StereoMaster is a GUI application for converting 2D videos into impressive 3D content using AI. It combines:
2025/02/09Initial commit.2025/02/16Update: Added downscale inpainting, anaglyph file merging, enhanced partial frames, improved StereoCrafter (mask dilation, blur, configurable chunking), refined color matcher (fixed preview vs. SBS mismatch), new outputs (4KHSBS, Right-Only, EXR export), plus various bug fixes.2025/02/28Early Blackwell support, improved color matching in inpainting, new mixed inpainting system, fixed missing frames and repeated frames at the end of each chunk, UI/UX enhancements, VDA split fix for batch_size, and improved installation process.
- Depth map generation using DepthCrafter or Video Depth Anything.
- Splatting (stereo shift) and filling (inpainting) with StereoCrafter.
- User-friendly GUI with:
- Dynamic adjustments for disparity, convergence, and depth parameters.
- Keyframes to modify 3D intensity over time.
- Multiple preview modes (Original, Depth, Anaglyph, etc.).
- Scene detection and merging of clips for longer videos.
- OS: Windows 10/11 (64-bit)
- GPU: NVIDIA (RTX 3000/4000/5000)
- Required VRAM: 12 GB (16 GB recommended)
- Python: 3.12
- Git: Required for dependency management
-
CUDA 12.6 (Stable)
- Use
requirements_cu126.txtfor standard RTX GPUs. - Torch versions are stable releases.
- Use
-
CUDA 12.8 (Nightly) for Blackwell Support
- Use
requirements_cu128.txtto run the project on the latest NVIDIA GPUs, including those based on the Blackwell architecture (RTX 5000 series), as well as older architectures (Ampere, Turing, etc.) - Torch is installed from nightly builds; there is no official xformers wheel yet.
- Use
Note: After you complete all these steps, you can automatically launch StereoMaster by double-clicking the file
Launch StereoMaster.batin the repository’s root folder.
-
Download here:
Python 3.12.8 (64-bit) for Windows -
Run the installer and check the box “Add Python to PATH.”
-
Click Next → Install → Finish.
-
Open CMD (Win + R → type
cmd→ press Enter) and run:python --versionYou should see:
Python 3.12.8
Windows sometimes has a placeholder Python alias in WindowsApps that can interfere with the real Python installation.
Delete it with:
del C:\Users\%USERNAME%\AppData\Local\Microsoft\WindowsApps\python.exe
Then verify again:
where python
Ensure it points to the real Python installation
-
Download from:
Visual C++ Build Tools -
Run the installer and select:
- C++ CMake tools for Windows
- Windows 10/11 SDK (latest version)
- MSVC v143 - VS 2022 C++ x64/x86 (14.40-17.10)
- C++ build tools for MSVC v143-VS 2022 for x64/x86 (latest)
- Spectre-mitigated libraries for MSVC v143 – VS 2022 C++ x64/x86 (v14.40–17.10)
(If you don’t see these options, check Individual Components.)
-
Download from:
Git for Windows -
Run the installer (accepting the default options is usually fine).
cd C:\
git clone https://github.com/murdavs/StereoMaster.git
cd StereoMaster
-
Create a new virtual environment:
python -m venv stereomaster_env -
Activate it:
call stereomaster_env\Scripts\activate -
Upgrade pip:
pip install --upgrade pip setuptools wheel -
Choose the requirements file you need:
-
For standard (CUDA 12.6 stable):
pip install --use-pep517 --no-cache-dir -r requirements_cu126.txt -
For Blackwell support (CUDA 12.8 nightly): includes compatibility with older architectures:
pip install --use-pep517 --no-cache-dir -r requirements_cu128.txt
(Torch with CUDA 12.8 is installed from nightly builds, and xformers is not yet available for that combination.)
Verify PyTorch & CUDA:
python -c "import torch; print(torch.__version__); print(torch.cuda.is_available())"
Expected:
2.6.0
True
-
Login in Hugging Face CLI:
huggingface-cli loginAdd some more info on how to get the token...
- You need to create a Hugging Face account if you don’t already have one.
- Once logged in, go to Settings > Access Tokens.
- Click New token and name it however you like.
- For simplicity, you can check all the permission scopes available (or only those you specifically need).
Note: When you paste your token in the console (e.g., after running
huggingface-cli login), you might not see it appear on-screen for security reasons, but it is actually there! -
Clone the required models into a
weightsfolder:cd C:\StereoMaster mkdir weights cd weights git clone https://huggingface.co/stabilityai/stable-video-diffusion-img2vid-xt-1-1 git clone https://huggingface.co/tencent/DepthCrafter git clone https://huggingface.co/TencentARC/StereoCrafterAt 6.2 "Clone the required models..."
- If you get an error on your first try, you might need to visit the model page (e.g., stabilityai/stable-video-diffusion-img2vid-xt-1-1) in your web browser.
- Go to the Files tab and agree to the license if prompted.
- After agreeing, the download should work without further issues (though it might take a while to complete).
-
Download depth models:
cd C:\StereoMaster mkdir checkpoints cd checkpoints curl -k -L -o video_depth_anything_vits.pth https://huggingface.co/depth-anything/Video-Depth-Anything-Small/resolve/main/video_depth_anything_vits.pth curl -k -L -o video_depth_anything_vitl.pth https://huggingface.co/depth-anything/Video-Depth-Anything-Large/resolve/main/video_depth_anything_vitl.pth
After all steps are completed:
-
Go to the root folder of StereoMaster:
cd C:\StereoMaster -
Activate the virtual environment:
call stereomaster_env\Scripts\activate -
Run the script:
python StereoMaster.py -
Alternatively, double-click:
Launch StereoMaster.bat(in the root directory) to launch automatically.
If you find StereoMaster helpful or would like to support further development, consider buying me a coffee). Thank you!
Provided “as is.” Use at your own risk; no warranty is given for potential data loss or system problems.
Below is a sample image of StereoMaster in action:
