Day 3 of making a cozy game with no dev experience. FAQ edition!
- Published
- 09/04, 04:31
- First discovered
- 09/04, 14:00
- Type
- Dev community · RSS
A developer, on Day 3 of creating a cozy game with no prior experience, shared their workflow using various AI models. They utilize Opus 4.6 for brainstorming, Fable 5 for planning, and Opus 4.8 for coding. While currently relying heavily on AI, their long-term goal is to learn coding and use Blender independently, moving away from AI-generated images. They expressed surprise at their current capabilities with AI, a good workflow, and creativity.
Day 3! Got way more responses than i expected on my last posts and couldnt get to everyone so heres the answers to the most common questions i kept seeing.
Workflow: Opus 4.6 writes prompts → Gemini creates 2D images → Meshy AI turns them into 3D models → Claude cleans them up in Blender (polygon reduction) → into Unity for testing. Every asset goes through this. I tried skipping Meshy and having Claude do 3D directly in Blender but it couldnt nail the look.
Animations: The models come out of Meshy with no skeletons. Claude handles rigging in Blender, then i test in Unity and go back and forth a lot. "Make it more squishy," "lean more side to side," stuff like that. Also had Claude research games with the feel i wanted.
Claude setup: Desktop app, no MCP or CLI. Claude just drives Blender and Unity through my computer.
Which model: Opus 4.6 for brainstorming (more creative than newer versions imo), Fable 5 for planning, Opus 4.8 for coding.
Art style: Drew a terrible sketch, uploaded it to Claude, had it write a detailed Gemini prompt, nitpicked until i liked it. Every prompt after that references the same style for consistency.
Long term: Ultimately im trying to learn how to actually code, or at least understand it, use Blender myself and not rely on AI generated images. For now though im pretty impressed with what im able to do with only AI, a decent workflow and some creativity.