DeepThinner v2.0

Deep Sample Optimisation for Nuke 16

Created by Marten Blumen

---

Overview

DeepThinner reduces deep sample counts per pixel through seven independent, artist-controllable passes. It dramatically improves downstream compositing speed, memory usage, and deep EXR write times with minimal to zero visual impact.

---

Installation

Pre-built Plugin (Recommended)

1. Download the correct plugin for your platform:

   • Windows: DeepThinner.dll

2. Copy the plugin file to one of the following locations:

   • Your personal Nuke directory: ~/.nuke/ (Linux/macOS) or %USERPROFILE%\.nuke\ (Windows)
   • A shared plugin directory on your NUKE_PATH
   • Your studio's custom plugin directory

3. Copy menu.py to the same location as the plugin (or merge its contents into your existing menu.py).

4. Restart Nuke. DeepThinner will appear under Deep > DeepThinner in the node toolbar.

Building from Source

Requirements

• Nuke 16 with NDK headers installed
• CMake 3.15 or later
• C++17 compiler:
  • Windows: Visual Studio 2022 (MSVC 19.x)
  • Linux: GCC 9+ or Clang 10+
  • macOS: Xcode 12+ / Apple Clang 12+

Windows

cd C:\path\to\DeepThinner
mkdir build && cd build
cmake -DNUKE_INSTALL_PATH="C:\Program Files\Nuke16.0v8" ..
cmake --build . --config Release
copy Release\DeepThinner.dll %USERPROFILE%\.nuke\

Linux

cd /path/to/DeepThinner
mkdir build && cd build
cmake -DNUKE_INSTALL_PATH="/usr/local/Nuke16.0v8" ..
make -j$(nproc)
cp DeepThinner.so ~/.nuke/

macOS

cd /path/to/DeepThinner
mkdir build && cd build
cmake -DNUKE_INSTALL_PATH="/Applications/Nuke16.0v8/Nuke16.0v8.app/Contents/MacOS" ..
make -j$(sysctl -n hw.ncpu)
cp DeepThinner.dylib ~/.nuke/

---

File Structure

DeepThinner/
├── DeepThinner.cpp    — Plugin source code
├── CMakeLists.txt     — CMake build configuration
├── menu.py            — Nuke menu registration
└── README.md          — This file

---

Quick Start

1. Create a DeepThinner node from Deep > DeepThinner (or press Tab and type "DeepThinner").
2. Connect it to your deep stream.
3. The default settings (Occlusion Cutoff + Contribution Cull + Smart Merge) are a good starting point.
4. Render, then open the Statistics group in the properties to see your reduction results.
5. Click Update Statistics to refresh the numbers at any time.

---

Passes

#	Pass	Default	What It Does
1	Depth Range	Off	Clips samples outside a near/far Z range
2	Alpha Cull	On	Removes samples with alpha at or below a threshold
3	Occlusion Cutoff	On	Drops everything behind the point where coverage reaches a cutoff
4	Contribution Cull	On	Removes samples whose visible contribution is negligible
5	Volumetric Collapse	Off	Collapses runs of low-alpha volume samples into fewer samples
6	Smart Merge	On	Merges Z-close and colour-similar consecutive samples
7	Max Samples	Off	Hard per-pixel sample cap; drops deepest first

---

Suggested Settings

Light Touch (safe for final comp)

• Occlusion Cutoff: 0.9999
• Contribution Cull: 0.0005
• Smart Merge Z tolerance: 0.005, color tolerance: 0.005

Moderate (good balance)

• Occlusion Cutoff: 0.999
• Contribution Cull: 0.001
• Smart Merge Z tolerance: 0.01, color tolerance: 0.01

Aggressive (previews, layout, WIP)

• Occlusion Cutoff: 0.99
• Contribution Cull: 0.01
• Volumetric Collapse: On, group size 8
• Smart Merge Z tolerance: 0.05, color tolerance: 0.05
• Max Samples: 64

---

Troubleshooting

Problem	Solution
No reduction showing	Ensure at least one pass is enabled. Check that input has deep data.
Banding in volumes	Reduce Volumetric Collapse group size or raise volume alpha max.
Edge fringing on transparent obj	Lower Smart Merge color tolerance or disable merge for that section.
Statistics say "No samples"	Make sure you've rendered (not just validated). Click Update Statistics.
Build error: NOMINMAX	Ensure you're using the provided CMakeLists.txt with Windows defines.
Build error: DeepOnlyOp.h	Use DeepThinner v2.0 source — v1 targeted an older Nuke API.
Plugin doesn't appear in menu	Check that menu.py is in your ~/.nuke/ or NUKE_PATH directory.

---

Compatibility

• Nuke 16.0v1+ (built against the Nuke 16 NDK)
• Tested on Windows 11,
• Passes through all channels unchanged — safe for any deep pipeline

---

Version History

v2.0

• Seven independent thinning passes (added Depth Range, Occlusion Cutoff, Contribution Cull, Volumetric Collapse)
• Colour-aware smart merge
• In-panel statistics with Update button
• Artist Guide and Technical Notes in the UI
• Thread-local scratch buffers for zero-allocation processing
• Nuke 16 API (DeepFilterOp)

v1.0

• Initial release with Z-tolerance merge, alpha cull, and max samples

---

License

This plugin is provided as-is for use in production and personal projects. Created by Marten Blumen.