Small curiosities, with room to become something.

A single average can describe a color that barely appears in the image. A single dominant swatch can lose the small detail that makes the photograph interesting. The early coarse bins needed to become a richer description of the actual pixels.

The plugin makes a small analysis copy, preserving the full composition and aspect ratio. Its longest side is bounded at 96 pixels. Nearby RGB samples are grouped, but each group keeps a real sampled pixel close to its mean as its representative. That distinction matters: the palette should show colors that came from the photograph.

Each image retains up to 256 measured color groups, two kinds of weight, and a spatial summary. Coverage describes how much of the image a color occupies. Attention adds an estimate of visual importance. The spatial summary keeps enough information to revisit a selected subject area without resampling the full photograph.

Every photograph contributes the same total weight to the collection display. A larger upload doesn’t get a louder voice simply because it contains more pixels.

A pointer position is converted into a hue from 0 to 359 degrees and a saturation from neutral to fully saturated. Circular displays use angle and distance from the center. Rectangular displays use horizontal and vertical position. All four views feed the same matcher.

Pointer capture keeps a drag connected even when a finger or mouse moves outside the canvas. Fine controls provide separate hue and saturation sliders for keyboard use and precise adjustment. Quick swatches provide another way in.

The field represents the current collection, including its metadata filters and text search. It stays still while someone changes the color, sorts results, or moves between pages. Otherwise, selecting a color would erase the surrounding colors and move the target underneath the pointer.

The RGB and HEX readout belongs to the selected search color. The picker uses a normalized brightness for that reference. The viewer’s palette swatches are a different thing: measured colors from the actual photograph, including their lightness.

The earlier wide-range behavior loosened two things at once: color similarity and the amount of matching color an image needed. At larger settings, the results could feel arbitrary. A predominantly red image shouldn’t become a green match just because the range is wide.

I separated tolerance from presence. The range now broadens a bounded neighborhood of related hues and saturation levels. It doesn’t lower the minimum presence rule. A candidate still needs enough matching color, a sufficient score, and a meaningful share of its chromatic palette.

Hue comparison uses both HSV hue and perceptual Oklab hue. That helps with natural olive and green neighbors that can be awkwardly spaced in HSV. Neutrals follow their own rule, instead of borrowing an arbitrary hue from almost gray pixels.

I kept two matching preferences. Similar palettes uses the image’s color coverage. Subject-aware palettes combines coverage with an attention estimate. The automatic estimate looks at contrast against the image border, position within the frame, and the amount of color present. It does not identify objects or read intent.

That estimate helps, but a bright sign could be more noticeable than the intended subject. So the image editor also lets the owner choose a subject area. Draw a box, redraw it, adjust its percentages with the keyboard, or return to Automatic.

The selected rectangle is stored as relative coordinates. The matcher intersects it with the spatial color cells, maps those cells back to the measured color groups, and builds a new attention distribution. It doesn’t crop the image or throw away its surrounding palette.

Subject-aware matching blends 35% coverage with 65% attention. The overall photograph still contributes. Manual focus guides that balance when the automatic estimate misses.

The first use case was camera and film stock. One photographer needed “Camera used for this shot” paired with “Nikon F3,” then “Film stock used for this shot” paired with “Kodak Portra 400.” Hard-coding Camera and Film would have solved only that one use case.

The editor instead gives each gallery three optional detail fields. The label belongs to the gallery. The value belongs to the individual image. A field-level switch decides whether the values also become a filter. Filled details still appear in the image viewer when filtering is off, while empty details disappear.

That same structure can describe an artist, material, location, client, project, season, or collection. It is open to the owner’s categories within three reusable fields, rather than promising a separate feature for every possible archive.

Previously entered values become suggestions. Bulk editing applies shared details to several images, while leaving unchecked fields alone. A recorded camera can be suggested when that metadata is available. Labels and switches don’t have to be re-entered for every photograph.

The field places each weighted color at fixed hue and saturation coordinates, then spreads its influence through a Gaussian density calculation. Overlapping contributions build concentration. Contour bands make those differences visible without drawing every individual sample.

Hue wraps around the edge of the map, so red doesn’t become two unrelated regions at opposite ends. The bandwidth stays fixed. This keeps the relative color contributions meaningful while the collection determines the shape.

Once the contour structure worked, it still looked too transparent on white and too dim on black. The 1.5.3 refinement raised the six bands to 68%, 78%, 87%, 94%, 98%, and 100% opacity. The density map, coordinates, and underlying weights stayed the same.

Light and dark backgrounds are remembered independently for each display. The goal was a clear instrument in either setting, without muting the photographs to make the interface feel restrained.

The editor developed alongside the front end. Six sections separate the starting design, images, layout and loading, color filter, image viewer, and style. Visual presets make the first choice concrete. A live preview connects every adjustment to its result.

The less common settings sit behind specific disclosures rather than filling the screen at once. Image details live with the images. Shared labels, bulk values, and subject focus use the same editing context because they all describe how a photograph participates in the gallery.

On smaller screens, the editor gives settings and preview their own space. The Save action stays reachable, inputs have room for readable text, and the image-detail dialog uses the available height. Reordering also has explicit arrows, so dragging is optional.

Filtering to zero results exposed a sideways jump. It was measurable: the desktop scrollbar disappeared, adding 15 pixels to the viewport and moving the centered scope 7.5 pixels. Reserving that space and stabilizing short result sets addressed the cause.

Rapid selections also needed a different rhythm from a single change. The control and readout respond immediately. Incoming thumbnails get a short opportunity to prepare. Retained photographs ease into new positions, outgoing images leave once, and newer selections replace work that has become stale.

Horizontal galleries needed to retain the position after a drag. Arrow navigation now aligns to actual image boundaries rather than assuming every card has the same width. A movement threshold separates a deliberate drag from opening a photograph.

Input hints follow the way someone is interacting. Touch, mouse, and keyboard get appropriate guidance. Native page scrolling remains available outside the color control and subject-selection area. Reduced-motion preferences remove nonessential movement.

The WordPress version also separates cached page markup from the current public gallery manifest. Saving a gallery publishes its current settings and images; the renderer revalidates them instead of trusting an old cached embed. The public manifest keeps internal editing data out of its response.

Here, visitors have their own saved browser gallery. They can change settings, add images, adjust subject focus, rename detail fields, enable filters, and save the result back to the working gallery on this page. The editor and front end use the same renderer and matching logic.