CarveAnything

A Python pipeline designed to transform 2D bas relief images into detailed 3D bas-reliefs using Depth Anything 3 depth estimation.

Note: Currently, the script doesn't convert a normal image to an image of a bas relief, it only converts the latter to a mesh. You will have to do this using a different method, such as a local image pipeline, Gemini, ChatGPT, etc. An example prompt that performs pretty well especially on animals:

recreate this exact image as a pure white matte marble bas relief. orthographic view. no frame border. compress depth of the background, enhance the depth of the subject. clear sharp lines.  identify and replicate the subject's particular features. keep the same aspect ratio. very light shadows. high resolution, crisp details, 4k, 8k. use the image as line art control net 

Note: The Gemini images perform best for the moment

Features

• Depth Estimation: Uses Depth Anything 3 for highly detailed depth map prediction.
• Automated 3D Relief Generation: Converts depth map directly into a mesh (OBJ).
• Interactive CLI Interface: Simple, prompt-guided execution with command-line arguments.
• GPU Accelerated: Optimized for CUDA-enabled PyTorch execution.
• CNC / 3D-Printing Ready: Designed to generate usable 3D geometries for CAM software, carving, and 3D printing.

Supporting Carve Anything

You can help further development of Carve Anything by contributing to the work or by financially supporting the creator here: OpenSource CNC Ko-Fi page Any bit of help counts a lot, so thank you all in advance!

Prerequisites

Before setting up CarveAnything, ensure you have the following installed on your host system:

• Python: 3.10 or higher recommended (Python 3.9 minimum)
• Git: Installed and configured
• NVIDIA GPU Driver: Version 550.58 or higher (for CUDA 12.4 support)

Installation

Follow these step-by-step instructions to set up CarveAnything on your local system:

1. Clone the Repository

Open your terminal or PowerShell and clone the repository:

git clone https://github.com/opensourcecnc/CarveAnything.git
cd CarveAnything

2. Set Up a Virtual Environment

On Windows (PowerShell):

python -m venv venv
.\venv\Scripts\Activate.ps1

On Linux / macOS:

python3 -m venv venv
source venv/bin/activate

3. Install Dependencies

Upgrade pip and install all necessary Python dependencies:

python -m pip install --upgrade pip
pip install -r requirements.txt

If the requirements installation fails on depth_anything_3, try:

pip install depth_anything_3 --ignore-requires-python

edit requirements.txt, remove depth_anything_3, then:

pip install --no-build-isolation -r requirements.txt

Note on PyTorch / CUDA: requirements.txt configures CUDA 12.4 PyTorch builds by default. If running on CPU only or requiring a different CUDA release, refer to the PyTorch Get Started guide.

Model Weights Setup

CarveAnything requires model weights from the Hugging Face DA3MONO-LARGE repository to perform depth estimation.

1.The Hugging Face repository: depth-anything/DA3MONO-LARGE

2. Download the following files:

   • config.json
   • model.safetensors

3. Place the downloaded files into the da-models/ directory in the root of the repository.

Usage

Before launching the program, always ensure your virtual environment is activated:

Windows (PowerShell):

.\venv\Scripts\Activate.ps1

Linux / macOS:

source venv/bin/activate

CarveAnything features an interactive terminal workflow. You do not need to pass command-line arguments when launching the program.

Simply execute the main script:

python CarveAnything.py

Once launched, the script will guide you with step-by-step on-screen prompts to input your parameters.

Generated output files (depth maps and 3D mesh files) will be saved automatically in the output/ directory.

License

See LICENSE for details.