Thanks a lot! 😄
If I include not just the first desk scene, but also the later version with the character and chair added, I think it took around 20 hours in total.
For the character, I imported a rigged model that I had previously used in my Three.js train project. I also imported the animations, and the LLM adjusted, arranged, and merged them into the scene.
As for “LLM2,” I’m not actually sure what I meant there, so it may have been a typo. Maybe I meant either Claude or Codex. For this project, I’m using GPT 5.6 Sol.
My PC is relatively high-end, with 64 GB of RAM and 10 GB of VRAM. But honestly, most of the time was spent iterating on prompts, making adjustments, and waiting while the model was thinking.
Thanks again!
@thfcizzy Hi😃! I started on July 23 from installing Blender 5.2, and had it finished by the evening of July 24. I’d say it took about six hours of actual work in total.
Sure! I used Blender 5.2 and installed the official Blender MCP add-on in Blender. Then, using the VS Code terminal, I set up the Blender MCP server locally and registered it with an MCP-compatible coding client. After that, I entered instructions in the coding client’s chat panel in VS Code, and the generated objects and edits were applied directly to the scene currently open in Blender.
The exact setup steps vary slightly depending on the coding client, so asking the AI you use for the detailed Blender MCP setup procedure should give you more specific instructions.