v0.3.0

This release dramatically improves performance and memory requirements. It does so by replacing the transformers backend and full model with a llama-cpp-python backend and quantized model.

Before installing, uninstall any previous release, then restart Blender.

Installation

Windows/Mac/Linux (CPU)

This version is slow, but supports more devices. If you meet the requirements, the GPU version is recommended.

Requirements

• Blender 4.2

Steps

• Download meshgen-cpu.zip in Assets below
• In Blender, go to Edit -> Preferences -> Add-ons -> Click the arrow in the top right then Install from Disk
• Choose meshgen-cpu.zip
• In addon preferences, click Download Required Models

Windows/Linux (NVIDIA GPU)

This version is much faster, but requires a high-end NVIDIA GPU.

Requirements

• An NVIDIA GPU with at least 8GB VRAM
• CUDA 12.1
• Blender 4.2

Steps

• Download meshgen-cuda.zip in Assets below
• In Blender, go to Edit -> Preferences -> Add-ons -> Click arrow in the top right then Install from Disk
• Choose meshgen-cuda.zip
• In addon preferences, click Download Required Models

Usage

• Press N -> MeshGen (or View -> Sidebar -> Select the MeshGen tab)
• Click Load Generator (this will take a while)
• Enter a prompt, for example: Create a 3D obj file using the following description: a desk
• Click Generate Mesh

Troubleshooting

• Find errors in the console:
  • Windows: In Blender, go to Window -> Toggle System Console
  • Mac/Linux: Launch Blender from the terminal
• Report errors in Issues