color-mixer-batch

Standalone CLI that runs both Snapmaker-Orca color-mixing algorithms on a table of target colors and reports each algorithm's best recipe side by side.

• FS — legacy Justin-Hayes degree-4 polynomial pigment blend (vendor/filament_mixer*, header-only model).
• Prusa — calibrated Yule-Nielsen spectral model (vendor/prusa_fdm_mixer*), median ΔE2000 ≈ 5.7 vs real FDM prints.

The two algorithm files are vendored unmodified from the Snapmaker Orca slicer — see vendor/PROVENANCE.md for the source commit and file SHAs.

Two ways to use it

① Web UI (recommended) — edit palette/targets in the browser, see color swatches side by side, auto-recompute on every edit, export CSV when done:

./web/start.sh        # builds CLI if needed, opens http://localhost:8008

Requires Python 3. No third-party packages (stdlib http.server only).

② CLI — batch-process CSV files directly (see Usage below).

What it does

For every target color in your spreadsheet:

1. Normalize the target to 8-bit sRGB (from hex / RGB / Lab / CMYK).
2. Run the FS reverse-match search → best (filamentA, filamentB[, C], ratios) recipe + its predicted color.
3. Run the Prusa reverse-match search → best recipe + predicted color.
4. Emit both recipes (hex, RGB, Lab, CMYK + ΔE2000) into one output CSV row.

Both algorithms use the same ΔE2000 yardstick (prusa_fdm_mixer), so their ΔE values are directly comparable — this is the same scoring the slicer uses internally.

Build

Requires only a C++17 compiler. No cmake needed.

Windows (MSVC) — run from a x64 Native Tools Command Prompt for VS:

build.bat

Linux / macOS / MinGW:

./build.sh

Or with cmake:

cmake -B build -DCMAKE_BUILD_TYPE=Release
cmake --build build --config Release

Output: build/color_match_batch[.exe].

Usage

color_match_batch \
  --palette  samples/palette.csv \
  --targets  samples/targets_hex.csv \
  --output   results.csv \
  --input-format hex

Required flags

Flag	Meaning
--palette <file>	CSV of physical filament colors (one #hex per line, optional ,label)
--targets <file>	CSV of target colors (first column = label, rest = color components)
--output <file>	Where to write the results CSV
--input-format <fmt>	hex / rgb / lab / cmyk — applies to the whole targets table

Optional flags

Flag	Default	Meaning
--min-percent <n>	0	Min per-component percent (0–50); higher excludes tiny admixtures
--exclude "1-2,3-4"	—	1-based palette-id pairs to skip (incompatible materials)
--rgb-scale auto|255|1	auto	How to read numeric RGB columns

Input CSV formats

palette.csv — one color per row, #hex first (or label,#hex):

#FF0000,Red
#00FF00,Green

targets.csv — first column is always a label; the meaning of the remaining columns depends on --input-format:

Format	Columns (after label)	Example row
hex	#RRGGBB	orange,#FF8800
rgb	R,G,B	orange,255,136,0 (auto-detects 0-255 vs 0-1)
lab	L,a,b	mid_gray,50,0,0
cmyk	C,M,Y,K	pure_red,0,100,100,0 (percent; auto-handles 0-100 vs 0-1)

A leading header row (e.g. label,color) is auto-detected and skipped.

Output CSV columns

Each row is one target; columns (in order):

label,
target_hex, target_R, target_G, target_B, target_L, target_a, target_b,
target_C, target_M, target_Y, target_K,
fs_type,    fs_ids,    fs_weights,    fs_hex,    fs_R,    fs_G,    fs_B,
fs_L,    fs_a,    fs_b,    fs_C,    fs_M,    fs_Y,    fs_K,    fs_delta_e,
prusa_type, prusa_ids, prusa_weights, prusa_hex, prusa_R, prusa_G, prusa_B,
prusa_L, prusa_a, prusa_b, prusa_C, prusa_M, prusa_Y, prusa_K, prusa_delta_e

• fs_type / prusa_type: pair (2 filaments) or triple (3 filaments).
• fs_ids / prusa_ids: /-joined 1-based palette indices (e.g. 1/3).
• fs_weights / prusa_weights: /-joined integer percents summing to 100 (pair: pctA/pctB; triple: wA/wB/wC).
• Color columns repeat for both target and each recipe's preview color.

Open results.csv directly in Excel, or paste-special it next to your source sheet.

Excel workflow

1. Put your filament palette on one sheet → Save As → CSV → palette.csv.
2. Put your target colors on another sheet → Save As → CSV → targets.csv.
3. Run the tool (one format at a time).
4. Data → From Text/CSV → import results.csv onto a new sheet.

Algorithm notes & caveats

• FS triple blends are order-sensitive. The FS polynomial has no native N-color form, so triples are folded left-to-right in filament-ID-ascending order (matching the slicer's blend_color_multi). Results are deterministic but would differ with a different fold order — this is inherent to FS.
• Prusa handles N colors natively via its Yule-Nielsen weighted average; no folding artifact.
• Lab/CMYK targets lose precision when normalized to 8-bit sRGB (both algorithms operate on uint8 channels). ΔE from a Lab target to its best match therefore includes this quantization floor.
• CMYK uses simple device conversion (R = 255·(1-C)·(1-K)), not ICC profile-aware. Fine for relative comparison, not for print-accurate prediction.

Repository layout

color-mixer-batch/
├── CMakeLists.txt        optional cmake build
├── build.bat             one-shot MSVC build
├── build.sh              one-shot g++/clang build
├── README.md             this file
├── src/                  tool sources
│   ├── main.cpp          CLI + CSV I/O
│   ├── match_search.{h,cpp}  reverse-match search (port of slicer logic)
│   └── color_io.{h,cpp}      hex/rgb/lab/cmyk parsing + formatting
├── vendor/               vendored algorithm files (see PROVENANCE.md)
│   ├── filament_mixer.h / .cpp / filament_mixer_model.h
│   ├── prusa_fdm_mixer.hpp / .cpp
│   └── PROVENANCE.md
└── samples/              example CSVs
    ├── palette.csv
    ├── targets_hex.csv
    ├── targets_rgb.csv
    ├── targets_lab.csv
    └── targets_cmyk.csv

Web UI (web/)

web/
├── app.py                stdlib-only HTTP server; spawns CLI, returns JSON
├── start.sh / start.bat  build-if-needed + launch
└── static/
    └── index.html        single-page UI (no framework, vanilla JS)

The server writes temp CSVs, invokes build/color_match_batch, parses the 42-column result back into JSON, and the page renders color swatches + recipe tables. Every edit triggers a debounced recompute (400ms) — the CLI runs in <100ms for typical palettes, so it feels instant. The page auto-loads demo palette/targets on first visit; edit freely or paste your own.