Run local 3D model generation from photos on your own GPU
If you want to create a 3D model from a single photo on your computer but prefer not to upload files to external servers, Modly is a desktop app that handles the entire process on your local GPU. It runs directly without installation or can be installed via an installer on Windows, Linux, and Apple Silicon macOS.
Workflow from photo to mesh
Modly structures the conversion from image to 3D mesh as a visual connection. After setting up a default connection like 'Image -> Generate Mesh -> Add to Scene' in the 'Workflows' tab, you can start generation in the 'Generate' tab. Errors or logs during generation are available in the 'Settings/Logs/Errors' menu, and if a connection is incorrect, the app displays a warning while keeping the current screen. Imported meshes can be smoothed or have their polygon count reduced within the app, and the results are saved back to the workspace.
Adding extension models via GitHub
Modly supports external models and processes through an extension system.
- Hunyuan3D 2 Mini series Available in three versions: Basic, Turbo, and Fast.
- TripoSG and Trellis2 GGUF Supported as separate official extensions.
Controlling the app with scripts
A CLI tool allows you to send commands to the running app without direct screen interaction. This tool uses only the Python standard library and outputs results in JSON format. It supports basic commands such as health, model, and workflow-run, while the generate command automates the process of taking image input, starting generation, and exporting the final mesh. It also includes an experimental command for running ComfyUI workflows.
Things to check before running
Running in development mode requires installing Node.js dependencies and setting up a Python virtual environment. macOS supports only Apple Silicon chips, while Windows and Linux use different window control methods. The license is MIT, but if you create a new app based on this project, you must credit the original project and the creator (Lightning Pixel) in the UI or documentation. It is recommended to verify in advance whether your GPU performance can handle 3D model generation.
