Annual Photo Album & Calendar Creator

Local-first desktop app that turns a folder of a year's photos into a printable photo book and a matching wall calendar. Indexes your source folder, scores each photo with on-device computer vision (blur, faces, scene tags, exposure), auto-assembles pages you can review and tweak, and exports both as PDFs ready for a print shop.

All photo data and analysis stays on your machine — nothing leaves the computer.

Features

Library

• Walks any source folder; extracts EXIF (dates, GPS, camera, lens) and writes a thumbnail for each photo into local SQLite.
• HEIC, JPEG, PNG, WebP. Cross-platform via libheif on macOS / built-in on Win.
• On-device CV pipeline: Laplacian blur (focused on face regions when present), YuNet face detection + SFace identity embeddings, OpenCLIP scene tagging, exposure scoring, pHash for duplicate detection.

Album

• Auto-selects the best photos per day (top-N by aggregate score) up to a per-month cap, then packs them into pages using a varied template rotation (hero, side-by-side, asymmetric trio, quad grid, six grid + mirrored variants) with a user-configurable max page count.
• Per-page template picker as a schematic-icon popover; reorder, regenerate, delete, swap two adjacent photos via mid-edge buttons, insert blank pages, adjust crop in-place with brightness / contrast / saturation sliders, face-aware default cropping.
• Drag-to-reorder sorter view shows every page as a thumbnail grid.
• Place text overlays anywhere on a page; ~20 curated Google Fonts.

Calendar

• 12 monthly pages, auto-populated from the year's photos with one representative photo per month (or several per page via the per-page template — side, pair, trio variants on top or bottom).
• Calendar grid with configurable rule style (boxed / spreadsheet-grid / lines / none), text + grid color, separate weekend (Sunday) color, Google Font for the grid, Mon/Sun week-start.
• Date numbers vertically centered under their column headers, scaled with page width so previews and exports render proportionally.

Style + paper

• Per-output paper size from a 14-preset catalog (A4/A4-portrait, US Letter, 20/21/28/30 cm squares, 8/10/12" squares, 30×20cm + 28×21cm landscape, 21×28cm + 20×28cm portrait). Album and calendar can have different sizes.
• Page bg color, slot gap, page margin, slot corner radius — separate values for album and calendar where it matters.
• Four built-in style presets (Minimal / Classic / Polaroid / Modern) that bundle gap / padding / corner / bg / calendar font / calendar color.

Export

• Per-output PDF export at a chosen target DPI (170 / 255 / 340). The scale multiplier is computed at capture time from the chosen DPI + paper size, so the label is honest regardless of preview thumbnail size.
• One PDF per output, written to a user-chosen path via the native save dialog.

Tech stack

Tauri v2 shell · SvelteKit (Svelte 5) renderer · SQLite via tauri-plugin-sql · Node sidecar (Sharp + exiftool-vendored) · Python sidecar (FastAPI + OpenCV

• OpenCLIP + imagehash) · jsPDF + modern-screenshot for PDF capture.

Development

Requires Node ≥ 20, Rust toolchain (Tauri v2), Python ≥ 3.11, uv, and libheif for HEIC support (brew install libheif on macOS; bundled in the Windows distribution).

npm install
cd sidecar && npm install && cd ..
cd py-sidecar && uv sync && cd ..
npm run tauri dev

First run note: the first time the Python sidecar starts after uv sync, OpenCLIP downloads the ViT-B/32 weights (150 MB) to `/.cache/torch/hub/checkpoints/. The first /embedor/tags` call takes a couple of minutes; subsequent runs are fast.

Tests

cd sidecar && npm test            # Node sidecar (Vitest)
cd py-sidecar && uv run pytest    # Python sidecar
cd src-tauri && cargo test        # Rust
npx playwright test               # UI smoke (non-Tauri)

Fixture generator

tools/fixtures/ produces a synthetic family-of-four photo set via ComfyUI (FLUX.1-dev) + ExifTool, for testing the index/CV/selection pipeline against realistic-but-controllable data. See tools/fixtures/README.md.

Architecture

See docs/superpowers/specs/2026-05-14-family-album-builder-design.md for the original design doc. The app has grown beyond that spec — the source is now the truth — but the high-level shape is unchanged: Tauri shell, SvelteKit UI, SQLite for project + photo state, two sidecars (Node for fast EXIF/Sharp work, Python for CV models that need PyTorch/OpenCV).

All photo data and ML inference is local. The Google Fonts CSS is the only outbound request, and only when a non-default font is used.