Since I'm a developer with no experience in 3D design, I just automated my process with freecad and vibe-code my models with GitHub copilot.
That gives me instant visual feedback plus 3d model code I can check in into git: https://github.com/micw/freecad-automation
Is this robust enough to run headless? I'm thinking of sticking it in a Celery worker to generate parametric variants, but I'm not sure if the LLM output is reliable enough to run without that visual feedback loop.
- generate the model via AI, using plain text
- review the result (using FreeCad for visual feedback)
- iterate until the model is fine
- tell the AI which parameters shall be customizable
(see my tablet holder example, there I have parametes for a Huawei MediaPad and also for an old Samsung Galaxy Tab, both using the same model)
2. 3D Model export
This can be done entirely headless and it's very robust. No AI involved, only the python code previously generated by AI.
Currently a generic 3mf file is exported, so It needs to go though a slicer and you need to apply your printer's settings (printer type, nozzle type, all that stuff).
But the generated 3mf is just a ZIP with 3d model and the settings within. So this could also easily be automated.
