I wanted a hobby project. Three conditions were locked from the start: build it with AI, generate the art with Stable Diffusion, and it has to make money. I can't draw and I can't compose. So SD and AI aren't really "weapons" for me — they're more like teammates that do the parts I can't do alone. That was the premise.
But the moment I hit "okay, so what do I actually make?", I went in circles several times. I'm just writing that process down. The conclusion sounds trivial, but for me every wrong turn on the way there was necessary.
First I crossed out everything that looked plausible
The first idea was a visual novel, since it pairs well with SD. But then it hit me: visual novels get made just fine without any AI. That defeats my whole "made with AI" purpose. Cut.
Then I went toward what I actually love. I like games with great game-feel, like Sanabi. I enjoyed rage-platformers like Only Up too. But those live and die by precise action animation — which is exactly what SD can't do. SD gives you a gorgeous single character, but making that character move naturally frame by frame is hell. My teammate (SD) was useless here.
I looked at Roblox, since I've built jump maps before. But Roblox is 3D and my SD produces 2D illustrations. Mismatch again. And my research showed Roblox's algorithm quietly buries games churned out lazily with AI. I briefly considered RPG Maker, but I've used it — it's clunky, slow, and worst of all its closed structure means you can't get AI coding help. For someone whose whole plan is "let AI make this easy," it was the worst possible choice.
Taste and reality collided head-on
Here came the painful moment. I honestly listed the games I enjoy: Zenless Zone Zero, League of Legends. All of them are either big 3D action or real-time PvP. Every single one is impossible to make solo. I wondered how what I enjoy and what I can build could overlap so little.
Then another memory surfaced. I used to love short story indies like RPG Maker games and Undertale — the kind you finish the ending of in a day. And I've played a lot of TFT. Those two were the clue. The intersection of what I enjoy, what I can build solo, and what fits SD — that was the auto-battler.
Weirdly, the auto-battler passed every wall I'd hit. Combat is automatic, so no precise animation needed (dodges SD's weakness). Characters, synergies, and items are all data, so AI can generate content and balancing for me (my teammates finally have a job). And it's a kind of fun I already know I like, proven by TFT.
Waifu or cute — and why cute won
Since it's a character game, I briefly considered going the "waifu" route, since that's what SD generates best. But research showed waifu auto-battlers score right down the middle (one sat at 56 on Steam), and more importantly that space is basically a graveyard of F2P gacha exploitation and adult shovelware. Cute mascot-style games (the Chiikawa vibe), on the other hand, are a much healthier market — proven by Super Auto Pets. Safer floor, higher ceiling. So I dropped waifu and went cute.
I tore apart every famous auto-battler
Once I had the direction, I analyzed TFT, Hearthstone Battlegrounds, Super Auto Pets, Backpack Battles, The Bazaar, Mechabellum, Storybook Brawl one by one. I learned a lot here.
What to borrow was clear: async snapshots (fighting teams other players built — competition without a live server), synergy combos (data, so AI helps well), XP level-ups. Anti-staleness I learned from Battlegrounds — varying the setup each run so one "correct build" never hardens. The common complaint that Super Auto Pets gets repetitive is exactly this problem left unsolved.
What to avoid I learned too. Storybook Brawl was a fine game but touched NFTs and died; being server-based, it vanished entirely when the studio folded. The Bazaar had a great system but self-sabotaged its first impression with pay-to-win monetization, and its information was so dense you had to keep a wiki open to play. All of that was a textbook of "do this and you die."
I got honest about the money
Here I asked myself the most important question. "If I knew in advance it wouldn't sell a single copy, would I still build it?" The answer was no. So I admitted my real motive is money — not some vague "I do it because it's fun."
But that alone was weak, because solo game dev has low expected value (the median Steam indie earns the developer only a few thousand dollars). So I added a second motive — proving that "you can build a game fast with AI, without the grind." That way, even if it doesn't sell, the methodology remains. It gives me a floor.
And I redefined the goal of game #1: not a hit, but validating whether revenue is even possible. A low-cost test of whether I'm someone who can make something that sells. Framing it that way settled everything. Build fast to lower risk, and open up the dev process as both proof and marketing.
So this is what I check, every time
Once the direction was set, I needed a compass to return to whenever I got lost mid-development. So I turned it into a checklist.
It comes down to three questions: Does it sell? Can it be fast? Is it seen? Every time I want to add a feature, I ask "can the validation survive without this?" — if yes, cut it. If I'm grinding by hand, I hand it to the AI. Before adding any art, I check the fun with plain boxes first. Gameplay never goes behind a paywall; only cosmetics are for sale.
I redefined success too. Even if game #1 doesn't sell, if I learn why it didn't, that's a successful validation. Real failure isn't "no sales" — it's quitting, exhausted, without ever knowing why. So building fast and cheap is the whole point.
The next steps are set. Build a prototype with just the core fun → release a demo and open the devlog → read the signal of whether wishlists grow → if there's a response, launch. This cycle itself will answer my question: am I someone who can make something that sells?
Maybe it's a trivial conclusion. "I'm making a cute auto-battler" is all it amounts to. But the things I crossed out, the walls I hit, and the motives I admitted along the way — I needed all of it. Now I just have to build it.
Top comments (0)