From Graphics to the Engine: Li Chi and His Meowa Experiment

Starting in the second half of 2024, AI’s prominence in the gaming industry surged.

Google DeepMind released Genie 2, which generates interactive 3D environments from a single image; in the video demonstration, players can walk, jump, and even interact with objects in the AI-generated world.Decart’s Oasis went a step further, releasing a real-time, Minecraft-style open world in October 2024—touted as the first fully AI-driven playable game.By 2025, products like Gambo had further lowered the barrier to game development: by entering a text description, users could obtain a working game prototype within minutes.

The “one-line game creation” trend is rapidly gaining momentum in the gaming industry.

At the same time, changes were also taking place on the other end of the game development spectrum. In February 2025, Andrej Karpathy introduced the concept of “Vibe Coding”—describing requirements in natural language and letting AI write the code. Tools like Claude Code, Codex, and Cursor rapidly matured within a year, and coding ability was no longer a barrier to game development.By mid-2026, more than 60 percent of Vibe Coding users were not professional developers. This meant that a large number of people who had previously been barred from game development by the coding barrier now had the opportunity to enter the field for the first time.

Investment from major companies is also accelerating. Mihoyo has publicly announced that it will invest 100 billion over three years to “go all-in” on large language models. According to industry rumors, nearly every leading game company has projects underway for AI-native games or AI NPCs. Both money and talent are pouring into this field.

But if you’re actually developing a game—not just creating a demo or technical showcase, but building a project that will be released to players and require long-term maintenance—you’ll encounter a more challenging problem: the gap between a vague gameplay idea and a playable prototype isn’t just about code.

The world of Genie 2 can only be sustained for ten to twenty seconds; the visuals in Oasis are “imagined” by AI on a frame-by-frame basis, so there are no exportable asset files; Vibe Coding has solved the coding aspect, but for game development, elements such as level design, art, and sound effects—these aspects have not yet truly been integrated into the development process by AI.

Art assets, in particular, are the biggest headache. General-purpose image-generation tools can produce a visually appealing concept art, but there’s a huge gap between a single concept art and a set of game assets ready to be imported into Unity or Godot—blurry edges, stylistic inconsistencies, incorrect dimensions, and a host of issues after slicing.For indie developers, coding can be done through “vibe coding,” and game designers can write their own scripts, but art assets are the biggest bottleneck. Setting aside the fact that “outsourcing” is a slow process, the high cost alone deters many developers.

When Li Chi left ByteDance AI Lab in 2023 to start his own business, he faced this same problem. His team initially wanted to develop AI-native games, but along the way, they discovered that there were no tools on the market capable of supporting the entire creative process—from game design to asset creation.So they decided to build their own internal toolset, starting with pixel-level art—the most challenging aspect—because pixel-level art has the strictest standards. If the AI could achieve “perfect pixels, consistent style, and engine-ready assets,” it would prove that the AI-generated content was truly “Game Ready.”

Later, this internal tool was spotted by an external competitor, who immediately expressed a willingness to pay to use it.

This is how Meowa began—not as a grand product plan born out of market research, but as the result of an AI game development team turning its own creative toolchain into a commercial product during the production process.

The following is a transcript of the conversation between Tea House Jun and Li Chi. (Some content has been edited to enhance readability.)

01

From Byte to Miao Jito

Teahouse Guy: ByteDance has had its fair share of setbacks in the gaming sector. Why did you decide to leave a major tech company to develop AI games?

Li Chi: I’m determined to make games for the rest of my life. So for me, the question has never been “whether to leave ByteDance,” but rather “how to keep doing this.”

I’ve developed some AI games and achieved some success with them.

At ByteDance AI Lab, I spent several years conducting research on game AI—I worked on the reinforcement learning training and deployment of game agents for projects with hundreds of millions of daily active users (DAU), and I also led the 3D Generative Agent project (which recreated Stanford’s AI Town within ByteDance). Outside of work, I’ve created numerous AI game demos and won awards multiple times at internal hackathons.This experience has shown me two sides of the coin: on one hand, how burdensome the industrialized production lines at large tech companies can be; on the other, just how awe-inspiring the capabilities emerging from cutting-edge AI are. I am firmly convinced that AI will inevitably disrupt the gaming industry—and I don’t want to keep working the old way.

ByteDance is an excellent company, but it has always been a bit indecisive when it comes to gaming. I left because I believe we are at a turning point: advances in AI are breaking down the talent barriers that major tech companies have built up over the years, and the gaming industry is about to bid farewell to the capital-intensive era of relying on massive workforces.At this juncture, AI will greatly amplify the agility of startups, allowing them to bypass the layers of bureaucracy at major companies and focus all their resources on what truly matters. This is a unique window of opportunity available only to startups in the AI era.

Teahouse Guy: How did Meowa first come about? Was there a pivotal moment when you realized it could become a standalone product?

Li Chi: If I had to pick one specific moment—there was a time when I demonstrated our internal pipeline to a friend who’s also working on AI games. After seeing it, he immediately said, “Can I use this? I’m willing to pay for it.” In that moment, I realized that Meowa might be more than just a tool for us.

Meowa started out as an in-house tool we developed to solve our own problems while working on *Meow Island*. At the time, we had tried nearly every AI tool on the market, and they all had the same issues—they looked too “AI-generated,” the pixels were blurry, the styles didn’t match, and the assets couldn’t be imported into the engine.Those tools focused on producing “a pretty image,” but game development requires “a set of standardized, consistent assets that can be integrated into the pipeline”—these are two completely different things. So we decided to build our own system specifically designed to generate pixel-level game assets and frame-by-frame animations.

Early meowa screenshots

That friend’s reaction was not an isolated case—later, we showed it to several other people in the industry, and the feedback was consistent: they had encountered the same issue. Combined with the fact that agent orchestration capabilities were beginning to be integrated, the tool’s potential far exceeded our initial expectations. That’s what ultimately led us to decide to turn it into a product.

So, for us, tools and games aren’t separate paths—they’re two branches of the same tree. Games help us identify the industry’s pain points, and tools enable us to deliver solutions to the industry more quickly.

Teahouse Guy: Could you tell us a bit about the Miao Jito team?

Li Chi: We currently have a team of six, and we’ve intentionally structured it so that half the team focuses on AI and half on games—only a team that truly understands both areas can create AI-native games and tools.

Less, our CTO, and I lead the model and technology side of the business. I previously worked as a game AI researcher at ByteDance AI Lab, where I led the reinforcement learning training and deployment of game agents for projects with hundreds of millions of daily active users (DAU) and managed the 3D Generative Agent project.Less previously worked at Alibaba AI Lab, where he was responsible for large-scale model and agent architectures. He was the lead creator of Alibaba’s open-source framework TinyNN and later served as the algorithm lead for the Multi-Agent Coding product at Ant Group.

On the product and game side, I oversaw the production of *Meow Island* and guided the product direction for Meowa.Lead Designer Bofan is a graduate of the Game Design Department at Communication University of China; his indie game *Round and Roll Farm* was signed by a top publisher. Lead Artist Jean previously served as Art Director for *Driven into the Void* and *Good Pizza, Great Pizza*, and has a strong background in both anime-style art and hit indie games.

I am directly in charge of the market at this stage. At this point, the founder is the most effective go-to-market (GTM) strategy—our early users were essentially convinced by the quality of the demos generated within the developer community, and the conversion path was very short.

02

Starting with Fine Arts

Teahouse Guy: Who is Meowa’s core user base right now?

Li Chi: The most important group is independent game developers. We currently have a community of over 10,000 developers who interact in our group chats every day to discuss their needs, report bugs, and share their progress—whether it’s pixel RPGs, farm simulators, horror games, or cultivation games, there are people working on every genre, and new titles pop up almost every day.

In terms of user demographics: the core consists of independent developers, but the community also includes many game design students, small teams of a few people, and game industry professionals working at major studios—most of whom use Meowa to work on side projects they’re truly passionate about in their spare time.In other words, Meowa’s users aren’t categorized by “professional status,” but rather by “creative autonomy”—as long as someone wants to make a game for themselves, they have a place in our community.

Why start with independent developers? Because they lack the resources of large companies to throw manpower at a project, yet they have full control over their own work—making them the earliest genuine users of AI game development tools.

Teahouse: Before using Meowa, how did these users typically handle pixel art issues? What are the real reasons they’re willing to pay?

Li Chi: Users are willing to pay for Meowa—on the surface, to save money and time, but fundamentally because they’re gaining access at an extremely low cost to something that was previously hard to come by: customized, high-quality, stylistically consistent art assets that can be integrated directly into the game development process.

In the traditional workflow, independent developers typically have only a few options for creating pixel art. Outsourcing allows for customization, but it’s slow, expensive, and involves high communication costs—creating a single set of animations for a character often takes anywhere from a few days to several weeks.Buying asset packs is cheap and immediate, but it’s difficult to match styles; characters, scenes, and UI often come from different creators, resulting in a patchwork look when combined, and generic asset packs lack uniqueness. Drawing the art yourself is suitable for the few developers with artistic skills, but for most programming- or design-oriented creators, art is the biggest bottleneck.Many others are deterred by the artistic barrier before they even begin developing a game.

So when users buy Meowa, it’s not just for the low price or speed. What they’re really buying is the ability to transform an art pipeline that was once unmanageable and unaffordable into a low-cost, iterative, and customizable creative workflow.

Teahouse: Many AI image-generation tools can produce attractive images, but what game developers need are “assets that can be used in a project.” How do you define when an asset meets the criteria for use?

Li Chi: We don’t use “good-looking” as a criterion for determining whether game assets are usable. For game developers, the real standard is whether an asset can be incorporated into a game project at low cost and with minimal effort.

Internally, we have several tiers of evaluation criteria for Game Ready assets.

First is the artistic standard. Assets must look like they belong in a full-fledged game, not like an isolated AI-generated image.They must feature a consistent aesthetic, clear outlines, restrained detail, clean color blocks, accurate perspective, and a consistent lighting and shadow palette. A particularly important aspect of pixel art assets is “perfect pixels”: edges must not be blurred, there must be no extraneous color noise, and there must be no dirty noise or semi-transparent transitions commonly found in AI-generated images.

Second are engineering standards.For game assets to be used in Unity, Godot, or other engines, they must have a clean, transparent background, clear dimensional specifications, appropriate white space, consistent grid alignment, and a format that allows for direct slicing, importing, and rendering. If users have to spend a significant amount of time cutting out backgrounds, trimming edges, or rearranging elements after receiving the assets, then they are not truly usable.

Third is the standard of consistency. Game development requires not just a single image, but an entire asset system. A character, a set of props, and a set of scene tiles must all belong to the same universe, share the same scale, and adhere to the same visual language.Many AI tools can generate a single high-quality image, but when ten images are placed together, there’s a stylistic drift. Meowa focuses more on consistently generating a complete set of assets with a unified style.

Specifically, for character animations, we check whether consistency is maintained from frame to frame—whether the body shape, clothing, weapons, and color palette remain stable; whether the movements have clear start, middle, and end phases; whether the anchor points are stable; and whether the animation can be imported directly into the engine as a sprite sheet.For tilesets, we check whether they can be seamlessly tiled together, whether the transitions at the edges are natural, whether they adhere to tileset conventions, and whether they appear consistent and legible when laid out to form a complete map.

Therefore, our definition of “usable” is: visually appealing enough to be marketable, technically ready for direct integration into the engine, and capable of being continuously expanded into a complete set of game assets with a unified style.

Teahouse: What’s the biggest difference between pixel art and regular AI-generated images? Why did you choose to focus on pixel art first?

Li Chi: The biggest difference is this: While general AI image generation aims for “aesthetically pleasing” results, pixel art demands that “every single pixel be correct.”

Regular images are continuous, so users may not notice if the edges are slightly blurry. However, pixel art is a highly structured and engineered art form, requiring extreme precision in outlines, edges, color blocks, color palettes, animation frames, and tile-matching relationships.A pixel art image will immediately appear unprofessional if its edges show anti-aliasing, if there is noise in certain areas, or if the colors are not restrained.

Why start with pixel art? First, pixel art has clear commercial value in the indie game market; many highly acclaimed indie games feature a pixel art style.Second, pixel art is a genuine pain point for a large number of indie developers—characters, environments, tile maps, UI, and animations all require consistent, long-term production, and traditional workflows are either expensive or time-consuming. Third, the standards for pixel art are very clear, making it an excellent way to test whether an AI game asset platform is truly “Game Ready.”

What were the challenges in the past? Large models are inherently better at generating continuous images than at understanding strict pixel grids. They are prone to blurry edges, semi-transparent transitions, and noisy artifacts.Maintaining stylistic consistency is also difficult—style drift occurs when ten images are placed together. Pixel animation is even more challenging—sprite sheets require stable body anchors and continuous motion logic; simply stitching together a few similar images does not constitute animation.

Teahouse: What is the key difference between Meowa’s technical approach and that of general-purpose image-generation models?

Li Chi: Meowa isn’t just a generic image generation model with a pixel art filter applied. Our approach combines a domain-specific model, high-quality game asset data, an understanding of asset specifications, and the Agent Harness workflow.

General-purpose models generate images that “look like pixel art,” but they don’t truly understand the requirements of game development—whether pixel edges are aligned, whether animation frames in a sprite sheet are consistent, whether tiles can be seamlessly tiled in all four directions, and whether the transparent background is clean. Meowa addresses a different need: rather than generating a single image, it generates a usable game asset.

The role of Agent Harness here is as follows: It does more than simply feed the user’s prompt to the model; it understands the type of assets the user wants, their intended use, and the development context, automatically breaks down the requirements, selects the appropriate generation workflow, ensures stylistic consistency, and reviews and refines the results.For example, if a user requests a character sprite sheet, the system needs to understand that the character must have a consistent body type and color palette, that there should be no drift between different animation frames, and that the final output must be exported in a format compatible with the game engine.

So the key difference is this: general-purpose models address the question of “Can it be rendered?”, while Meowa addresses the question of “Can it actually be used in a game?”

03

Going Beyond Fine Arts

Teahouse: Do Meowa users currently view the platform more as a way to experiment with AI-generated images, or do they see it as a real production tool? How do you determine whether a user is seriously creating game assets?

Li Chi: In the early stages, any AIGC tool will attract a group of early adopters. These users are more likely to approach the tool with a “let’s see what the AI can generate” mindset; they’ll randomly experiment with various styles and themes, generate a few individual images, and then move on after viewing them—they typically don’t become consistent paying customers.

Game assets created by users using our tools

However, based on Meowa’s current data and community feedback, we’ve already seen a group of users who explicitly view it as a genuine production tool. The most telling metric is that our current paid user rate has exceeded 11 percent. For an early-stage AIGC product, this indicates that users aren’t just satisfying their curiosity—they’re willing to pay for actual production value.

When determining whether a user is seriously creating game assets, we don’t just look at how many images they’ve generated; rather, we examine whether their behavior exhibits the characteristics of “project-based production.”

Users who come to play typically explore a wide range of options. They might generate a cyberpunk character today, a medieval monster tomorrow, and a Japanese-style avatar the day after. The styles, themes, sizes, and uses are all disjointed.

Users who are actually creating game assets are a completely different group. Their behavior clearly centers around a single project: first, they define a character’s design, then request that different animations for that character be generated; first, they create a set of UI icons, then continue to add elements like backpacks, skills, shops, and buttons; first, they create a set of scene tiles, then repeatedly request that the edges blend naturally, the tiles be tilable, and the style remain consistent.Their concern isn’t merely “Does this image look good?” but rather “Can I actually use this image in my game?”

Several clear signals: contextual continuity, where users consistently generate assets based on the same worldview, the same character, and the same art style;Modification requests have become increasingly specific, shifting from “make it look a little better” to “keep this character unchanged and generate a 4-frame walking animation” or “unify the icons using the same color palette”; export practices have become more engineering-oriented, with a focus on transparent backgrounds, size specifications, sprite sheets, and engine-compatible formats; and there has been repeated fine-tuning to ensure stylistic consistency.

Teahouse: There are already some AI pixel art tools available overseas, and major Chinese companies are also developing more comprehensive AI game development platforms. Compared to them, what do you think are Meowa’s strengths?

Li Chi: We have indeed systematically evaluated a large number of relevant products both overseas and domestically. Our assessment is that most products currently fall into one of two categories: one is a single-point asset tool—such as those for generating pixel-art characters, props, and backgrounds—and the other is a platform that stacks features, linking modules such as game design, coding, art, and testing through programmatic workflows.

Both types of products have value, but they share a common problem: they shift the complexity onto the user.

Users still need to understand how to write prompts, how to tune model parameters, how to maintain a consistent style across assets, how to integrate code, and how to organize workflows. For professional teams, this is a productivity tool; but for everyday creators, the barrier to entry remains high. The more features there are, the more likely it is to become something that “looks powerful but is hard to use.”

Meowa’s core strength lies in the fact that we are not simply creating a “collection of features,” but rather an Agent Harness that truly understands user intent. Its goal is to build an intelligent bridge between user expressions and the underlying complex systems.

Users don’t need to understand models, parameters, or the entire development process—they simply need to describe the assets, gameplay, or gaming experience they want. Agent will help them break down their requirements, select the right tools, generate content, and ensure a consistent style.

The first key difference lies in the barrier to entry. Meowa is designed to serve not only professional game developers, but also a vast number of ordinary users who lack development experience yet have a desire to express themselves and create. We believe that the true value of an AI game platform lies not merely in making professional teams 10 times faster, but in enabling people who previously couldn’t make games at all to create their own games for the very first time.

The second area of differentiation lies in our understanding of competitive advantages. As capabilities like Codex and Claude Code rapidly evolve, “writing code” in game development will become increasingly commoditized. In the future, the true distinction between AI-generated games will not simply lie in whether code can be generated, but in aesthetic sensibility and gameplay design capabilities—what we internally refer to as “Taste.”

Game development isn’t simply about piecing together features. Whether a character looks good, whether a scene has a cohesive style, whether gameplay is engaging, or whether the balance of game mechanics keeps players coming back—none of these issues can be solved by merely stacking up models. Meowa’s vision is to become the most tasteful AI game creation platform on the market.We aim to embed refined aesthetics, seamless interactions, and well-crafted gameplay design into the Agent’s workflow and decision-making system.

Teahouse: What valuable signals does Meowa gather every time a user generates, edits, retries, or exports content? Do these signals help Meowa better understand “game assets”?

Li Chi: Every time a user generates, selects, modifies, retries, or exports content on Meowa, they are essentially providing us with extremely valuable preference data. This isn’t just a simple usage log—it tells us what kinds of game assets users find “appropriate,” “attractive,” and “usable.”

We will focus on several types of signals.

Selecting a design. Meowa typically generates multiple candidate results at once, and the fact that users ultimately choose one and discard the others serves as very strong feedback on their preferences. This tells us which styles, colors, contours, levels of detail, and pixel rendering techniques users prefer for a given design requirement.

Modification requests. Users constantly submit requests for changes, such as “make the outline cleaner,” “make the style more consistent,” “make the movements more exaggerated,” “change it to a 4-frame walking animation,” and “keep this character and regenerate attack animations.”These modification instructions are crucial because they directly expose the shortcomings of the generic model in game asset generation and help us retroactively optimize the Agent Harness’s understanding and execution capabilities.

Retry signal. If a user repeatedly generates, fine-tunes, and modifies assets of the same type, it indicates that the current results do not fully meet their needs. We can analyze the cause of the failure: is it inconsistent art style, unclean pixels, poor character consistency, disjointed movements, or an inability to tile the images?

Exporting assets. Exporting is a very strong form of positive feedback.If a user is willing to export assets, it indicates that these assets aren’t just “visually appealing,” but are already close to being ready for use in game projects. This is especially true when users export sprite sheets, tilemaps, icon sets, transparent PNGs, or engine-compatible formats—it shows they aren’t just messing around, but are actually producing assets.

These actions create a continuous data flywheel: users generate assets, the system records their choices and modifications, and Meowa labels, analyzes, and organizes these signals, which are then used to optimize the underlying models, generation workflows, quality checkers, and Agent Harness.In the long term, what we’re accumulating isn’t just ordinary image data, but preference data and quality standard data from the game asset production process.

Compared to general-purpose image-generation models, can Meowa gain a better understanding of game assets by leveraging user data? The answer is yes. And this is precisely the core of our strategy: General-purpose models understand “images,” but they don’t understand “game assets”—these are two distinct things.

In earlier versions, users had to go through many rounds of prompts to ensure a consistent art style across different assets; now, after just a few rounds of natural language interaction, they can finalize a unified art style for an entire game. In earlier versions, it took more than a dozen iterations to ensure the fluidity of a walking animation; now, through natural language interaction, it can be finalized in just 2–3 rounds.

Behind it all lies the same mechanism: the data flywheel.Every time a user on Meowa “chooses this and rejects that,” they are tagging the model with preference labels—we use this real-world data to continuously train the model using RLHF. General-purpose image generation models don’t have access to this data, so they will always be better at understanding “a visually appealing image,” while Meowa will become increasingly adept at understanding “a set of game assets that are ready to use.”

Teahouse Guy: There’s a common view in the industry that, in the gaming sector, AI is currently “deeply integrated at the tool level but only superficially integrated at the experience level.” Does Miao Jito’s experience happen to validate this view?

Li Chi: You’re half right. It’s true that “penetration is deep at the tool layer,” but “penetration is shallow at the experience layer”—that’s a result, not a measure of investment.

The experience layer has never lacked funding or talent.Mihoyo officially announced a three-year, 100 billion investment to go “all in” on large language models, and Liu Wei has publicly stated his goal of creating a “personalized for every player” gaming experience; as far as I know, nearly every major game studio has AI-native games or AI NPC projects running internally. The investment is definitely there—but the results haven’t materialized yet, and that’s the real issue.

The mechanism behind this is as follows: The tool layer has clear paths and predictable returns, so it spreads quickly; the experience layer, on the other hand, offers a much broader scope for exploration, but the paths have not yet been validated, so the vast majority of attempts are superficial, and no one has truly gone all the way. This is not a matter of “insufficient investment,” but rather a matter of “insufficient depth of exploration.”

For AI games to truly break into the mainstream, the industry doesn’t need more funding—it needs more people who are willing to see this through to the end.

Teahouse Guy: What was the biggest uncertainty you faced when developing AI-native games in the past? What types of AI games do players tend to accept more readily?

Li Chi: The biggest uncertainty in developing AI-native games is that this is completely uncharted territory—no one knows exactly what an AI-native game should look like. Both major companies and startups have tried their hand at it, but very few have sent a clear enough signal to the industry that “this is the direction to take.”We’ve seen quite a few companies choose to give up or pivot after initial exploration. To date, the AI gaming sector has yet to experience its own “IceFrog moment”—a prototype with the potential to become a phenomenon.

But precisely because no one has done it before, every direction is worth exploring. We believe AI will definitely bring players a unique and engaging experience, and that some form of AI gameplay will become the next DOTA.We don’t expect Meow Island to be that DOTA—but if it can become that AOS, serve as an early prototype, and allow those who come after us to build upon our work to create even better AI games, thereby pushing the industry forward just a little bit, we’ll be more than satisfied.

After experiments like *Meow Island* and *Cute Paws Brawl*, we’ve seen that what truly hooks players is free interaction, an emergent sandbox experience, and the feeling that “I can have a real, lasting impact on this world.”What players want isn’t to chat with AI or battle against it—it’s to make a difference in the world through their own actions.

That’s why Meow Island has moved toward a more in-depth user-generated content (UGC) experience. We’ve developed a set of technologies that use AI to generate assets in real time, allowing players to use AI to create their own pets, buildings, and even entire islands. Everyone’s Meow Island is one-of-a-kind—there’s no other like it. This sense of “the world belongs to me” is the fertile ground from which emotional connections truly grow.

Teahouse Guy: If Meowa continues to grow, will you develop it into a single-pixel art tool, or expand it into a complete AI game development toolchain? What role will AI-native games like *Meow Island* play in Meowjito’s future?

Li Chi: The art features you see now are just the tip of the iceberg when it comes to Meowa. From the very beginning, Meowa was positioned not as a standalone tool, but as an AI-powered game creation platform.Currently, we have opened three production pipelines: design, art, and music/sound effects. The Agent design tool helps creators turn vague ideas into developable Game Design Documents (GDDs); the art pipeline covers pixel art, high-definition graphics, UI, animation, maps, Spine, and more; and the music/sound effects pipeline includes ambient music, combat music, and various game sound effects.

Pixel art is probably what people see the most right now, because it was the first proven entry point—only after demonstrating that AI-generated content can truly be “game-ready” do the subsequent capabilities become meaningful.

Meowa.ai Product Screenshots

Next, we are developing an end-to-end game production chain and testing pipeline to make the entire process more seamless. However, we will proceed cautiously with the rollout of these capabilities; we will not sacrifice the quality of any individual step in the pursuit of completeness.

As for *Meow Island*, it is the first AI-native game to emerge from this platform and embodies everything I’ve learned about AI gaming over the past three years. It serves not only as the best demonstration of our platform’s capabilities but also as our introductory gift to the AI-native gaming industry.We want to use it to prove one thing: AI is not limited to being a supporting tool; it can fundamentally transform the way games are produced and experienced—and we hope it will carve out its own place in the history of AI gaming.

Conclusion

AI gaming has yet to have its own “IceFrog moment.” From proof-of-concept to a full-fledged creative platform, and from tool capabilities to the player experience, there are still many steps that need to be worked out along the way.

But I believe Meowa will be part of the ultimate answer.