Setup & Use

Setup & Use

This page walks new users through installing Video Duplicate Finder, the first launch, and a first scan.

Setup

1) FFmpeg & FFprobe

VDF requires ffmpeg and ffprobe. If they are missing, the program offers to download them on first launch. Alternatively, place them in the application directory or make them available on the PATH. If the automatic download fails, see the project readme for manual installation instructions.

2) Choose a database format (first launch only)

On first launch you are asked to choose a graybyte database format:

• 16x16 - fastest, but more false positives. The pHash algorithm is not available with this format.
• 32x32 (recommended) - slightly slower on the first scan, fewer false positives, and pHash is available.

This choice is permanent. You cannot switch formats later without deleting the entire database and starting over.

3) Start the program

• Windows: start VDF.GUI.exe.
• Linux/macOS: change to the VDF directory and run ./VDF.GUI (or dotnet VDF.GUI.dll).

The main screen

A scan is set up in two steps on the home screen:

Where to look

Add the folders you want to scan. You can also drag and drop folders onto the window (hold Alt while dropping to add them as excluded folders). Each folder shows a file count once it has been indexed; counting runs in the background and never delays scanning. Subfolders are included by default.

What to look for

Instead of tuning individual settings, you pick a scan profile. Each profile is a bundle of matching settings named after what you want to find:

Profile	What it finds	Notes
Exact & near duplicates	Copies, renames and re-encodes of the same video or photo	Fastest. Similarity 98%, no extras.
Edited & altered copies	Also crops, watermarks, flips and quality changes	Recommended default. Similarity 92%, flipped comparison and black/white pixel handling on.
AI scan	Also cropped, mirrored and re-edited copies, plus clips cut from longer videos, all detected visually	Adds the AI matching passes. No audio decoding.
Deep clean	Everything above, plus audio-fingerprint partial clip detection	The most thorough combination and the slowest first scan.
Custom	Whatever your own settings say	For people who know the knobs.

The profiles only manage a handful of matching settings (similarity threshold, flipped comparison, black/white pixel handling, and the partial/AI toggles). Everything else, like filters, performance and database settings, always stays yours.

If you change any managed setting in the settings screen, the selection automatically switches to Custom. Nothing is lost: your custom values are remembered, and picking Custom later restores them.

Press Scan to start. Advanced settings on the same screen opens the full settings view.

During the scan

The scanning screen shows overall progress plus a separate progress bar per drive, each with an SSD or HDD chip showing how the drive was classified (see Performance & Parallelism for why that matters). You can pause and resume at any time; pausing saves your progress. Stop aborts the scan but keeps everything already written to the database, so the next scan continues where this one left off.

The first scan is always the slowest because every file has to be read and analyzed. Results are cached in the database, so subsequent scans only process new or changed files.

Reviewing results

Duplicates appear in groups. For each file you see thumbnails, resolution, duration, codecs, size and more. From here you can:

• Compare two files side by side in a dedicated comparer window
• Auto-select files per group by quality, duration, file size or a custom expression, then delete, move or recycle the marked files
• Filter the list by path, size, group type and other criteria

When you are done, New scan takes you back to the setup screen with your folder list intact.

Common adjustments

Too many false positives (files flagged that are not duplicates)

• Raise the similarity threshold, or switch to the Exact & near duplicates profile.
• Increase Frames sampled per video (Settings > Scanning). More frames mean more evidence per comparison. Note that changing it re-extracts every video on the next scan.

Duplicates are not found

• Lower the similarity threshold slightly, or switch to a broader profile.
• Enable logging in the settings and check whether the files are being excluded and why (duration difference is a common reason).
• If the copies are cropped, zoomed or otherwise edited, try the AI scan profile.

Scan is slow

• See Performance & Parallelism. The two biggest wins are usually the native FFmpeg bindings and a sensible parallel worker count for your storage.