Building Halo Infinite Forge maps with local LLMs, DCjson and Blender MCP
swagonflyyyy · reddit · 2026-10-27
The author uses a local LLM (Q3.8-FN-NVFP4 quant, Ninfer-v3 fork) to auto-generate Halo Infinite Forge level blockouts — previously thought impossible since EAC blocks writing MVAR files and there's no public API or MCP.
Workflow:
- Community modding tools extract structure/properties of 800+ assets
- Agent generates DCjson files; a Python script places assets in Forge with coordinates, scale and rotation
- Per-room folders with layered DCjson files (walls/floor/ceiling/doorways) concatenated up to a root file for the whole map
- Nested AGENTS.md per room describing purpose, dimensions and connections
- Blender MCP lets the agent inspect converted .blend visualizations
Findings:
- Heavy trial and error, but each iteration improves
- Smaller models only handle one room at a time
- Local LLMs unreliable for long-horizon multi-room work
- Parallel subagents only trustworthy in short bursts
- Agent restricted to a handful of primitive assets to keep layouts manageable
The author is still weighing whether fully delegating room design is worth it given the required micromanagement.
More from coding & agent
- Cursor launches Agent SDK: agents as code with Markdown prompts and TypeScript tools — soleio · 2026-10-27
- Vibecheck, the 'usable formal methods' hackathon, sells out two floors in SF — spikedoanz · 2026-10-27
- How do you verify AI agent outputs beyond unit tests and LLM-as-judge? — Practical_Green5775 · 2026-10-27
- AI agents may soon become the only way to run code, not just write it — IgorCarron · 2026-10-27
- Hermes Agent used as browser QA tester finds 8 issues in a bug-seeded site — techlatest_net · 2026-10-27
- Open-source FableCut lets AI agents edit a real video timeline via remote MCP — FastPresence9799 · 2026-10-27