DP Image Color Analyzer Small

You've got a reference image and you want to know what's actually going on with its colors - not just eyeball it, but get a palette and, ideally, a description you can drop straight into a prompt. That's the whole job of this node. Point it at an image, tell it how many colors you want pulled out, and it hands back a swatch strip plus a text summary of the color theme.

It's the stripped-down sibling of the pack's full DP Image Color Analyzer - same core idea, fewer knobs. If you just need "give me the dominant colors and a description," this is the one to reach for instead of the bigger node.

How it works

Under the hood it's dominant-color extraction: the node looks at the pixels in your image and clusters them down to the N most representative colors, the same general technique behind any palette-generator tool. Where it goes further than a plain swatch grid is the second output - a string describing the color theme, which is meant to be usable as prompt language rather than just a label. That's genuinely handy if you're trying to match a mood board or a reference photo's palette in a generation: instead of squinting at a swatch and guessing "dusty rose and sage, I guess," you get a phrase you can paste straight into a positive prompt or a note node.

Inputs and outputs

Only two inputs, both required:

• image - the image to analyze.
• num_colors - how many dominant colors to extract, 1 to 12, default 5. Five is a reasonable starting point; push it up if the image is busy and five colors feel like they're averaging away real detail, pull it down to 2-3 if you want a punchy, simplified read (great for logo or flat-illustration palettes).

Two outputs:

• color_images - an image showing the extracted colors (a swatch strip). Preview it directly, or feed it downstream if you want the palette itself as a visual asset.
• color_theme - a string describing the palette. Wire this into a text-preview node to read it, or splice it into a CLIPTextEncode prompt if you want a generation to lean into the same color mood as your reference.

How to install it

Search ComfyUI-Desert-Pixel-Nodes (or "Desert Pixel") in the ComfyUI Manager and install, or do it by hand:

cd ComfyUI/custom_nodes
git clone https://github.com/DesertPixelAi/ComfyUI-Desert-Pixel-Nodes

Restart ComfyUI. The pack's README doesn't list any node-specific Python dependencies beyond a standard ComfyUI environment (Python 3.8+, PyTorch with CUDA) - this is a pure image-processing node, no model downloads involved. After restart you'll find it under the DP category in the node browser, alongside the rest of the pack's ~105 nodes. One thing worth knowing about the pack itself: it also installs a universal node-theming system (right-click any node → "DP Color Themes") that applies to every node in your graph, not just DP ones - a nice-to-have if you like color-coding a busy workflow, easy to ignore if you don't.

Common issues & troubleshooting

The color_theme string comes back vague or generic. Palette-description tools are only as good as the clustering - a photo with lots of gradient/lighting variation will average toward muddier, less "nameable" colors than a flat illustration or a color-graded still. If the description feels off, try lowering num_colors so the node isn't forcing distinctions between colors that are barely different in the source image.

This is a small, low-traffic pack. ComfyUI-Desert-Pixel-Nodes doesn't show up much in community troubleshooting threads - it's a grab-bag utility pack rather than one with an active support community, so if something behaves unexpectedly you're mostly reading the node's own behavior rather than finding a matching Reddit thread. Worth knowing going in: you're closer to "figure it out from the graph" than "search and find someone else's fix."

Wrong node for the job. If you actually want every individual color swatch as separate outputs, or the fuller theme-detection/SD-friendly-description feature set the README mentions, check the full DP Image Color Analyzer (non-Small) - this node is deliberately the pared-down version.