Blender and 3D AI tools
A set of Open WebUI tools that let a local AI model drive Blender, turn images into textured 3D models, and slice and run 3D prints on my Bambu P2S.
- Language
- Python (Open WebUI tool plugins)
- Tools
- Blender Bridge · Image to 3D · Printer P2S · Homelab Terminal · LLM Server Status
- 3D models
- Qwen-Image-Edit, Hunyuan3D-2mv shape and Hunyuan3D Paint texture
- Printing
- OrcaSlicer CLI and a Bambu Lab P2S
- Access
- Private homelab use. No download
These are plugins for Open WebUI that give a locally hosted model real tools instead of just a chat box. They’re the glue between my GPU machines, Blender, and my 3D printer.
The tools
- Blender Bridge (v2.1). Runs Python in a headless Blender, inspects the scene, and renders. The render is returned to the model as an image, so it can look at its own work and shown in the chat.
- Image to 3D (v0.6.1). Takes a picture from the chat and, when only a front view is given, first generates the left, back and right views with Qwen-Image-Edit. It then builds a shape with Hunyuan3D-2mv and paints the texture. Model files land on the 3D workstation and in Nextcloud.
- Printer P2S (v0.1.1). Status, slicing with OrcaSlicer’s command line, starting prints, and camera checks for my Bambu Lab P2S with its AMS.
- Homelab Terminal (v0.9.4). Lets the model run shell commands on my machines over SSH.
- LLM Server Status (v2.1). Reads real GPU and model status from llama-server and
nvidia-smiinstead of letting the model guess.
Honest notes
- The image-to-3D pipeline gives good results when the subject fills the frame. A dragon head looked great; full-body characters have small faces in the source image, so face detail is limited.
- Fine part placement in Blender can’t be judged on placeholder boxes, so I stop at “roughly right” until there’s real geometry.
- These plugins contain my own network addresses, so I’m not publishing the files.
Used in: The Long Rot and the 3D-printed hardware projects.