Frequently Asked Questions

PhotoSort reads your photos locally in your browser, analyzes them with the CLIP AI model, and groups them into sessions based on time and location. Each session becomes a named folder arranged in a Year/Month-EventName structure.

For example:

• 2025/08 - Rome Trip
• 2025/11 - Wedding Munich
• 2025/12 - Christmas at Home

Your original photos are never moved — PhotoSort creates copies in your chosen output folder.

A session is a group of photos taken within a continuous 4-hour window. If the gap between two consecutive photos (sorted by capture time) exceeds 4 hours, PhotoSort starts a new session.

Sessions map naturally to real-world activities: a birthday party, an afternoon hike, a day sightseeing in a new city. Each session becomes a separate named folder in your output.

A typical library of 5,000 photos might contain 80–200 distinct sessions depending on shooting habits.

During setup, you enter your home city. PhotoSort reads GPS coordinates from the EXIF metadata in your photos and uses reverse geocoding (via OpenStreetMap) to determine the location of each session.

Sessions that took place outside your home city area are classified as trips. Sessions in your home city area are classified as local events. This distinction shapes both the folder labels and the human-readable names generated by Gemini AI.

Photos without GPS data are treated as local events by default, unless other signals (like airport content tags) suggest travel.

PhotoSort uses the CLIP model to detect 15 content categories:

• Pets and animals
• Selfies and portraits
• Food and restaurants
• Nightlife and bars
• Concerts and live music
• Nature and hiking
• Beach and water activities
• Skiing and snow sports
• City sightseeing and landmarks
• Parties and celebrations
• Weddings and formal events
• Shopping
• Home and indoor life
• Airports and travel hubs
• Sports and fitness

These tags feed into the Gemini naming system to produce descriptive, human-readable folder names.

Yes. Date-based sorting is the foundation of how PhotoSort works. It reads the EXIF DateTimeOriginal field from every photo to get the exact capture date and time. Photos are organized chronologically into Year/Month folders first, and then further grouped into named events within each month.

If EXIF data is missing, PhotoSort falls back to the file's last-modified date so no photo gets skipped.

You can rename any folder in the preview before confirming the sort. The default structure is Year/Month-EventName. Custom structure templates are planned for a future update.

All Gemini-generated folder names can be edited inline in the preview screen before any files are copied.

If the Gemini API is unavailable or the request times out, PhotoSort automatically uses a structured fallback name based on the session's geocoded location and top content tags. A beach session in Mallorca in July would become 2025-07 - Beach Mallorca. The fallback runs silently — you still get a clean, descriptive folder name without any error.

PhotoSort skips copying a file if a file with the same name and size already exists in the output location. This prevents duplication when you run PhotoSort multiple times on the same library. Full duplicate detection by image hash (which would catch renamed duplicates) is planned for a future update.

Videos (MP4, MOV, AVI, MKV, M4V) are sorted alongside photos into the same session-based folder structure using their file date metadata. PhotoSort does not run CLIP visual analysis on video frames, but videos are placed in the correct event folder based on when they were recorded — alongside the photos from the same session.

No. PhotoSort's AI (CLIP) runs entirely inside your browser using WebAssembly. Your photos are read from your local disk and processed in memory — they are never sent to any server.

The only external network requests are:

• OpenStreetMap Nominatim: GPS coordinates only (no photos, no filenames)
• Gemini API: Short text descriptions only (date, location string, content tags — no photos)

Your image files never leave your computer.

Never. PhotoSort is completely non-destructive. It creates copies of your photos in the output folder you select. Your source files are not touched in any way — not moved, not renamed, not modified, not deleted.

If you don't like the result, delete the output folder and run PhotoSort again with different settings. Your originals are always safe.

PhotoSort requires Google Chrome (version 86+) or Microsoft Edge (version 86+). These are the only browsers that currently support the File System Access API that PhotoSort uses to access your photos locally without uploading them.

Firefox and Safari do not support the File System Access API as of 2026 and are not compatible with PhotoSort. Support will be added automatically once these browsers implement the API.

PhotoSort relies on the File System Access API to read your entire photo library from disk and write the organized output — all without uploading anything to a server. This API was designed specifically to enable powerful local file operations in web apps.

As of 2026, only Chromium-based browsers (Chrome and Edge) implement this API. Once Firefox and Safari add support, PhotoSort will work in those browsers automatically.

Yes. PhotoSort is a web application that runs in the browser, so it works on any operating system that can run Chrome or Edge — including Windows 10/11, macOS, and Linux. There is nothing to install or update beyond your browser.

Images: JPEG, PNG, HEIC, WebP, TIFF

Videos: MP4, MOV, AVI, MKV, M4V

RAW formats (CR2, ARW, NEF, DNG, ORF) are not currently supported but are on the roadmap.

There is no hard limit. PhotoSort has been tested on libraries of 30,000+ photos. Processing time scales roughly linearly:

• 1,000 photos: ~5–15 minutes
• 5,000 photos: ~25–75 minutes
• 10,000 photos: ~50–150 minutes

Speed depends on your CPU, whether GPU acceleration is available, and your disk's read speed. The actual file copying after the AI analysis completes is much faster.

Yes. Any drive that appears as a local drive in your operating system can be accessed through the file picker. This includes external USB drives, Thunderbolt drives, and network-attached storage (NAS) devices mounted as local drives. The processing still happens locally in your browser — only the file reading and writing happen on the external drive.

The CLIP model (Xenova/clip-vit-base-patch32) is a neural network with over 150 million parameters. It was trained on hundreds of millions of image-text pairs and understands the visual content of photos well enough to classify them into dozens of categories without any task-specific fine-tuning.

Even in its compressed, quantized ONNX form optimized for browsers, the model weighs approximately 350MB. This is a one-time download. After the first use, the model is permanently cached in your browser's storage and PhotoSort works completely offline for AI analysis.

It is a one-time fee of €9.99. There are no monthly or annual charges, no renewal reminders, and no features locked behind a higher tier. Pay once and use PhotoSort forever, including all future updates.

PhotoSort lets you run the full AI analysis and preview the complete proposed folder structure before paying. The 7-stage AI analysis runs for free. You only need to purchase to confirm the sort and have files copied into the organized structure.

This means you can see exactly what PhotoSort will do with your photos before you spend a cent.

Yes. The full AI pipeline — CLIP analysis, session clustering, trip detection, geocoding, and Gemini naming — runs in your browser before you need to purchase. You can see the complete proposed folder structure, rename folders, and review which photos go where. Purchase is required only to copy files into the organized output.

If you're unhappy with PhotoSort after purchase, contact us within 30 days and we'll issue a full refund — no questions asked. Since you can preview the full output before purchasing, refund requests are rare, but we honor them without hesitation.