The Mood Board Maker is 100% client-side. Each tile is a plain typed object (an image reference, a colour swatch, a text note or a shape) held in the page, and a pure engine draws them onto an in-browser HTML5 canvas. When you upload a photo it is read straight into a local data URL and composited on-device — it never leaves your computer and is never sent anywhere. Sampling a colour draws a downscaled copy of your image to a detached off-screen canvas and reads its pixels locally. Export renders the same scene to a canvas and saves it with canvas.toBlob, and Save/Load stores the whole board (images embedded as data URLs) only in this browser’s localStorage. There is no analytics, no tracking and no CDN in the processing path; the only optional network use is consent-gated ads. The tool works offline once cached, and closing the tab discards anything you did not save or export.

Unlike cloud tools that bundle a huge stock library, this maker deliberately ships NO external stock imagery, because pulling real photography would require calling a third-party server and would break the “nothing leaves your device” promise. Instead you bring your own images: click “Upload image tile(s)” (you can add several at once) and each becomes a tile you can drag, resize, rotate, round the corners of, give a white polaroid frame, and crop-to-fill (cover) or fit (contain). The six starter templates seed a cohesive colour palette, a couple of text notes and empty image FRAMES laid out for you — you just drop your own photos into the frames. Local vector shapes and colour swatches let you build a rich board even before you add a single photo.

Colour sampling uses fast bucket quantisation. When you pick a selected photo and choose “Dominant colour” or “Extract palette”, the engine draws a downscaled copy of the image (about 48–64 px on the long edge) onto an off-screen canvas, reads the pixels, skips transparent ones, and groups the rest into 4-bit-per-channel buckets. It averages each bucket, orders them by how many pixels fell into each, de-duplicates colours that are perceptually close, and drops the top one (dominant) or top six (palette) onto the board as swatch tiles showing their hex codes. It is a quick approximation of the prominent colours rather than a designer’s hand-picked or perceptually-clustered palette, so treat the swatches as a starting point and nudge, recolour or delete them to taste. Everything runs locally — the image is never uploaded to compute the palette.

Because the same render routine draws the live preview and the export, the file you download matches the board exactly (true WYSIWYG). Pick a PNG width of Standard (1600 px), High (2400 px) or Ultra (3200 px) — the height follows your board shape — then download it or copy it straight to the clipboard. Save serialises the board to JSON (with each photo embedded as a data URL) into this browser’s localStorage, and Load restores it and re-decodes the images; this keeps one board per browser and never touches a server. Note the practical limits: localStorage quota is usually around 5 MB, so a board with several large photos may be too big to save (just export the PNG in that case), and canvas text renders with the fonts installed on the viewing device, so metrics can vary a hair between browsers.