games” is a new idea, but because the next generation of AI products may be built around persistent creative artifacts.
For indie founders, that is the useful takeaway.
Do not stop at generating something impressive.
Build the system that helps the customer maintain, iterate, test, ship, and improve the thing they are actually trying to create.
That is where AI generation starts becoming a product instead of a demo.