Building something from scratch in Blender with a coding agent can get crazy expensive in tokens. It can take days, too. I had Codex build a raccoon wizard that way. It did a pretty good job. When I tried generating most of the parts locally and had Codex fit them together in Blender, it already had something to work with.
Meshy and Tripo can give you great results, but a few wrong turns can get expensive. Local 3D models can do some really good work too, once you figure out which workflows and settings work for what you’re making. I built Mumblebox 3D Asset Foundry because I wanted my agent to be able to bridge those gaps.
The MCP is working well. The agent can run workflows, check previews, and grab exports. What I’m working on now is teaching it when to use each workflow. If I need a character head from a reference, let my GPU make a starting mesh and have the agent check it and fit it in Blender. If I need a bracket with exact dimensions, the agent can make that right in Blender. I want it to be able to make a whole batch of assets locally and spend its time and tokens where they actually help.
Pinokio handles setup and dependencies, and I can run the same visual wizards myself whenever I want. I also plan to integrate Tripo, Meshy, and Hi3D. The idea is to do most of the work locally and have paid services available for the jobs where they’re worth it. Foundry on Pinokio
Two Blender repos I already use are EL4CTEO’s blender-skills for things like modeling, UVs, and rigging, and 1Pio’s blender-workbench for checking geometry against references. I’ve gotten a lot out of both. blender-skills · blender-workbench
How are you writing your SKILL.md files? What did you put in yours that actually helped the agent choose the right tool? Is there a skill for writing skills that you’ve had good luck with? And if you have a Blender skill or repo you keep coming back to, I’d love a link and a little about what you use it for.
I’m really kind of humbled by how willing people here are to help and share what they’ve learned. I appreciate it.
— Brian
Building something from scratch in Blender with a coding agent can get crazy expensive in tokens. It can take days, too. I had Codex build a raccoon wizard that way. It did a pretty good job. When I tried generating most of the parts locally and had Codex fit them together in Blender, it already had something to work with.
Meshy and Tripo can give you great results, but a few wrong turns can get expensive. Local 3D models can do some really good work too, once you figure out which workflows and settings work for what you’re making. I built Mumblebox 3D Asset Foundry because I wanted my agent to be able to bridge those gaps.
The MCP is working well. The agent can run workflows, check previews, and grab exports. What I’m working on now is teaching it when to use each workflow. If I need a character head from a reference, let my GPU make a starting mesh and have the agent check it and fit it in Blender. If I need a bracket with exact dimensions, the agent can make that right in Blender. I want it to be able to make a whole batch of assets locally and spend its time and tokens where they actually help.
Pinokio handles setup and dependencies, and I can run the same visual wizards myself whenever I want. I also plan to integrate Tripo, Meshy, and Hi3D. The idea is to do most of the work locally and have paid services available for the jobs where they’re worth it. Foundry on Pinokio
Two Blender repos I already use are EL4CTEO’s blender-skills for things like modeling, UVs, and rigging, and 1Pio’s blender-workbench for checking geometry against references. I’ve gotten a lot out of both. blender-skills · blender-workbench
How are you writing your
SKILL.mdfiles? What did you put in yours that actually helped the agent choose the right tool? Is there a skill for writing skills that you’ve had good luck with? And if you have a Blender skill or repo you keep coming back to, I’d love a link and a little about what you use it for.I’m really kind of humbled by how willing people here are to help and share what they’ve learned. I appreciate it.
— Brian