This mobile game has been made using AI for a year now – and no one noticed

This article is part of AI Week.

For about a year now, all of the content for the mobile game Sunrise Village has been generated by AI. But far from generating a backlash from players, it seems that no one has realised that anything has changed.

“Player numbers and engagement have remained stable,” says Thomas Lehr, director of engineering at the game’s developer, InnoGames. “The results speak for themselves: zero community complaints about AI or perceived quality differences. It seems players simply haven’t noticed any change.”

But now that the AI cat is out of the bag, is InnoGames braced for a revolt? Will players view the game less favourably than before? “Honestly, some players might have mixed feelings about it – that’s understandable,” muses Lehr. “But what players really care about is a good experience: fun gameplay, engaging stories, polished content. Our playtesting makes sure we deliver that.”

“The alternative was shutting down the game. Given that choice, I think most players would rather keep playing – and a year of stable engagement shows the content works.”

The harsh economic world of mobile free-to-play titles – which require constant content generation and vast marketing spend – means that typically, only the biggest hits survive. Supercell’s Squadbusters is a good example: even with around 58.7 million downloads and approximately $66.2 million in revenue to date (according to AppMagic figures), it was still deemed enough of a failure for development to be halted just 18 months after launch.

Sunrise Village, a cosy farming game first released in 2021, hasn’t exactly been a complete miss for InnoGames: in 2025, it brought in total revenue of between around $160,000 and $190,000 a month, based on AppMagic estimates. That might sound like a lot, but bear in mind that this is before things like the 30% store cut, taxes, and marketing costs are deducted, let alone the salaries of the game’s developers. Furthermore, the game’s monthly active user count peaked a couple of years ago, and its revenue has roughly halved since then.

In short, says Lehr, Sunrise Village was “underperforming financially,” and had reached a point where InnoGames couldn’t justify keeping a team of roughly 25 people working on it full time. “Traditionally, the answer would be to shut down unprofitable games,” he says, “but that’s becoming increasingly unpopular with players, as the Stop Killing Games initiative shows.”

AI offered an alternative. By generating content using AI, the Sunrise Village team has gone from 25 to between two and four people, with the previous developers being moved onto a new project. “Without AI, the game would have been cancelled,” says Lehr. “With AI, we can keep it running profitably with a minimal team while maintaining the high quality of content that players expect.”

“That said, we still think the human touch matters also for a game with heavy AI involvement like in this case. We don’t just take whatever the AI produces – we guide it, refine it, and own the final result. AI does the heavy lifting, humans still shape what players actually experience.”

A rapid revolution

Lehr joined InnoGames around 10 years ago, rising from technical project lead to director of engineering. He says he first got interested in AI through ChatGPT and GitHub Copilot around two and a half to three years ago, which he found “genuinely mind-blowing.”

“As an engineer, I didn’t expect that something as complex as coding could be ‘auto-completed’ so effectively. That was the moment I realised this technology would be disruptive.”

Since then, he’s been driving AI initiatives across InnoGames. “We have central teams for art and engineering that serve as AI support, but we also encourage individual research from employees,” he says. “It’s very much a company-wide effort.”

The advancements he’s seen in AI coding over the past couple of years have been astonishingly rapid. “We’ve gone from basic autocomplete to what I call ‘agentic coding’: writing detailed specifications in natural language, then having AI translate that into executable code and tests. This is different from the misleading ‘vibe coding’ – a term that’s become popular for describing approaches where you accept AI output without fully understanding or controlling it. Fast results, but rather random and often unmaintainable.”

Lehr says that on smaller projects, coding agents now outperform humans in both speed and quality. “Tools that used to take days or even weeks are now done in hours.” Humans remain a crucial part of the equation though, as someone still needs to review the code, steer the AI, and make sure the output matches the expectations.

Upfront investment

Getting the AI functionality up and running wasn’t a trivial process. “The initial development took about two months of dedicated engineering effort,” says Lehr, which involved one or two engineers. A key insight was that they were able to use AI to actually create the content-creation tools, which would then use AI to generate content.

The biggest challenges now are the large, legacy codebases, says Lehr, which need to be made explicit and retrievable by the AI. “This is not a solved problem – it requires significant effort and experimental approaches to make codebases ‘AI-friendly’. That’s our current goal: workflows that let agents take a task from the virtual board, plan it, and deliver an implementation for the engineer to review. The review itself can also be AI-assisted.”

Content for Sunrise Village is now created using what InnoGames calls the “AI Stage Designer” – a custom toolset integrated into Unity that uses LLMs to handle story and level creation. It involves six production steps:

1. Story and quests. Natural language narratives are converted into a storyline and structured quest data, with auto-generated triggers and validation.
2. Intelligent balancing. Using a knowledge base of roughly 100 game items across multiple tiers, the tool automatically calculates construction requirements, XP rewards, and difficulty scaling.
3. Asset discovery. The system visually analyses every asset and auto-generates searchable labels.
4. Layout and placement. Level structures, progression paths, and unlockable areas are generated.
5. Object spawning. Collectibles, treasures, and obstacles are placed around the levels.
6. Cutscenes. Cutscenes are generated that integrate with the story.

Lehr says the process is overseen by humans throughout, and every level gets full playtesting before release, with final tweaking done by humans to give the content an individual touch.

The AI process is based around OpenAI’s GPT-4o model, which Lehr notes is “already quite an outdated model by today’s standards, but it was a breakthrough in intelligence when we implemented the toolset. Newer models have improved significantly, but we’ve found that stability and predictability matter more than being cutting-edge – each model change requires extensive testing, as different versions can behave quite differently.”

“Across InnoGames more broadly, we provide access to the frontier models from Google, Anthropic, and OpenAI for all employees, and use different models depending on the task. Anthropic Claude is particularly strong for engineering tasks. Google Gemini is currently leading for visual content analysis. It’s project-dependent, based on which model performs best for specific use cases.”

“We also use various image and video creation tools, and AI plays a significant role in our localization workflows as well.”

InnoGames has found that one of the most efficient uses for AI is automatically labelling the many thousands of assets that get created for every game. “Sunrise Village has roughly 10,000 visual assets, and finding the right one for a particular scene was incredibly time-consuming,” says Lehr.

“Our AI labelling tool visually analyses every game asset and auto-generates searchable labels. What used to take hours of manual browsing now takes seconds. You can search for ‘cosy fireplace’ or ‘autumn tree’ and immediately find relevant assets.”

InnoGames isn’t the only games company to have discovered the power of AI for asset retrieval. During the Machine Learning Summit at GDC 2025, Lóránt Mikolás from Call of Duty developer Infinity Ward gave a talk about how the studio uses an AI-based multimodal search engine to find and retrieve 3D models, photogrammetry assets, animations, and more.

Lehr says InnoGames is currently in the process of rolling out the AI labelling tool to other games. “Besides story and quest creation, it’s been one of the most immediately impactful applications of AI in our Sunrise Village workflow.”

Replaced by AI

One of the biggest worries about the introduction of AI is that it will replace jobs. But Lehr insists that at InnoGames, the aim is to increase productivity rather than cutting roles.

“We’re expecting significant productivity gains that will enable us to pursue more projects simultaneously and deliver more of our roadmaps – and thus more value for players,” he says. “Growth-oriented, not cost-cutting.”

“The staff members who transitioned from Sunrise Village to our new project Cozy Coast were actually happy about it. They got to stop doing routine content creation tasks and focus on creative work for a new title. That outcome aligns perfectly with our overall AI strategy.”

But are those artists and designers who were taken off Sunrise Village really happy about it?

Lémuel Wuibout is a senior lead artist at InnoGames, and he led a team of around seven artists on Sunrise Village before being moved across to the new merge-two game Cozy Coast. He’s an industry veteran, with a career in games going back more than 30 years, including a stint at Crytek.

“All the art team was pretty excited about moving from Sunrise Village to this new project,” confirms Wuibout. “Most of the artists were working on Sunrise Village for maybe more than five years, some of us even seven years. So it’s always a bit refreshing when you move to a new project and you have to do everything from scratch, like define the art style and get the new characters.”

But how does he feel about the fact that the game he worked on for so long is now being created by AI? “I will say that I’m quite happy about it. And I also have quite some admiration for the developers, all the tools that they created, and everything that they did to make it work. And it seems to be a natural flow or a transition, I will say, because on Sunrise Village, we were already having the problem of how do we create more content faster for the players.”

“It’s legitimate to be a bit scared by how things are moving forward, because there’s still a lot of unknowns”

Lémuel Wuibout

Wuibout had already been using procedural generation to create things like textures and terrain, so for him, using generative AI to do something similar felt like “the next step.”

That said, he is concerned about the future. “At the current state, the way we use AI, I’m not feeling threatened or scared by it, but you never know what’s going to happen in two, five, or 10 years. So of course, it’s legitimate to be a bit scared by how things are moving forward, because there’s still a lot of unknowns and things that we don’t have any control over.”

Wuibout stresses that InnoGames has been upfront about its use of AI to enable more projects and create content more efficiently, without reducing headcount. “And so far, I think that’s actually enhancing all of the artists. The way we use AI is more like [as] a concept artist: rather than starting from a blank page, you can start with a couple of images generated by AI. Maybe you will photobash them until you get a base of the canvas that you can work on, and then be more efficient because you don’t have to draw everything. You can just polish, and retake, and adjust everything so that you have a nice image in the end.”

But is that really a more efficient process than just drawing an image from scratch exactly as intended? Wuibout says that at the beginning, the introduction of AI didn’t make the process any quicker. “But AI is making so much progress so fast that the more it goes, the more it’s actually time saving.”

It took a while to get here, though. “At the beginning, you need to learn, you need to understand the different types of models, what you can do, what you cannot. There was also how to prompt correctly, what to prompt. So there was a learning phase that was obviously not improving your efficiency that much, but then we started using Scenario AI, where we can train our own models: so we can use our internal library of assets, train the model with it, and then you get good results.”

Adapting to a completely different way of working didn’t come easily at first. “It was a bit tricky prompting and finding the right words,” remembers Wuibout. “And sometimes it could be a bit frustrating… Especially as artists, when we have something in mind and we want to express it, we are expressing it by drawing directly, making a sketch and not really prompting using words.”

All in all, however, Wuibout says it’s been “interesting” to learn a “new skill” along with others at InnoGames. “It’s something that we are sharing in the company: when you talk to other artists, [you say], ‘How did you manage to get to that result?'”

Interestingly, despite AI now being used at InnoGames to generate initial images, concept artists are still being used there. Wuibout says the concept artist for Sunrise Village moved onto a new project, “and I can say that even on Cozy Coast, I also hired a freelancer as a concept artist to help us.”

“It definitely made us more efficient”

Lémuel Wuibout

So what exactly is the role of a concept artist in an AI-dominated pipeline? “Let’s say that it’s a bit more specialized to a certain degree,” says Wuibout.

“The role of concept artist slightly changed. It evolved a bit into having a keen eye on the result, being able to understand what’s wrong, what needs to be fixed, and also look at the details. There’s always things that you need to fix with AI. Some shadows can be wrong, some lighting can be wrong, some elements, some proportions can be wrong. So you need to have an eye on this to try to understand what needs to be fixed.”

Wuibout emphasises that artists are needed to tweak AI-generated images before they’re used in the actual game, particularly for things like consistency with the art style – although he says the problem of inconsistency has lessened considerably since they started using Scenario and training the AI on InnoGames’ own images. He’s seen the technology progress enormously within a short space of time. “Recently, I will say that Gemini with Nano Banana is quite insane at understanding the style and providing something in a similar style.”

Crucially, he says the new workflow has dramatically sped things up. “It definitely made us more efficient,” he says. “And when I see how fast we were able to move on Cozy Coast compared to the time it took us on Sunrise Village to get to the same point, it clearly allowed us to do more.”

Pushback

Although Wuibout is sanguine and even positive about switching over to an AI-first workflow, it’s a safe bet that many, if not most, artists in the games industry will be far less agreeable to the prospect. In fact, in a recent GamesIndustry.biz survey of people working in the industry, 84.2% of respondents said that they thought no amount of AI-generated content was acceptable at any point the the development of a game, with only 3.2% agreeing that AI-generated art should be used for the minority of finished content, and just 0.9% agreeing that it should be used for the majority of content.

Lehr admits that InnoGames has experienced pushback against AI from its employees. “I think that’s natural and healthy,” he says. “Some team members had concerns about job security – that’s a legitimate worry when you see headlines about AI replacing workers. Others worried about losing their skills over time, becoming ‘prompt operators’ rather than craftspeople who truly understand their domain.”

“We addressed these through open communication. On job security, we were clear from the start: our strategy is to use productivity gains to take on more projects and create more value, not to reduce headcount.”

“On skill development, we’ve found AI can actually be a powerful learning accelerator. It’s a patient and knowledgeable peer that explains concepts from different perspectives, helps you explore adjacent skills, and lets you tackle challenges slightly beyond your current level. Used intentionally, it broadens expertise rather than replacing it.”

The future

There’s a risk, of course, in building your development pipeline on technology owned by someone else. It’s well known that many of the big AI companies are currently unprofitable – so what happens if a company like OpenAI dramatically increases its prices a few years down the line, once businesses have come to rely on its services?

“It’s a valid concern,” says Lehr. “Vendor dependency is a real risk. But we don’t see dramatic price increases in the foreseeable future, and we believe the risk is manageable.”

Lehr thinks that strong competition between providers will keep prices in check, and points to the roughly tenfold reduction in costs over the past two years as a result of algorithmic, hardware, and efficiency gains. He adds that open-source alternatives also tend to follow just 6–12 months behind frontier models, and notes that InnoGames takes a model-agnostic approach, so it isn’t “locked-in” to using a particular AI model. “If any provider dramatically increases prices, alternatives are around the corner,” he says.

Right now, the plan at InnoGames is to double down on AI. “Building on the Sunrise Village success, we’re expanding AI content generation across InnoGames,” says Lehr. “We’re already rolling out tools like our asset labelling system to other titles, and applying our learnings to games that are still actively growing – not just end-of-life titles.”

He says that InnoGames sees AI as a “productivity multiplier,” allowing game designers to generate more ideas, iterate faster, and validate concepts more quickly. He points to artists using AI to explore concepts more quickly, create mood boards, generate variations, and create rough drafts to refine, as well as engineers building better tooling for content creators.

“This isn’t just for struggling games,” he says, “it’s about enabling better outcomes: bigger feature sets, more content, faster iteration cycles, and ultimately more value for players.”

“Those who don’t adopt will find themselves at a severe competitive disadvantage”

Thomas Lehr

Given that, he thinks AI content generation will become widespread across the mobile sector. “The productivity gains are too significant to ignore,” Lehr concludes. “Companies that figure out how to effectively integrate AI into their workflows will be able to do more with the same resources – and likely in better quality, since humans can focus on creative and high-value work, while AI handles the repetitive tasks consistently. More games, more content, longer lifespans for beloved titles. Those who don’t adopt will find themselves at a severe competitive disadvantage.”

It’s difficult to get a handle on exactly how widespread AI-first workflows are in mobile right now. Given the enormous public backlash that games companies have faced over AI use, very few are willing to speak about it publicly. Mario Wynands, CEO of the New Zealand-based mobile and PC developer PikPok, says that it’s “hard to tell” how widespread AI adoption is given that he doesn’t have direct insight into other companies’ workflows.

“However, it is very clearly having a growing presence, at least on the UA [user acquisition] side of the mobile business, with the increasing obvious usage of generative AI UA creative,” he says. “Localization, asset creation, UA campaign management, market and metrics analysis, and game prototyping in hyper casual are other ‘high volume’ aspects of successful mobile games at scale where I expect market leaders and AI-centric startups to be investigating or adopting AI aggressively.”

Ultimately, the consumer will decide whether AI-generated content is acceptable. We’ve already seen enormous pushback against the technology in the PC and console space, but so far there has been relatively little outrage about AI among mobile game players – another reminder that the mobile industry is almost an entirely separate world from the PC and console space. It remains to be seen, however, what level of AI-generated content mobile consumers are willing to accept, as well as what artists and designers in the mobile industry will tolerate.

For Lehr though, the transition to an AI-first workflow seems inevitable. “Within the next two years, I expect AI-assisted content generation to be standard practice across the industry – at least in mobile games. The tools will only get better and more accessible.”