Source‑available • Windows first • CUDA ready

AI powered 2D to 3D conversion for cinema and VR

VisionDepth3D brings pixel accurate, real time parallax shifting to your 2D videos. Built for creators who want clean depth, natural convergence, and flexible export targets.

1.1k+ downloads 40+ GitHub stars VR and cinema ready
VisionDepth3D 3D Generator
VisionDepth3D Depth Estimation
VisionDepth3D Frame Tools

Highlights

Core capabilities designed for clean depth and comfortable viewing. Every feature is built to help creators ship better 3D faster.

Per pixel parallax

GPU accelerated depth weighted shifting for precise stereo alignment and minimal artifacts.

Zero parallax tracking

Subject aware zero parallax to keep the focus natural during cuts and complex motion.

Edge aware masking

Suppresses ghosting by protecting edges with smart shift masks and pixel healing.

Scene aware dampening

Smooths rapid parallax changes to maintain comfortable depth across scenes.

Floating window control

Simple controls for windowing to reduce divergence and improve screen edge comfort.

Flexible export

Half SBS, Full SBS, VR180, anaglyph, and interlaced outputs for your target device.

The VisionDepth3D Method

A proprietary workflow for real time 2D to 3D conversion, tuned for cinema and VR.

  • Depth weighted continuous parallax shifting
  • Subject aware zero parallax tracking
  • Edge aware shift masking to reduce ghosting
  • Floating window stabilization
  • Scene aware parallax dampening
  • CUDA accelerated stereo warping
  • Depth of field simulation with pixel healing

Documentation is provided for learning. Redistribution of the method text or derivative use is not allowed.

Built in tools

Everything you need to go from flat footage to clean 3D. Pick your models, preview in real time, and export.

Depth estimation

25+ models supported including DPT, ZoeDepth, Depth Anything v2, Marigold, and more.

Upscaling

Real ESRGAN based upscaling for crisp final output. Custom model support included.

Interpolation

RIFE based frame interpolation to push smoother motion when you need it.

GPU pipeline

CUDA and ONNX runtime support. TensorRT friendly setup for advanced users.

Batch rendering

Process entire folders and paired video plus depth map lists with progress indicators.

Audio tools

Attach and offset audio, rip streams, and keep sync during export.

Download

Grab the installer and start converting. Windows 10 or 11 is recommended. NVIDIA GPU is recommended for best performance.

  • Runs on CPU or GPU. CUDA support recommended for speed.
  • Source available on GitHub. License: personal and non commercial.
  • No telemetry. Your renders stay on your machine.

Formats supported

  • Half Side by Side
  • Full Side by Side
  • VR180
  • Anaglyph
  • Interlaced

Minimum requirements

  • Windows 10 or 11
  • 8 GB RAM minimum, 16 GB recommended
  • NVIDIA GTX 10 series or better recommended

FAQ

Does it run on AMD or Apple GPUs

Yes on CPU. GPU acceleration is focused on NVIDIA with CUDA and ONNX. Apple Silicon support is planned through Metal backends in the future. AMD GPU users can run on CPU or try DirectML when available.

Can I use my own depth model

Yes. Place your model weights in the weights folder and select it from the model dropdown. ONNX and PyTorch options are available.

What license does the app use

VisionDepth3D is licensed for personal and non commercial use under the VisionDepth3D License Agreement. Commercial or revenue generating use requires written permission from the Licensor. The Method description on this site is informational and does not grant rights to create derivative works of the method text. Third party components remain under their own licenses. See the LICENSE for full terms or email redsky90@gmail.com for commercial licensing.

Where can I get help

Open an issue on GitHub or post in the community thread. You can also contact the maintainer through Itch.io.

License

Summary of terms. This is not a substitute for the full license.

Grant

Free personal and non commercial use. No automatic rights for business or revenue generating use.

Restrictions

No redistribution, sublicensing, resale, reverse engineering, or derivative works of the software or method text.

Ownership

All rights remain with the Licensor. Third party components are governed by their own licenses.

Third party notices

Depth models such as DPT, MiDaS, ZoeDepth are Apache 2.0. FFmpeg is GPL or LGPL depending on build and is invoked as an external process.

No warranty

Provided as is without warranty. Use at your own risk.

Governing law

Ontario, Canada.

Read the full license: LICENSE on GitHub. Commercial licensing: redsky90@gmail.com.