Boneyard Tools

Visual Duplicate Finder

Find duplicate and near-duplicate images by how they LOOK, not by an exact file match. Drop in a whole batch of photos and an AI vision model (CLIP) gives each one a visual fingerprint, then compares them by similarity, all in your browser. Because it compares the picture and not the bytes, it catches resized copies, re-saved JPEGs, light crops, and small edits that defeat ordinary duplicate finders that only match identical files. Group near-identical images and see the one to keep plus its duplicates, or list the most similar pairs ranked by a similarity score. Drag the threshold slider to make matching stricter or looser and the groups update instantly, with no re-analysis. Nothing is uploaded; the model downloads once on first use, then is cached.

How to find duplicate images

  1. Add the images you want to check (drag a batch in, or browse). Two or more.
  2. Wait a few seconds while each image is analyzed in your browser.
  3. Drag the similarity slider to tune matching, then read the duplicate groups or the most similar pairs.

Examples

A folder of near-duplicate photos

12 images, including a JPEG and a resized copy of the same shot
3 groups, 5 duplicates of 12 images, with the copy to keep marked in each group

Frequently asked questions

Are my images uploaded anywhere?

No. Every image is analyzed entirely in your browser with a CLIP vision model running via transformers.js (WebGPU accelerated when available, WebAssembly otherwise). Your photos are processed on your device and never leave it. Only the model itself downloads, once, then is cached for instant reuse.

How is this different from a normal duplicate file finder?

Ordinary finders compare file hashes, so they only catch byte-for-byte identical copies. This tool compares how the image looks. It turns each picture into a visual fingerprint and measures similarity, so it also catches resized copies, re-saved JPEGs, format conversions, light crops, and small edits that change the bytes but not the picture.

What does the similarity threshold do?

Each pair of images gets a similarity score from 0 to 100 percent. The threshold is the cutoff for calling two images duplicates. A higher threshold is stricter and only flags near-identical images; a lower one catches looser, more loosely related matches. Changing it re-groups the existing results instantly, with no re-analysis.

What is the difference between duplicate groups and most similar pairs?

Duplicate groups cluster near-identical images together and mark one to keep plus its duplicates, which is ideal for cleaning up a library. Most similar pairs simply lists every image pair above the threshold, ranked by similarity, which is useful when you want to review matches one by one before deleting anything.

How many images can I check at once?

As many as your browser's memory allows. Each image is analyzed once and its fingerprint is cached, so adjusting the threshold afterward is instant. Very large batches take longer to analyze the first time and use more memory, since the comparison looks at every pair of images.

Related tools