I ran into this wall too. Someone here on HN said their general test is “make a browser only simple rts game with no AI and no multiplayer” What makes a game fun is very different than the engine of a game. My kid asked me to make a game where you brew potions. Okay, done. Adding ingredients, having physics to drop the items rather than “okay they just appear in the pot” (or worse it just says “okay they’re in there!”) Especially for kids there’s a physicality you have to capture to make it both fun and understandable to a seven year old.

Getting cutesie stylized 3D models is something that’s trivial with an RTX 5090, a ChatGPT pro subscription (unlimited image generation), you run Trellis2 plus a few other open source things in a pipeline that your agents can queue and it’s astonishing how much cool stuff comes out the other side. But the graphics don’t make the game fun at all, they’re just set dressing for the fun.

There’s been a lot of learning going from 0. “Okay, 3D model of a character. Oh, this model is useless since it isn’t in a T Pose I can’t rig it. What’s a rig? Okay, there’s a rigging ML model. Download that. Okay, how do I animate it? Oh, cool there’s a model for that. Oh wait my model has holes in it, that looks weird. Okay there’s an ultra shapes library that helps improve geometry. Whoops, that strips all the textures and shaders. Okay, trellis2 has a mode that takes an existing model and retextures it. Okay wow these look good, the characters are walking around! This goblin is break dancing! Okay uhh, what do you actually do in this game?”

Like it feels like that trap you can get stuck in when one part of something is trivially easy, so I have like 500 random 3D assets that are honestly pretty good looking for a game where the core gameplay loop is not developed at all because I have no idea what would make it feel fun. Because I can prompt and say “oh wouldn’t a Christmas village be cool?” And I wake up the next morning with 50 3D models of Christmas village stuff and characters and I say “wow, neat!” (It takes maybe 8 minutes end to end for the pipeline to generate one 3D model, so I just run it overnight). But then I have to manually place them in the world (if you let the AI do it in unreal engine 5 it places them via coordinates which become impossible to move inside the unreal editor).

The fun part is “wow, this is something I’m making with my kids, and it’s unique to us”. That’s what keeps me at it. I’ve never seen my kid so engrossed and excited to help me with something, she’s the one coming up with ideas and saying “what if we did this and that?” and then seeing those things become real is really neat. The bottleneck is there’s a dozen agents that can work on different parts of the game but it’s a chaotic mess.

Still, I’d imagine this is how people learn is by making something that’s a piece of junk then making something that’s better. I don’t plan on releasing my pieces of junk unless I feel like they’re actually fun.