Your trip, ready to print: the date and place on each photo, the lab's sizes, nothing uploaded.
Free · offline · open source · macOS, Windows and Linux
⬇ Download for macOS, Windows or Linux
Website · Manual · Releases · English · Tiếng Việt
You shot 800 photos on a trip and want twenty of them on paper. Getting there today is tedious:
- Prints don't say when or where. In a few years, nobody remembers.
- Labs auto-crop your photos to 2:3, 3:4 or 1:1, and that can cut off heads.
- Online tools upload your photos. GPS data reveals where you live.
Fotollection does all of this on your computer: 1,000 photos open in 0.12 s, an export runs at 18 photos a second, and the print keeps your camera's detail and colour:
Against a web editor's usual export (simulated with its standard settings: the colour profile dropped, 2000 px, 4:2:0, quality 80), measured on an Apple M3; the method and every number are under Performance.
- Open the trip's folder. JPEG, HEIC from the iPhone or camera RAW, thousands at once; best shots are suggested.
- Keep the ones to print. The date and place go on each photo in a film-camera face, the crop fits the lab's size, a look gives the colour.
- Export for the lab at full quality, GPS removed, and take the files to the kiosk: dm, Rossmann, CEWE, or 7-Eleven, Lawson and FamilyMart in Japan.
Finished prints as Fotollection exports them. Photos of Prague © @hoangphi01.
| 📅 Date and place on the print | from the photo's EXIF, in a film-camera LCD face, places found offline |
| ⏱️ Camera and phone together | a camera's wrong clock fixed from one phone photo of the same moment, and the place taken from your phone's photo of the same scene |
| ✂️ The lab's own sizes | 21 print sizes of Europe's, Japan's and Korea's shops and kiosks, your own sizes, or the photo's original shape, with a 3 mm border so the kiosk's trim cuts nothing |
| ✨ The best shots, suggested | bursts and similar shots grouped, the sharpest with eyes open suggested with its reasons; stars, flags and filters, Lightroom's keys |
| 📷 Every kind of photo | JPEG, PNG, HEIC from the iPhone, and camera RAW: Fujifilm RAF, Panasonic, Canon, Nikon, Sony, Olympus and DNG all work today |
| 🌈 Looks, presets and LUTs | one tap gives a photo its colour; your own Lightroom presets and LUMIX Lab or .cube LUTs work too; colour tools as Lightroom's |
| ✉️ Postcards | a photo on one side, the address side on the back, typed or written by hand; the Japanese はがき and the Korean 우편엽서, the address filled in from its postal code (AddAssist) |
| 🎚️ Three modes | Simple (keep or drop, then print), Collage (two photos on one print, or postcards) and Pro (every control) |
| 🔒 Private by design | no account, no cloud, no telemetry; GPS removed on export; the only network request is the update check, which you can turn off |
New in v0.14
- v0.14.3: an address book for postcards, a sender on each card, and looks, LUTs and presets in Collage mode.
- v0.14.2: the app in 日本語 and 한국어, the はがき and the 우편엽서 with the address filled in from its postal code, Japan's and Korea's print sizes.
- v0.14.1: grain and vignette in millimetres on the print, highlight recovery for blown skies in RAW, the app in Deutsch and Magyar.
Interface in English, Tiếng Việt, Deutsch, Magyar, 日本語 and 한국어. Every step with pictures: the manual. Every version: CHANGELOG.md. Every requirement: docs/PRD.md.
Fotollection works end to end today; v1.0.0 in December 2026 is its promise of stability. Two releases come first: v0.14.4 in November, which sorts a phone's whole camera roll (camera shots, screenshots, saved images, Live Photos, duplicates, events), and v0.14.5 with more looks; after it, one release a month, each with its theme, and maintenance first, planned to 2028-11. Coming later: a face-aware crop, a photo-book preview and QR codes on the back of prints. Every requirement and its quarter: docs/ROADMAP.md; the checks left before v1.0.0: docs/STATUS.md.
Get the latest version from the Download page or the Releases page: a .dmg for macOS 11+ (Apple Silicon and Intel), a -setup.exe for Windows 10/11 (the .msi is for IT departments), and an AppImage or .deb for 64-bit Linux (which distributions). The app asks GitHub once at start whether a newer version exists (Settings → Updates turns it off) and never downloads anything by itself. Installing over an older one keeps your settings and projects.
Important
Builds are unsigned for now (ADR-0009): a certificate costs money a one-person open-source project does not have yet. The code is public, every build comes from GitHub's own CI, and the app goes online only to check for a new version. On first launch:
- macOS: System Settings → Privacy & Security → Open Anyway
- Windows: SmartScreen → More info → Run anyway
- Linux: nothing to approve; make the AppImage executable, or install the
.deb(below)
Verify each download against SHA256SUMS.txt on the release page. See the code signing and privacy policy.
The same app on every system, with these differences (ADR-0034):
| macOS | Windows | Linux | |
|---|---|---|---|
| Culling, stars, flags, best shot, print basket | ✓ | ✓ | ✓ |
| Print sizes, crop, straighten, zoom, date stamp, info frame | ✓ | ✓ | ✓ |
| Offline places, dates and places by hand, camera clocks | ✓ | ✓ | ✓ |
| Export for print and for social media | ✓ | ✓ | ✓ |
| Looks, Lightroom presets and LUTs | ✓ | ✓ | ✓ |
| JPEG and PNG | ✓ | ✓ | ✓ |
| Camera RAW | ✓ ¹ | ✓ | ✓ |
| HEIC (iPhone photos) | ✓ | ✓ ² | not yet ³ |
| Shoot with a camera over USB | not yet | ✓ ⁴ | — |
| "Open with" from the file manager | ✓ | ✓ | ✓ ⁵ |
| Check for updates downloads the new version | ✓ | ✓ | ✓ ⁶ |
- Every system develops Bayer RAW with Fotollection's own engine (Fujifilm X-Trans through rawler until v1.1); on macOS, Panasonic and Olympus RAW go to the system first, for its lens corrections, which Fotollection's engine does not make yet. On Windows, Microsoft's Raw Image Extension, when installed, is only a last fallback.
- With Microsoft's free HEIF Image Extensions and HEVC Video Extensions.
- HEIC needs an HEVC decoder, which Linux does not ship and Fotollection will not bundle (patents). A later version will use the
libheifyour distribution installs (FR-068). Until then, convert HEIC photos to JPEG. - Sony a6400 first (FR-063).
- With the
.deb. An AppImage is not registered with the file manager unless a tool such as AppImageLauncher does it. - The AppImage downloads the new AppImage; the
.debopens the release page.
Which distributions, and how to install
Works on 64-bit Linux with glibc 2.35 or later, x86_64 or ARM64:
- AppImage: Ubuntu 22.04+, Debian 12+, Linux Mint 21+, Pop!_OS 22.04+, elementary OS 7+, Zorin OS 17+, Fedora 36+, openSUSE Tumbleweed, Arch, Manjaro and others of that age.
.deb: Ubuntu, Debian, Mint, Pop!_OS, elementary OS and Zorin OS.
Works with a step:
- Ubuntu 24.04 and newer: the AppImage needs FUSE 2 once:
sudo apt install libfuse2t64. You can also run it with--appimage-extract-and-run. - NixOS:
appimage-run Fotollection_*.AppImage. - Fedora Silverblue and other immutable systems: use the AppImage.
- Nvidia graphics: Fotollection turns off WebKit's DMA-BUF renderer by itself, which avoids a blank window.
Does not run on:
- glibc older than 2.35: Ubuntu 20.04, Debian 11, RHEL, CentOS, Rocky or Alma 8, openSUSE Leap 15;
- musl systems such as Alpine;
- 32-bit machines;
- ChromeOS without its Linux container.
Install:
# One line for any distribution above: the signed APT repository on Ubuntu, Debian, Mint and
# their family (updates then come with sudo apt upgrade), the AppImage in ~/.local elsewhere
curl -fsSL https://fotollection.org/install.sh | sh
# Or the APT repository by hand
curl -fsSL https://fotollection.org/apt/fotollection.gpg | sudo tee /usr/share/keyrings/fotollection.gpg > /dev/null
echo "deb [signed-by=/usr/share/keyrings/fotollection.gpg] https://fotollection.org/apt stable main" | sudo tee /etc/apt/sources.list.d/fotollection.list
sudo apt update && sudo apt install fotollection
# AppImage (any distribution above), x86_64; use _aarch64 on ARM
chmod +x Fotollection_*_amd64.AppImage && ./Fotollection_*_amd64.AppImage
# .deb (Ubuntu, Debian, Mint…), from the folder you downloaded it to; _arm64.deb on ARM
sudo apt install ./Fotollection_*_amd64.debMinimum and recommended
| Minimum | Recommended | Notes | |
|---|---|---|---|
| Operating system | macOS 11 Big Sur, Windows 10 64-bit, or 64-bit Linux with glibc 2.35 (which distributions) | macOS 13 or later, Windows 11, or Ubuntu 24.04 | Apple Silicon and Intel Macs. Windows uses the WebView2 runtime, built into Windows 11 and installed automatically on Windows 10. Linux uses WebKitGTK 4.1, bundled in the AppImage and installed with the .deb |
| Processor | Any 64-bit CPU from the last ~8 years | Apple Silicon, or a recent 4+ core x64 | Work is spread over all cores. Every number in Performance is from an Apple M3 |
| Memory (RAM) | 4 GB | 8 GB or more | An export stays under 2 GB even for 5,000+ photos (measured peak 1.45 GB, NFR-004). Since v0.11.0, the preview keeps up to a quarter of the memory so photos open at once (FR-091) |
| Disk | 100 MB free plus the cache | 1 GB free | The installers are 33–118 MB, within the 300 MB budget (NFR-008). The cache takes about 25 MB per 1,000 photos (thumbnails about 22 KB and best-shot analysis about 4 KB each) and is safe to delete. Exports need space for the full-resolution files |
| Display | 1280 × 800 | 1920 × 1080 or a Retina display | The toolbar adapts to narrow windows |
| Internet | Not needed | — | Everything runs offline: faces, places and previews included. Nothing is uploaded |
| HEIC (iPhone photos) | macOS: built in | — | Windows: install HEIF Image Extensions and HEVC Video Extensions from the Microsoft Store (ADR-0017). Linux: not yet (ADR-0034) |
| Camera RAW | macOS: built in | — | Windows and Linux: built into Fotollection, nothing to install (ADR-0029). With Microsoft's Raw Image Extension installed, the system's codec is a last fallback on Windows (ADR-0023) |
Fotollection is built for enthusiast and professional photographers, so speed is a requirement, not a nice-to-have. Every number below is measured, not estimated, on an Apple M3 (8 cores, 16 GB) with real 16 MP JPEGs. The first measurements, of 2026-10-23 on v0.1, are kept as the baseline; each section names its date, and the measurements are run again every month against that baseline (ROADMAP). Each chart has a reference to read it against: the budget, the memory ceiling, or the browser. Full results: docs/RESULT.md.
| What | Budget | Measured | |
|---|---|---|---|
| Open a folder (import + EXIF + first thumbnails) | < 1 s per 1,000 photos | 0.12 s per 1,000 (5,035 photos in 0.62 s) | ✓ |
| Preview update after a settings change | < 50 ms (p95) | 18 ms p95, 17 ms p50 | ✓ |
| Export throughput | ≥ 8 photos/s at 24 MP | 18.0 photos/s at 16 MP | ✓ |
| Peak memory | ≤ 2 GB at any batch size | 1.45 GB with 5,035 photos | ✓ |
| Best-shot analysis, per photo (2026-10-24) | ≤ 20 ms, ≤ 40 ms with faces | 17 ms, 28 ms with faces | ✓ |
On a real trip (one folder of 20 MP Lumix GX9 photos, 422 files) the export reached only 7.6 photos/s with a 2.14 GB peak on 2026-11-03. The fix, in v0.9.7, exports its JPEGs at 10.1 photos/s, and the whole run peaks at 1.31 GB, within both budgets (RESULT.md §20).
Since v0.11.0 a preview of every photo is prepared in the background and kept on disk, so any photo opens at once, RAW too (FR-091). On one trip's 422 photos, each version measured from its own code:
Memory must not grow with the number of photos. A batch 53 times larger costs 0.1 GB more at peak:
| Photos | Peak memory | Export | Library in memory |
|---|---|---|---|
| 95 | 1.35 GB | 18.0 photos/s | — |
| 5,035 | 1.45 GB | 15.4 photos/s | 13 MB |
The 5,035-photo export runs for more than 5 minutes, and the M3 throttles under that load, which explains the lower throughput.
The obvious alternative is a web app that does the same work in the browser (createImageBitmap + OffscreenCanvas, one Web Worker per core). Both ran on the same 95 photos on the same machine, 2026-10-23, on v0.1 (to be measured again on the current version):
| Fotollection (Rust) | Browser (Chromium 156) | ||
|---|---|---|---|
| Batch export | 18.0 photos/s | 17.9 photos/s | About the same |
| Warm preview, 1600 px (p50) | 17.4 ms | 18.5 ms | Fotollection faster |
| Peak memory | 1.35 GB | 1.78 GB | Fotollection 24 % lower |
On raw export speed the browser is just as fast, because Chromium uses libjpeg-turbo with SIMD. Fotollection is now faster on previews. Speed alone is not the argument for a native app, though:
- The browser can only encode JPEG with 4:2:0 chroma, which halves colour resolution. Fotollection exports prints in 4:4:4, with twice the chroma data, in the same time.
- Fotollection uses 24 % less memory, and its export is bounded, as the large-batch test shows. A browser tab has a hard ceiling.
- Fotollection writes the files with atomic renames and never writes GPS, so where you live never leaves your machine. The browser only produced blobs and never wrote them.
- Colour profiles (ICC), camera RAW and HEIC are out of reach in the browser.
| Step | Time | Share |
|---|---|---|
| Crop + resize | 9.7 ms | 58 % |
| Date stamp | < 0.1 ms | ≈ 0 % |
| JPEG encode | 7.0 ms | 42 % |
| Total | 16.7 ms | budget 50 ms · Chromium 18.5 ms |
The encoder used to be the bottleneck: 16.2 ms, 62 % of a 26 ms preview. Previews are only ever shown on screen, so they now encode with 4:2:0 chroma and plain Huffman tables, which looks the same at 3× zoom. Print exports keep 4:4:4. That one change took the preview from 26 ms to 17 ms.
While a colour slider is dragged, the look is drawn by the graphics card from the same table Rust builds, so a step costs 1–2 ms instead of about 26, and the photo follows the slider; the date stamp, the frame and the border are left out, and Rust's render takes over when the drag ends (ΔE2000 to it: mean 0.13, p99 0.77). A machine without a capable GPU keeps the CPU path; the app decides by itself. Exports stay on the processor, so a print comes out the same on every computer (ADR-0039).
A photo is decoded and encoded on one core, so speed follows how many are worked on at once, within the memory budget. The app takes six: on a real trip folder with RAW files, eight was no faster and used 0.2 GB more.
Speed means nothing if the print comes out flat. Prints made through a web editor often look shallow, with flat, blotchy colour and less depth than the original. So Fotollection's export is checked pixel by pixel against the original (bench/quality/audit.py), next to a typical web-editor export.
| Measured against the original (median) | Fotollection | Typical web editor |
|---|---|---|
| Colour error, ΔE2000 mean (camera JPEGs; 1 is the threshold of a visible difference) | 0.19 | 1.39 |
| Colour error, 95th percentile | 1.06 | 3.50 |
| Fine detail kept (camera JPEGs) | 100 % | 13 % |
| Colour saturation kept (Display P3 photos, like an iPhone's) | 99.9 % | 86.8 % |
| Contrast and tonal levels kept | 100 % | 100 % |
| Pixel fidelity (PSNR) | 52.6 dB | 37.9 dB |
| Colour profile (ICC) kept | 20 / 20 | dropped |
Why a web-editor print looks flat, and what Fotollection does instead:
- The colour profile is dropped. Phone photos are in Display P3, a wider colour space than sRGB. Without the profile, the lab reads the colours as sRGB and they print about 13 % less saturated. Fotollection keeps the photo's own ICC profile in every export.
- The photo is shrunk, then blown up again. A 2000 px image printed at 10 × 15 cm loses 87 % of the fine detail, so surfaces turn smooth and colours look blotchy. Fotollection exports at full resolution, and only resizes, with a Lanczos3 filter, when a print size needs it.
- The colour is halved, and the JPEG is squeezed harder. Common editors use 4:2:0 chroma, which halves colour resolution, at quality 80. Fotollection exports prints at 4:4:4 and quality 90 by default, adjustable up to 100.
The test used 20 photos from a Canon compact, plus 5 of them converted to Display P3 with macOS's own profile. The web editor is simulated with standard settings (sRGB assumed, profile dropped, 4:2:0, q80, 2000 px, upscaled back for print). Reproduce it with cargo run --release -p fotollection-core --example quality_export -- <out> <photos…>, then uv run bench/quality/audit.py <photos> <out>.
The Share online export makes light sRGB files with the camera details, sized for each platform. Measured on the same photos, then recompressed the way Facebook does (RESULT.md §10):
| Median | Original size | Facebook, 2048 px |
|---|---|---|
| File size (camera JPEGs, 3.08 MB) | 1.42 MB | 0.51 MB |
| Similarity to the original (SSIM, 1 = identical) | 0.981 | 0.979 |
| Colour error, ΔE2000 mean | 0.96 | 1.27 |
| After Facebook: colour of Display P3 photos kept, against the usual way | — | 99.6 % against 86 % |
Speed is not enough for a feature that suggests which photo to keep, so best-shot suggestions are also scored for accuracy. 95 real photos (people, food, streets, night shots, sunsets) were labelled by hand: 22 groups of shots of one moment, and for 6 of them the photo that should be kept. eval_bestshot compares the app's groups and picks with those labels.
| Step | Groups found (of 22) | Pairs found | Wrong groups | Acceptable picks |
|---|---|---|---|---|
| dHash (before) | 3 | 9 % | 0 | 2/2 |
| + layout and colour signature | 13 | 67 % | 0 | 3/4 |
| + zoom from focal length | 15 | 73 % | 0 | 5/6 |
| + fair sharpness, hidden faces | 15 | 73 % | 0 | 6/6 |
- dHash (a 64-bit hash of brightness steps) flips its bits in flat areas such as sky. Two near-identical shots of a hotel scored as far apart as two unrelated scenes. It is replaced by a 32 × 24 layout signature compared by correlation over small shifts and zooms, plus a colour histogram that separates different scenes with a similar layout.
- Re-shooting closer is common, for example 5 → 9.9 mm on the church at night. Zooming no longer splits a group: the focal lengths in EXIF tell the comparison how much closer the second shot is.
- Fair comparisons. A shot with the face turned away used to be compared on whole-image sharpness against eye sharpness in the other shot. Now every shot in a group is measured the same way, and a face turned away or hidden counts against a shot.
- Still missed: 7 groups where the pose or the framing changed a lot, such as a second pizza shot from the other side or a selfie taken upside down. Those need an image-understanding model, which is a next step. The 2 extra groups the app made are pairs labelled as ambiguous, not mistakes.
The analysis stays within its budget: 17.4 ms per photo, 27.6 ms with faces (NFR-014). The labels hold file names only, since the photos are private: label your own folder in the same format (bench/bestshot/truth-103_09.json) to score it.
Reproduce everything with bench/run.sh <photo-folder> and bench/batch.sh <photo-folder>; score best-shot suggestions with cargo run --release -p fotollection-core --example eval_bestshot -- <photo-folder> <labels.json>. The method is in docs/OPTIMIZATION.md §2, and the budgets every change must respect are in §3.
git clone https://github.com/hoangphi01/fotollection.git
cd fotollection
scripts/bootstrap.sh --install # macOS / Linux (Windows: scripts\bootstrap.ps1 -Install)
npm run dev # start the appEvery tool version is pinned: rust-toolchain.toml, .node-version, Cargo.lock, package-lock.json and, for the Python tooling managed with uv, uv.lock. A fresh clone therefore builds the same on any machine. See docs/DEVELOPMENT.md for prerequisites, commands and build-disk cleanup (scripts/clean.sh).
To build from source you also need about 10 GB free for Rust build files (a debug and a release build; scripts/clean.sh reclaims them), Rust (installed by rustup, version pinned in rust-toolchain.toml), Node.js 22 or later, and Xcode Command Line Tools (macOS), the MSVC Build Tools (Windows), or WebKitGTK 4.1 and GTK 3 development packages (Linux, as scripts/bootstrap.sh installs them). The full list is in docs/DEVELOPMENT.md §1.3.
Why it is fast and safe: every pixel and file is handled in Rust, never in the webview; no unsafe code in the core (unsafe_code = "forbid", only the two crates that call the operating system opt out); a core with no networking at all, checked in CI; memory bounded by design (1.45 GB at 5,035 photos); typed calls between the interface and the core.
| Layer | Technology |
|---|---|
| Core (Rust) | Pure-Rust codecs: zune-jpeg with jpeg-decoder for rare layouts, jpeg-encoder, png (ADR-0012, ADR-0033); HEIC through the OS (ImageIO / WIC, ADR-0017); camera RAW through Fotollection's own reader and rawler, the system's codec first on macOS (ADR-0029, ADR-0030); colour management with qcms (FR-027); fast_image_resize, skrifa, rayon; GeoNames for offline places (ADR-0003) |
| Shell | Tauri 2: the system webview, no bundled Chromium |
| UI | React 19, TypeScript, Tailwind CSS 4, shadcn/ui, Lucide, react-markdown for the manual inside the app |
| Typed IPC | tauri-specta: TypeScript bindings generated from Rust |
Design: docs/ARCHITECTURE.md. Decisions: docs/adr/.
All docs live in docs/ and change together with the code (ADR-0006).
| Doc | What's inside |
|---|---|
| PRD | Requirements (FR-xxx, NFR-xxx) |
| Architecture | Crates, pipeline, IPC contract |
| Results | Latest measured speed and memory |
| Optimization | Budgets, methodology, benchmark log |
| File formats | .fotollection, .fotosettings, .fotopreset and every other file type |
| Style rules | Rust, TypeScript, UI, git and docs conventions |
| Security | Threat model, privacy, supply chain |
| Roadmap / Status | Quarters, milestones and progress |
| User guide | How to use the app, step by step; also the manual on the website |
| Error codes | Every ERR-… code the app can show: cause and what to do |
| Docs guide | What to update when something changes |
| Development / Testing / Release | Setting up, tests, how a version ships |
| Lessons | What each debugging session taught |
| Changelog | What changed in each version |
Contributions are welcome, from people and from AI agents. Please read CONTRIBUTING.md first: it explains where each fact lives, how to keep the docs in step with the code, how to read and add error codes, and how to open issues and pull requests. In short:
- Every change traces to a requirement ID (
FR-…,NFR-…), and IDs are never reused. - Docs are updated in the same commit, following docs/DOCS_GUIDE.md.
- A bug report brings the
ERR-…code and the app's Copy diagnostics text. - Pull requests are welcome; a contributor agreement (CLA, ADR-0008) is planned but not set up yet. Good places to start: issues labelled good first issue and docs/ARCHITECTURE.md.
Report security issues through a private GitHub security advisory, not a public issue.
Fotollection is free, made by a student between classes and a part-time job. If it saves you time, you can sponsor it on GitHub or buy me a coffee on Ko-fi, monthly or once, or, from Vietnam, scan the VietQR code on the website.
GPL-3.0-only © hoangphi01.
Fonts, data, models and libraries Fotollection bundles
- The UI tests (
scripts/ui-check.sh, never shipped) use axe-core by Deque Systems under the Mozilla Public License 2.0, a development tool only (ADR-0042). - JPEG encoding uses
jpeg-encoder, which is based in part on the work of the Independent JPEG Group. - The text inside the photo uses DSEG14 Classic Italic by keshikan, and the info frame uses Inter; both under the SIL Open Font License 1.1 (
crates/core/assets/fonts/). The interface uses Be Vietnam Pro and, for figures set mono, Geist Mono (both OFL 1.1), bundled so it looks the same on Windows, macOS and Linux; Japanese and Korean use Noto Sans JP and KR (OFL 1.1), cut to the interface's characters (ui/src/assets/fonts/,scripts/cjk-font.sh). - The website's support section draws its VietQR code with qrcode-generator by Kazuhiko Arase, MIT licence (
site/assets/js/qrcode.js, kept as published). QR Code is a registered trademark of DENSO WAVE INCORPORATED. - Postcard text uses Xanh Mono Regular and Italic by Yellow Type Foundry, and Patrick Hand by Patrick Wagesreiter, both under the SIL Open Font License 1.1 (
crates/core/assets/fonts/, sources and checksums inFONTS.md). The Japanese postcard's text uses Klee One SemiBold by Fontworks (the Klee Project Authors), SIL Open Font License 1.1, cut to JIS X 0208. The Korean postcard's uses Nanum Pen Script by NHN and Gaegu by the Gaegu Project Authors, both SIL Open Font License 1.1. - The welcome's language list and the website's language menu show round flags from circle-flags by HatScripts, MIT licence (
ui/src/assets/flags/,site/assets/flag/). - AddAssist's Japanese addresses come from Japan Post's postal-code data (郵便番号データ), whose copyright Japan Post does not claim, data of 2026-10-23, and Korea's from the Ministry of the Interior's road-name addresses (도로명주소, "no condition on use"), November 2025 (
crates/core/assets/addassist/). - Place names come from GeoNames (CC BY 4.0), bundled for offline geocoding: towns and cities with at least 1,000 people, snapshot 2026-10-24 (
crates/core/assets/geo/). - Trademarks. The camera-brand logos in the info frame (Apple, Canon, Fujifilm, Hasselblad, Leica, Lumix, Nikon, Noritsu, Phase One, Ricoh, Samsung, Sony) are trademarks of their owners. They are public-domain renderings from Wikimedia Commons (list and sources), used only to show which camera (or film scanner) took a photo. Fotollection is not affiliated with, sponsored or endorsed by any of these companies, and a logo is removed on request. Adobe and Lightroom are trademarks of Adobe Inc.; LUMIX is a trademark of Panasonic Holdings Corporation. Fotollection reads presets and LUTs made with their software but is not connected with or endorsed by either.
- Camera RAW on Windows (and on macOS for what ImageIO cannot read), except the formats Fotollection's own reader has taken over (DNG for now, ADR-0030), is decoded by rawler by Daniel Vogelbacher and contributors, under the GNU Lesser General Public License 2.1 (ADR-0029). Fotollection's complete source is published, so it can be rebuilt with a modified rawler.
- Best-shot suggestions bundle the YuNet face detector (MIT, © 2020 Shiqi Yu) and Google's MediaPipe FaceMesh V2 landmark model (Apache License 2.0), converted to ONNX. Sorting a camera roll bundles Microsoft's TinyCLIP ViT-39M image tower (MIT), its weights stored as int8. Licences and checksums:
crates/core/assets/models/.





