AI-generated concept image showing the construction process of a game scene from wireframe to solid form
When making an action game, it takes only a few seconds for a designer to write down “you can counterattack after dodging.” It is not until the team actually playtests that they learn how much time is left after a dodge for the player to counterattack, whether enemy movements can be seen clearly, and what happens if the button is pressed twice in a row. The purpose of a prototype is to turn these arguments into questions that can be verified with a controller as early as possible.
On November 2, Moonshot AI released a game development and creative work solution. The solution combines the Kimi K3.1 series models, Kimi Code, Kimi Work, and professional tool plugins, and can connect to and operate projects in software such as Unity, Unreal Engine, and Blender. In the officially demonstrated Godot case, the system built a 2.5D action game prototype based on reference materials and gameplay requirements, including sword swings, dodges, and blocks, and used rendered images and playtest recordings to check the results of modifications.
Another demonstration, in cooperation with Unity China, pushed delivery to the engineering level. According to the publisher, the system ran continuously for about 20 hours in one demonstration without human intervention and delivered a native project that could be further developed. This demonstration recorded the result of one specific task. Engine versions, asset scale, and existing code structures used by different projects will change the difficulty of the work.
For the person responsible for project approval, a playable version is already enough to test a gameplay idea. For developers who take over the project, the first things they may need to look at could be another set of questions: whether the project can open in the team's environment, whether scenes and assets are easy to find, where the core rules are concentrated, and whether the next change will break existing functionality. Even if two teams receive the same prototype, they may reach different acceptance conclusions.
This is also what makes engine integration worth watching. Being able to enter a project and view the running results means AI has an opportunity to complete a round of modifications and verification on its own. Teams then need to decide which operations can be completed automatically in an experimental branch and which assets or settings must be recorded and reviewed by a person once changed.
The next valuable demonstration could start with an unfamiliar developer taking over the project. Have them modify a combat rule, rerun the prototype, and then hand it to a designer for playtesting. This relay process can show whether the team has actually received a demo worth watching or a project that has already entered the production pipeline.
Source · Kimi official announcement · IT Home report