Overview

The basic workflow is:

1. Load an image
2. Separate it into spectral bands
3. Save a spectral project
4. Load that project in the Combine tab
5. Edit the spectral sensitivity and density curves
6. Preview the result live
7. Export a reconstructed image or a 3D LUT

Important Limitation

This tool is still an approximation.

A normal RGB image does not contain the full real spectrum of the scene, so the spectral reconstruction is estimated from RGB data. That means the result can be visually convincing and useful for experimentation, but it is not a true spectrometer measurement.

Supported Input Formats

• PNG
• JPG / JPEG
• TIFF / TIF
• BMP
• CR3 (when rawpy is installed)

Main Features

• Separate tab for RGB/RAW → spectral band separation
• Combine tab for spectral reconstruction and look design
• Editable Red / Green / Blue spectral sensitivity curves
• Editable Red / Green / Blue H&D / characteristic density curves
• CSV import/export for individual R / G / B curves
• Live reconstruction preview
• Preview brightness control that does not affect exports
• Noise cleanup and film grain shaping controls
• Multiple spectral band export formats
• Multiple final image export formats
• 3D LUT export (.cube)
• LUT input and output color space selection
• Scene illuminant prior for spectral separation
• Background threading for heavy operations

Files Created by Separation

When separation runs, the program creates a spectral project folder. That folder usually contains:

• spectral_cube.npz
• source_reference_rgb.npz
• metadata.json
• bands/ (individual spectral band files)

Recommended Folder Usage

Keep the entire project folder together.

Do not move only one file out of it. The Combine tab expects the project data to remain intact.

Dependencies

This tool uses the following Python packages:

• numpy
• pillow
• matplotlib
• tifffile
• rawpy (optional, required for CR3 support)
• tkinter (usually included with Python on Windows)

CR3 loading issues

• Install rawpy
• Make sure the installed wheel includes usable LibRaw support

TIFF export issues

• Install tifffile

Performance Notes

• Separation and export run in background threads, so the UI should stay responsive
• Full-resolution separation can still take time, especially with many bands
• 31 bands is a good default
• 61 bands is a higher-quality mode, but slower and heavier
• 128³ LUT export can be much slower and heavier than 32³ or 64³

Color Pipeline Notes

• The reconstruction path is film-style, using spectral sensitivity + H&D density mapping
• Preview brightness is preview-only
• TIFF 32-bit float and RAW linear .npy preserve the most linear detail
• PNG and TIFF 16-bit are better for ready-to-view images
• The LUT path uses the current:
  • spectral sensitivity curves
  • density curves
  • density x-range
  • visible wavelength range
  • LUT input/output modes

LUT Notes

The program exports .cube 3D LUTs.

Make sure the LUT input color space in the app matches the color space of the footage or image the LUT will be applied to.

If the input assumption is wrong, the LUT can appear:

• too bright
• too saturated
• color shifted

Recommended Default Settings

General work

• Bands: 31
• Wavelength range: 380 to 720 nm
• Band export: TIFF 16-bit or TIFF 32-bit float
• Final export: TIFF 32-bit float for maximum retained detail

Fast testing

• Bands: 21
• Smaller preview max side
• LUT size: 32 or 64

Higher quality

• Bands: 61
• TIFF 32-bit float band export
• RAW linear .npy or TIFF 32-bit float final export

Main Controls

Separate Tab

• Start nm / End nm
  Wavelength range used for the spectral approximation

• Band count
  Number of spectral grayscale slices created

• Smoothness
  Regularization that keeps the recovered spectrum smoother

• Energy regularization
  Stabilizes the inversion

• Preview max side
  Resolution used for preview images and the preview cube

• Band export format
  Format used for the saved individual spectral band files

• Scene illuminant
  Auto or manual light-source prior used during separation

Combine Tab

• Visible from / to nm
  Wavelength range used during reconstruction

• Sensitivity log ceiling
  Top scale of the spectral sensitivity editor

• Density log ceiling
  Top scale of the density editor

• Characteristic-curve exposure axis in lux·s
  Displays the density x-axis in lux-seconds instead of relative log exposure

• Final export format
  Output format of the reconstructed image

• 3D LUT size
  Cube size for LUT export

• LUT input mode
  Assumed input color space for LUT generation

• LUT output mode
  Output encoding / color space for LUT generation

• Preview brightness
  Preview-only brightness balancer

• Noise cleanup
  Reduces coarse digital-looking noise in the preview/exported positive image

• Film grain
  Adds a finer grain layer after cleanup

Known Practical Limitations

• RGB → spectrum is underdetermined, so some film differences remain subtle
• Landscapes in neutral daylight may not show very large differences between film stocks
• Stronger differences usually appear in:
  • tungsten scenes
  • LED scenes
  • skin tones
  • saturated reds
  • mixed lighting
• Accurate spectral sensitivity curves help, but dye absorption behavior and scan transforms also matter for realism

PyInstaller / EXE Build

Use the generated build files:

• spectral_tool_app_v18_fixed.spec
• build_spectral_tool_fixed.bat

Place them next to:

• spectral_tool_app_v18_fixed.py

Then run the batch file.

Build output is usually placed under:

• dist\SpectralBandSplitter\

Troubleshooting

App does not open

• Confirm the Python version is compatible
• Confirm tkinter works on your system
• Confirm the matplotlib TkAgg backend is available

CR3 loading fails

• Install rawpy

TIFF export fails

• Install tifffile

Preview looks too bright

• Reduce Preview brightness
• Export TIFF 32-bit float or RAW linear .npy to inspect unclipped linear output

LUT looks wrong in Photoshop or another app

• Make sure LUT input mode matches the document or footage color space
• Make sure the host application is not interpreting a Rec.2020 LUT as sRGB, or the reverse

Summary

Spectral Band Splitter is a spectral-approximation and film-reconstruction tool designed for experimentation, curve-based look design, and LUT generation from RGB or CR3 input.

It is most useful when treated as:

• a spectral look-development tool
• a film-style reconstruction tool
• an experimental color pipeline

rather than a physically exact spectrometry system.