AI policy
I'm one developer. To make games as polished as I want them, I use AI tools. This is how.
The difference is the human work
Slop is "build me X", shipped as it came out. That isn't what happens here. I build my own tools: DIG for art, the playtester for playtesting, and Foundations for what every project stands on. Then I go back and forth. A game starts as ideas in my head, and it takes seeing them for real, and changing them, to find out what they should be. Making it is a process of discovery, not a spec handed over.
A language model is a funhouse mirror of humanity. Getting a true picture out of it takes time, effort and understanding.
What ships
Art made with DIG is finished art, and it ships in the games. Each game's page lists what in that game was made with AI, matching its Steam AI disclosure. For War of Attrition, that's What's in it.
Credit
Where I can credit someone directly, I do. I also give vibe credit: the references I fed into each game. A model is a pattern machine, so those references shape the mashup. The Fractal Nature of the Universe, for one, owes its feel to 3Blue1Brown, Primer, The Coding Train, Kerbal Space Program, Opus Magnum and Geometry Wars.
The stack
- Code: Claude, by Anthropic, a closed model. It is the only coding agent I use. It also runs inside Isekai's Storyteller and the playtester's personas.
- Art: open-weight image models, run on my own machine through DIG. Juggernaut XL v9 by KandooAI and RunDiffusion is the workhorse. DreamShaper XL by Lykon paints the watercolours, and Sinful Strokes by mnemic restyles War of Attrition's cards. ControlNet and IP-Adapter models from xinsir, TheMistoAI and h94 hold the shapes.
- Later: open-weight language models may join the playtester, for volume.
I'd hire artists if I could. At this scale, I credit the tools instead.