Physics-based Technicolor 3-Strip (Process IV) Emulation

[Naveen Chandran]

Hi everyone,

I’m a developer working on the color science side of post-production, and I’m looking for some feedback from cinematographers on a tool I’ve been building.

I’ve been trying to accurately reconstruct the Technicolor 3-Strip (Process IV) look in DaVinci Resolve as a DCTL Tool. I found that standard LUTs and RGB curves completely fail to capture how those vintage cameras actually responded to light, so I decided to build a mathematical model of the physical pipeline instead.

I’ve created a DCTL that simulates the original photochemical process from capture to print. I’d love for anyone who frequently shoots or lights for vintage looks to throw some challenging footage at it and tell me how it responds to your exposure.

Here is a quick breakdown of what the math is actually doing under the hood:

• The Camera Capture: It un-mixes the broad spectral overlap of modern digital sensors to approximate what light looked like passing through the original narrow-band Wratten #25, #58, and #47 filters inside the 3-strip prism.

• Subtractive Density: Instead of digital saturation (where colors get brighter), it uses the Beer-Lambert Law to simulate physical CMY dye absorption. If you push the saturation of a color, it physically gets denser/darker, just like a real imbibition print.

• Optical Scatter: It uses true Laplacian blur kernels for the light scatter/halation, rather than standard digital Gaussian blurs.

I have a fully functional demo available, as well as a comprehensive PDF manual breaking down the history and physics of the camera and printer mechanics: https://nxcolor.com/3-strip

I’m really curious to hear your thoughts on how this handles skin tones under different lighting temperatures, and if the halation/scatter feels optically authentic to you. Any technical critique is incredibly welcome!

Thanks, Naveen

[David Mullen ASC]

The tricky thing is that the 3-strip camera used a prism to split light through a green and magenta filter, the green light going to panchromatic stock.

The magenta light went to a blue-sensitive stock that had no anti-halation backing and was dyed red, and placed in front of panchromatic stock. The blue-sensitive stock only responded to the blue wavelengths in the magenta light and then was filtered by the dyed base to send only red wavelengths to the panchromatic stock, which was somewhat soft as a result.

So the red record was always softer and a bit grainier which affected skin tones since they have more red in them.

The red & blue records picked up a magenta flare from the lack of an anti-halation backing in the blue record.

In 1951, Technicolor altered the camera system to become tungsten-balanced, by replacing the gold-flecked prism with a silver coating, and probably by making the blue sensitive stock faster in some manner.

 

 

[John Rizzo]

Hello Naveen

 

we tried your demo and it crashes our GPU  (we are running it a MAC PRO WITH 4 GPU s .

 

regards

[Naveen Chandran]

Hello John,

Which version of Resolve are you running?

Also I've tested on mac Silicon (Metal) and Windows (CUDA and OpenCL), does it show any compilation error dialogs or does it just crash?

Best Regards,

Naveen

[Naveen Chandran]

Just a quick update on this — I’ve released v2 of the tool.

A fair bit has changed since I first posted here. I spent some time refining how the dye transfer stage behaves, particularly the interaction between layers as the gelatin loads up, and also reworked parts of the bipack/halation side to get something that holds together better under exposure.

I’ve also added broader input support so it can be used earlier in the pipeline without needing conversion.

And the tool itself is here (demo available):
https://nxcolor.com/3-strip/

Would be interested to hear any thoughts.

What's New in V2:

Expanded Input Format — Replaces Input Encoding with a new Input Format selector covering 15 camera color spaces including ARRI LogC3/LogC4, Sony S-Log3, Canon C-Log 2, RED Log3G10, Panasonic V-Log, and more. Gamut conversion is handled automatically.
Wide-Gamut Rendering Fix — In v1.1, DaVinci Intermediate and ACEScct inputs were decoded from their log encoding but their wide-gamut primaries were never converted to Rec.709 — colors were systematically wrong for both scene-referred formats. v2 adds correct gamut matrices for all 15 input formats, applied in linear light before tone mapping. Spatial effects now process in the correct domain for scene-referred footage.
IB Print Acutance — New pre-dye sharpening stage that recreates the three-dimensional pop characteristic of original imbibition prints. Adds edge snap and midtone depth before dye processing.
Gelatin Softness — Simulates the resolution limit of the physical gelatin matrix — softening fine surface texture like skin and fabric while leaving edges intact. The diffuse quality unique to real IB prints.
New Dye Physics — Adds Density Coupling and Midtone Dye Boost. Together these model how real dye-transfer color behaves: saturation darkens rather than blooms, midtones carry the richest dye, and highlights desaturate gracefully.
Refined Grain Engine — Improved IB dye grain with smoother large-scale modulation, a new long-range density variation layer, corrected dye-pool skew, and softer reticulation. Re-tuned defaults better reflect the coarser, organic texture of real IB prints.
Major Performance Upgrade — All convolution cores rewritten for significantly faster processing. Approximately 5× faster at 4K on wide-gamut inputs. Halation and bloom now scale correctly to full slider range at any resolution.
Cleaner UI — Redundant enable/disable toggles removed throughout. Several controls consolidated into the new dye physics system.