Talking AI Game Startups with a Former Genshin Impact Core Creator｜“This Time, I’m Going to Be the ‘Bad Boy’ of Gaming”
Summary
恶少 has put 奈洛科技’s 10-person team and newly completed Pre-A round behind a thesis more specific than “AI makes games”: Craney turns a spark of inspiration into a complete virtual world. The product uses inspiration cards to capture characters, plots, factions and even clothing textures, then organizes local creations into a coherent world; it remains in early closed beta. The endgame is not to own another hit game, but to make the best content of the future carry the “Made with Craney” gene.
His market view is that game hits cannot be planned around users, revenue or team size, but that the next 米哈游 or 鹰角 is “bound to happen,” and tool companies can define how those studios build ahead of time. 明日方舟 launched in 2019 and 原神 in 2020, followed by hits including 重返未来：1999, 点点互动 and 黑神话; he uses the sequence to push back against the internet-style view that once a dividend cycle ends, no new companies can emerge. In content, concentrated returns come from resonance between a work and its users, not from a synchronized migration of traffic.
AI’s most important impact on game production may not be cost reduction or efficiency gains, but lowering the third and most overlooked—and potentially largest—of three barriers: professional skills, team size and cross-disciplinary collaboration. Art, engineering and design require radically different training, so founders once had to assemble a complete team before they could begin. Now a fan artist, fan-fiction writer or mod creator can use AI to cover missing capabilities and turn an idea into a discussable prototype. “It may not reach release quality in the first phase, but it will be a beginning.”
原神’s industrialized production model was not about controlling creators with SOPs, but using technology to decouple creative relationships. Each version took a full year to build but had to ship every 42 days, so more than a dozen teams worked in staggered parallel. Each creator first built a mini-game in a personal “sandbox,” then a shared pipeline assembled them into gardens, nations and a world. 恶少 believes the real moat is preserving consistency without letting cross-team dependencies and project management extinguish inspiration.
米哈游’s positioning as “not a game company, but an ACG company” shifted monetization away from stats and competition toward characters, stories and emotion. 恶少’s explanation is that character backstories and the love-hate relationships between them are the world’s most valuable assets, so the business model naturally revolves around characters. The host quotes Gemini’s summary: “create for love, pay for love.” Six-week updates were also not an arbitrary KPI, but a player expectation inherited from 崩坏学园 2 and 崩坏 3 that balances freshness against update pressure.
恶少’s shift from “star designer in the spotlight” to “architect behind the scenes” became a key motivation for the startup. Before 原神’s first PV, he spent roughly 60% of his time personally making characters, monsters, puzzles and quests, and 40% building an editor; later, new teammates produced content that amazed him within the team’s framework, convincing him that helping others express themselves was more valuable than insisting on his own benchmark. He jokingly calls his urge to participate in every great piece of content the “bad-boy” instinct of gaming, while tools let that impulse expand from one game to hundreds or thousands of works.
Craney’s potential market extends beyond games, but the claim that “film and games will ultimately merge into interactive content” remains a long-term view with an uncertain timeline. 恶少 believes interactive short dramas and interactive films already use workflows close to games; he also argues that Gen Z and Gen Alpha, facing information overload, increasingly prefer single-player experiences, vertical interests and niche IP—creating the paradox that “the more connected the internet becomes, the more single-player our games are.” That offers demand-side clues for individualized worldbuilding, but is not yet a validated end state.
The largest execution risk is visible by design: the product is still being co-created, and one important feature has already undergone a major overhaul. The version the host tested generated more than 1,000 images from a prompt and spread them across an infinite canvas; 恶少 said bluntly that the feature had since been removed, then clarified that it had not disappeared entirely and had changed substantially. Current users span fan artists, fan-fiction writers, interactive-fiction creators, animators, filmmakers, short-drama makers and tabletop roleplayers. The key signal at this stage is that previously unanticipated workflows such as “ensemble casts” are reshaping the product from the outside in.
Deep dive
1. 奈洛 is building the “inspiration-to-world” architecture layer with 10 people
At 34, a Peking University graduate who identifies as ISTP, 恶少 describes 奈洛科技 as “a group of people who like making games and are very good at making games, but at this moment don’t want to merely make games.” The team has 10 members and recently completed a Pre-A round.
His one-line positioning for Craney is “the architect of virtual worlds”: not a tool that simply takes a prompt and generates one image, but a system that starts with an inspiration and progressively expands it into characters, relationships, settings and a complete world.
The team is drawn almost entirely from different game companies, but the product remains in early closed beta and co-creation. 恶少 treats developer feedback as product input, not routine iteration after a fixed workflow has already been established.
2. An old game made him see AI’s inflection point as a “steamroller”
The game that impressed 恶少 most recently was the old title 信长之野望：创造. Its “Creation” mechanic lets a faction break through resource constraints and rapidly raise troops once it crosses a threshold, eventually sweeping across the battlefield “with the force of a steamroller.” He compared it to AI’s “wow moment.”
He prefers old games for 3 reasons: new releases are easily distorted by media sentiment, classic designs have already survived the test of time, and old games are often discounted. “Life is short, so we should try to play only classic old games.”
3. He joined 米哈游 because of a demo that looked destined to fail
While working on 天下 HD at NetEase, 恶少 believed the team had delivered what was probably the best domestic answer at the time on the feel, controls and visceral impact of a 3D action game. But it had not solved progression or monetization. The project became a major commercial success, yet he once swore he “never wanted to make another 3D action game in this lifetime.”
米哈游 later approached him with a demo of 崩坏 3. His first reaction was not that he was looking at a hit, but: “Oh my God, this is going to fail.” The art, controls and investment were all far beyond normal levels; his rational side warned that the ROI would be difficult to assess and even wondered whether the project could survive 3 years.
What changed his mind was a different conviction: he did not know how much commercial success the project could achieve, but he believed these people would “get it launched no matter what difficulties they encountered.” Passing on it could become a lifelong regret; joining could become the most boast-worthy experience of the next several years. He chose to join on the spot.
4. A $0.40 stool explains the pursuit of perfection better than an SOP
When 恶少 joined, 米哈游 had roughly 50 people: about half on 崩坏学园 2 and the other half on 崩坏 3. 蔡哥, 大伟哥 and 罗爷, the 3 founders, felt less like bosses than older classmates by 1 or 2 years. The whole team was young and energetic.
Designers, artists, animators and VFX artists often clustered around the same workstation, spending hours debating whether to move something 1 pixel left or 1 pixel right. When someone took a founder’s chair, the “super-I” founder would silently watch for a while and leave, so the team bought plastic stools costing roughly 3 yuan each.
The stools eventually appeared behind many workstations. A debugger could sit down and keep polishing details, while brainstorming required no meeting-room booking. 恶少 believes the cheap prop carried “the pursuit of perfection,” agile discussion and cross-functional co-creation more accurately than any slogan.
5. “ACG company” puts worlds and characters ahead of the game format
On his first day, a founder told 恶少: “米哈游 is not a game company, but an ACG company.” He was skeptical at first, then spent the next 10 years understanding the weight of that statement.
His thinking gradually moved beyond gameplay, monetization and progression. The founders did not begin by deciding to make a commercially successful game; they began with a story they wanted to tell and a world they wanted to show. The game was simply the medium they ultimately chose.
At 原神’s first annual meeting, 恶少 used 崩坏 3’s data to make a double several times larger, turned the huge number into a target, and was laughed at for “daring to dream.” After launch, he realized that seemingly absurd target was “just a rounding error.”
6. 原神’s core was not copying Zelda, but turning elemental interaction into world grammar
The initial motivation was straightforward: everyone on the team had played The Legend of Zelda: Breath of the Wild and wanted to recreate its sense of freedom. But converting a single-player experience into a mobile, action, anime-style and monetizable product offered no obvious path; simply “copying it” was not enough.
On the later debate over whether 原神 copied Zelda, 恶少’s view was blunt: “Anyone who has made a game will realize that copying Zelda is physically impossible in itself.” The real challenge was finding the project’s own core.
The key decision-makers went into a “black room” and worked backward from wow moments. 蔡哥’s most memorable moment was grass catching fire and rain putting it out; 恶少 instead heard Pokémon’s water, fire and grass, and a system of elemental interactions that could run through exploration and interaction.
That association became the project’s grammar. The team produced a proposal that very night, built around anime-style heroines and elemental interactions, and started executing the next day. 恶少 decided at that moment that “we were actually making a truly great game.”
7. From project kickoff to the first PV, he was a star designer before hitting the scale ceiling
恶少 divides 原神’s development into 3 stages. The first ran from project kickoff to settling on the elemental-interaction core; the second lasted through the first PV at the end of 2019, by which point Mondstadt, the characters, monsters, combat, levels and core gameplay were largely in place.
In the second stage, he came closest to being a pure creator: about 60% of his time went to personally making characters, monsters, mechanisms, puzzles and quest lines, while the other 40% went to building editors and tools for the internal team. “I was also the person hand-making the details,” so he knew exactly what the tools needed to solve.
The bottleneck came with scale. He wanted every corner of the world to reach his personal standard of perfection, and his energy was consumed by details. He could not transfer the lessons from making 3D action games to the new teammates joining rapidly, leaving him feeling that experience was not being reused effectively.
The breakthrough also came from newcomers. Working within the team’s framework, they found their own methods and produced content that amazed him. 恶少 admits that creators have large egos, but the result forced him to reassess his belief that “only I can do it best.”
8. Technical Designer pushed him from the spotlight into architectural work
恶少 consequently decided to give up the identity of “star designer in the spotlight” and become “the architect behind the scenes”: instead of personally defining all content, he would design a system that allowed more creators to express themselves consistently.
米哈游’s founders supported the transition and gave him ample resources to establish a new Technical Designer organization. The name was borrowed from the “technical designer” role in overseas AAA teams, but the actual responsibilities were explored internally from scratch.
He describes TDs as the “special forces” and “pathfinders” of a large team: they tackle problems for which the industry has no mature answer, build the underlying framework, and let later arrivals create on top of it instead of repeating every earlier mistake.
9. All platforms, 16 languages and unified data came from “doing it on day one”
The third stage was also the third transformation for 原神 and for 恶少. The project would cover mobile, PC and console, with mobile split between Android and iOS, and support 16 languages from day one. It was being built as a global product rather than a regional one.
The path emerged step by step. First came the decision to support mobile and PC. Since the game was already being produced to AAA standards, and consoles were the largest platform for AAA games, there was no reason to stop there. More difficult was connecting user data across platforms, which 恶少 called “the first case in the world.”
The host saw this as a major strategic shift, but 恶少 stressed that 米哈游 does not always place ROI at the top: “Since we want to make it cross-platform, we should do it on day one.” The ACG positioning and accumulated technical capabilities supported the decision.
10. A 42-day update cycle runs on parallel sandboxes, not creators forced into process
Updates every 6 weeks, or 42 days, came from the established rhythm of 崩坏学园 2 and 崩坏 3: any longer and players would lose freshness; any shorter would feel excessive. The real contradiction was that every new 原神 version took a full year to develop internally.
The apparent solution was simple: more than a dozen teams would start in staggered cycles, each working for a year, so one team could deliver every 42 days. The hard part was that all versions belonged to the same coupled world. Characters, regions and systems entered one another’s content, so they could not be assembled like 12 independent products.
When the host pressed for the specific mechanism, 恶少 deliberately moved the answer away from SOPs and toward a technical framework. Rules require continuous human management and may weaken initiative; the team’s real goal was to eliminate the work dependencies between functions, colleagues and teams that blocked expression.
Each creator first completes a mini-game in a personal sandbox without having to resolve every collaboration relationship up front. A pipeline he calls “highly proprietary and almost magical” then links the sandboxes into gardens, nations, regions and a world.
11. Characters are the world’s most expensive asset, so monetization means “paying for love”
Before 崩坏学园 2, 米哈游 tried different business models. After the founders decided to offer users a different experience, 崩坏学园 2 suddenly broke out. The new model gradually became a tradition for later projects, rather than a standard answer derived from theory in advance.
For content-driven games, 恶少 believes players consume “not those stats, not that competition, and not that progression,” but the content of the world itself. The host quotes Gemini’s summary: “create for love, pay for love.”
Character stories, backstories and the love-hate relationships between characters are the emotions the team most wants to convey, and the world’s most valuable assets. If the content revolves around characters, then the business model “should of course revolve around those characters.”
12. Game hits do not require a technology dividend; content resonance can concentrate in any year
The host cited roughly $100M in 原神 development and marketing spending. 恶少 said he really had worried early on that the company might collapse within 3 years, only later realizing that 崩坏学园 2 had made far more money than expected.
Internet companies often surge around a common dividend such as mobile or AI, with fewer dark horses after the dividend is exhausted. Games are different, 恶少 argues: companies’ breakout years are more evenly distributed because the decisive variable is whether an individual work resonates with players.
His timeline runs from 崩坏 3 in 2016 to 明日方舟 in 2019 and 原神 in 2020, followed by hits including 重返未来：1999, 点点互动 and 黑神话 emerging in 2022, 2023 or even after 2024.
鹰角’s debut title 明日方舟 was “reportedly” said to have generated 600M yuan in first-month revenue and 6M DAU. 恶少 preserved the rumor status of the figures, but was unequivocal about the trend: “These things will continue to happen in the years ahead.”
13. Greatness cannot be planned; what you can see is what creators want to express first
Responding to the question raised by Greatness Cannot Be Planned, 恶少 believes games are especially difficult to plan around users, data, revenue or scale. The leading companies that emerged around Caohejing over the past decade or more usually did not start with “I want to make a great game.”
A more common founder archetype is a fan artist, fan-fiction writer or game mod creator. Some had even made a female-character mod for The Elder Scrolls V: Skyrim. They first wanted to turn the emotional investment of derivative works into an original world; the game was simply the way to present the story to an audience.
As a result, these companies’ first products often looked simple on day one, sometimes leaving analysts unable to make sense of them. The host added that 米哈游, 莉莉丝, 牧童, 鹰角 and 散爆 all sit near Caohejing, yet in a neighborhood where a 300-to-500-meter walk reveals no sign of an industry miracle.
14. The stagnant pool 5 years later came from copying headcount and engines while failing to protect the spark
Around the Spring Festival before and after 原神’s first PV, The Wandering Earth premiered. 恶少 was exhilarated, believing Chinese film would produce more works at that level and that many great Chinese game teams would emerge after 原神.
About 8 years have passed since 原神’s project kickoff and exactly 5 years since launch at the time of the interview. Looking back at the industry, he described it as “a stagnant pool.” Many teams had fallen into a trap: 原神 used a lot of people, a certain engine and an editor, so they assumed they only had to copy the same headcount and tools.
He believes what was copied was precisely not the essence. The goal of industrial pipelines, editors and tools was not a flashy interface or “industrialization” for its own sake, but to help creators enter flow and nurture the small spark inside them into a towering tree. Heavy-handed process can instead grind away the instinct to express.
That became a direct motivation for the startup: rather than serve giant companies that may continue to drown inspiration, find people still in school, creating alone or just forming small teams, and help them become “the next 米哈游” or “the next 鹰角.”
15. Craney weaves details into complete worlds through inspiration cards
Craney is built around “inspiration cards.” A card can capture a character profile, biography, plot, school or setting, or zoom all the way in to a character’s ornament, pendant, clothing detail or a single texture.
The cards are not merely for generating assets. They let scattered inspirations accumulate and form relationships, eventually expanding from characters and worldbuilding into a complete virtual world. This extends 恶少’s 原神-era architectural idea of decoupling local creation first and assembling it into a unified whole later.
Because the basic objects are worlds, characters and narratives rather than engine-specific assets, he believes the same method can apply to films, television, comics and tabletop roleplaying scripts.
16. After completing a “homegrown open world,” he turned the urge to create into an urge to empower
As a junior obsessed with Dragon Nest and The Elder Scrolls, 恶少 would work for his supervisor in the lab, play games underneath while a meeting took place above, and make mods for games. Amid the domestic content drought of 2008–2010, he seriously asked for the first time: “When will we be able to make games this good in China?”
He liked seeing works used by large numbers of users and become a spiritual or emotional anchor for them, so he chose 3D action and open-world games. He assumed a Chinese open world would not arrive until he was in his 40s or 50s, but completed the goal less than 10 years after graduation—an experience he compared to “China’s men’s national football team reaching the World Cup.”
Once the goal was achieved early, his motivation began to change. Content made by the designers around him excited him just as much, and he discovered that the world was “far richer and more colorful than I thought.” Consuming and helping others complete their works could generate energy no weaker than making them himself.
The host summed it up: “The tool you’re making now is the content you’re making.” 恶少 agreed. The tool can be loved by others while giving their ideas a way to be expressed, so making another game of his own was no longer his only impulse.
17. “The bad boy of gaming” is the real reason he chose a platform over a single title
恶少 says he prefers “half-defined” missions to ones driven entirely by personal impulse. He was not initially an anime-style fan in the strict sense, but entered the category by participating in someone else’s creative goal and discovered new enjoyment there.
When he sees an excellent film or game, he feels a little jealous and disappointed: “Why wasn’t I on this team?” He jokes that wanting to meddle in every piece of great content is a bad trait, but it also makes him naturally suited to building common tools and platforms.
The host translated it as being “the bad boy of gaming”: no longer participating in only one title, but entering hundreds, thousands or even 100,000 works. 恶少 hopes that one day the very best content will carry the team’s mark: “Made with Craney.”
18. The beta has already overhauled a feature—and exposed a real demand for ensemble casts
Craney is currently being tested through co-creation. The team helps developers build works while feeding their feedback back into the product. 恶少 places particular value on creators’ own life stories, which continually introduce the team to niche subjects it had never encountered.
The old version tested by the host returned more than 1,000, sometimes 2,000 or 3,000, images from a simple prompt and spread them across an infinite canvas. 恶少 did not defend the design, saying directly that the product had gone through a “major teardown and transformation.” The feature was later removed, though he clarified that it had not disappeared entirely and had changed substantially.
Early users included visual creators, fan-fiction writers and interactive-fiction writers; outside interest has expanded to animation, film, short dramas and tabletop roleplaying. This suggests that “worldbuilding” may cross media, while leaving the team to choose between vertical depth and a general-purpose platform.
One unexpected demand is “ensemble casts.” Users want characters from the same school or faction to share uniforms, patterns and other visual markers, while preserving subtle differences and memorable details in each character’s clothing and appearance.
19. Film and games may converge into interactive content, but the timeline remains unknown
恶少 offers what he calls a “very bold” prediction: at some point in the future, the industry may stop distinguishing between film and games. Users could watch, immerse themselves and interact firsthand; interactive content would become their shared form, and the creators might ultimately be the same people.
The host argues that if this content has not yet been produced, its creators, usage habits and product requirements do not yet exist either. Would building tools for them be too detached from reality? He also says he has no idea when this future will arrive.
恶少’s practical landing point is interactive short dramas and interactive films. Their interaction structures are already closer to today’s games, so creators do not need to wait for the end state; existing game-making methods and workflows can serve them now.
20. AI’s biggest value is that creators no longer need to assemble 3 kinds of people first
恶少 believes cost reduction and efficiency gains are merely AI’s “smallest impact factor” in content. Game creation has 3 real barriers: professional skills such as drawing, modeling and coding; teams numbering in the thousands; and the most easily overlooked barrier, cross-disciplinary collaboration.
The third is especially difficult. Programmers often come from engineering backgrounds, while artists receive more artistic training; these people “often can’t even pee in the same place.” For many teams, the biggest obstacle is not money or headcount, but getting radically different people to express one world together.
In the past, a creator who came from fan art and wanted to make a game usually had to hire a programmer first. A fan-fiction writer likewise had to fill missing functions. With AI, an individual can cover absent capabilities first and turn an initial expression into a work others can see and discuss.
恶少 does not oversell current capabilities as the end state: “Even if this ability in the first phase cannot fully land or reach a quality that can be released to players, it will be a beginning.” AI first changes how creators start expressing themselves and assembling teams.
21. The richer the information environment, the more young users prefer single-player, vertical and niche experiences
As a 90s-born user, 恶少 first lived in a small world around him and only encountered computers, broadband and the larger internet after entering university. His generation therefore liked MMOs, guilds, lordship and “if you’re my brother, come chop me”—the core experience was that the more people, the more lively things became.
Those born in the 00s and 10s began with the huge virtual world supplied by the mobile internet. Excess information instead created “information panic.” They prefer retreating into highly vertical interests and finding like-minded people; the value of Douyin and Xiaohongshu is closer to information distribution and convergence.
This produces a technically counterintuitive trend: “Why is it that, as networks become more advanced, the games we play are becoming more single-player?” 恶少 sees the anime-style products centered on single-player formats after 2010 as an expression of this generational shift.
He uses anime culture’s migration as a sample. In 2008 and 2009, while studying computer science, he already had 4 东方 Project fans in the neighboring dorm room hugging anime-girl body pillows; many later entered the game industry. Back then, cosplayers felt embarrassed taking the subway; today ACG is mainstream, while more niche symbols such as Cthulhu, the SCP Foundation and the Triangular Institution are beginning to serve new identity needs.
22. He treats creation as a system that can be preserved, traced and debugged—and ultimately bets on people
恶少 says the most valuable asset he took from 米哈游 “may have been memes.” He compressed bugs, colleague photos and iconic development moments into jokes. When a hard drive failed, IT initially wanted to give up after hearing that the project assets could be restored from elsewhere; once they learned there were memes on it, they said, “Then I’ll try a little harder,” and ultimately recovered them.
Memes are not merely mementos for him, but content training inside a 128×128-pixel frame: find the comic anchor, compress the emotion, then make sure viewers receive it accurately. He studies industry history the same way, always trying to “find the father” of a design. He has even floated the deliberately provocative claim that all game content creation comes from Dune and The Lord of the Rings, tracing the lineage through shields, RTS, Age of Empires, StarCraft and Warcraft.
A NetEase vice president once required designers to analyze “why this feeling arises” while playing games. 恶少 extended that training to life: when an emotion appears, he mentally steps outside himself to observe it, looks for replicable patterns in positive emotions and locates the underlying problem in negative ones. After founding a company, the team’s anxiety, confusion and disappointment likewise became risk signals to repair early.
If given $3M to make 3 angel investments, he could think of only one person: a university classmate working on “artificial hair follicles.” Today, treating baldness may require harvesting follicles from one’s own head and transplanting them; his classmate wants to solve the problem with implanted artificial follicles. He did not force himself to name 3 people and accepted the host’s suggestion: “Give him all $3M.”