📊 PIP 高级颜色分析

You've got a render and you want text (or a LoRA, or a background) that matches its palette. Eyeballing hex values from a preview never goes well. PIPAdvancedColorAnalyzer, from the PIP_ArtisticWords pack, does the boring sampling work for you: it pulls the dominant colors out of any image, reports average brightness and saturation, and draws you a little analysis chart you can actually look at.

What it does

Feed it any IMAGE, and it returns the dominant color and a secondary color as hex strings (#rrggbb), the image's average brightness and saturation as floats (0–1-ish), and a generated IMAGE - an analysis panel showing swatches with their percentages. It's a read-only utility; nothing gets written to disk.

How it works

The interesting part is the sampling strategy. Instead of scanning every pixel (slow on 4K images), it auto-computes a sample count from the image dimensions - roughly sqrt(total_pixels) * 1.5, clamped between 2000 and 50000 - then samples on a uniform grid. Colors get bucketed with a Counter, the top bucket is the 主导色, and a secondary color is picked that's sufficiently distinct from it, so you don't get "dark gray" as the "different" second color. There are two hidden inputs in the code (auto_sample and sample_points) but they're not exposed in the node UI - the defaults are sensible, and you can't really break them.

The inputs and outputs that matter

Inputs are minimal, and that's a feature:

• image - the IMAGE to analyze (required).
• num_dominant_colors - how many dominant colors to consider (2–20, default 8). It mainly affects how good a distinct secondary color the node can find, so leave it near the default unless you're chasing a very specific swatch.

Outputs (Chinese labels, as this pack likes):

• 主导色 (STRING) - the main color as a hex string.
• 辅助色 (STRING) - a distinct secondary color.
• 平均亮度 (FLOAT) - average brightness.
• 平均饱和度 (FLOAT) - average saturation.
• 分析图 (IMAGE) - a rendered analysis panel with color swatches and percentages. Preview it or save it; it's the only visual output.

Why you'd wire it in

The honest use is pairing it with the pack's own color pipeline. Feed the 主导色 hex into PIPColorWheel to build a scheme, or match against the color_match logic of the PIP Artistic Text Generator. Beyond that it's a general palette-extraction tool - the analysis image alone is handy documentation of what a frame is actually made of.

Installing it

Part of the same pack, installed once:

cd ComfyUI/custom_nodes
git clone https://github.com/chenpipi0807/PIP_ArtisticWords.git
cd PIP_ArtisticWords
pip install -r requirements.txt

Or search "PIP Artistic Words" in ComfyUI Manager. Restart afterwards. Deps are just Pillow, numpy, opencv-python and webcolors - no models, no GPU.

Gotchas

• It prints a lot. On every run it logs the computed sample count, grid size, and strategy to the console. Fine for development, mildly annoying when it's just sitting in a workflow.
• "Dominant" means most pixels, not most important. A big flat sky will dominate over a small vivid subject. That's an inherent limit of bucket-counting, not a bug.
• Hex strings are lowercase. #4096ff, not #4096FF. Most color inputs (including the pack's own) accept both, but if you're string-comparing outputs somewhere, normalize case first.
• The analysis image is generated by Pillow and uses the pack's bundled 方正风雅宋 font for labels - if that font file ever gets deleted, it silently falls back to a default bitmap font.