How to Find Visually Similar Files on Mac
Arun Gopidas
Five years of client projects. Dozens of external drives. Thousands of JPGs, PNGs, exports, and screenshots scattered across a file system that made sense at the time and makes no sense now.
You need to find images that look like this one. Not the same file — similar ones. Other versions. Other crops. Other exports made at different points in the project. Images saved with whatever name made sense in the moment.
macOS has no tool for this. Finder searches by name. Spotlight searches by name. There's no "find me images that look like this" feature anywhere in macOS — which means designers have been doing this manually for years: scrolling, squinting, opening folder after folder.
reverscan adds visual similarity search to your Mac, locally, using Apple Vision.
The Problem with Large Image Libraries
A working designer accumulates images fast. Client assets. Brand photography. Stock images. UI screenshots. Mockup exports. Reference images. Iteration after iteration saved with slightly different names — or the same name in different folders, which is worse.
When you need to find something visual — a specific product shot, a logo variant, a UI export from eight months ago — you have almost no good options on macOS:
Spotlight: can find by filename, but you probably don't remember the name
Finder: same problem, plus it's slow on large directories
Smart Folders: require metadata attributes, useless for visual content
Photos app: only sees your Photos library, not project files scattered across drives
None of these understand what an image looks like. They understand what it's called.
How Visual Similarity Search Works in reverscan
reverscan builds a visual index of every JPG and PNG in your file library using Apple Vision's VNGenerateImageFeaturePrintRequest. This generates a compact numerical vector — called an embedding — that encodes the visual characteristics of each image: colour palette, composition, textures, shapes, spatial relationships.
When you want to find visually similar files, you drop an image into reverscan. It generates an embedding for that image and runs a cosine similarity comparison against every embedding in your index. Files that score 0.50 or higher are returned as results, ranked from most similar to least.
The top 15 matches appear as a thumbnail strip. Results with a similarity score of 0.75 or above are marked "high confidence."
This all happens locally. Apple Vision runs on-device. No image is ever uploaded or processed externally.
[SCREENSHOT: reverscan visual similarity results — a row of image thumbnails showing files that are visually similar to the dropped image]
Setting Up Visual Search
Step 1 — Index your images
Open reverscan and click + Index Entire Computer for your home folder, or + Index Drive / Folder to add an external drive or specific project folder. Visual embeddings are generated automatically during indexing — this is not a separate step.
Indexing runs in the background. A home folder typically takes under a minute. A large external drive with thousands of images takes longer, but you can keep working while it runs.
Step 2 — Drop an image to search
Go to the Search tab and drag any JPG or PNG into the drop zone. reverscan immediately returns visually similar matches from your index.
Step 3 — Explore the results
Click any thumbnail to reveal the file in Finder. Results from drives that are currently disconnected show with an "Unplugged" badge — you'll know exactly which drive to plug in.
[SCREENSHOT: the drop zone in reverscan's Search tab, with a drag-and-drop illustration]
Try it free → Download reverscan — visual similarity search is included, no extra setup needed.
What Visual Similarity Search Is Actually Good For
Finding all versions of an image
Exported the same comp at different sizes, different crops, different colour treatments? Drop any one of them in — reverscan surfaces the rest, regardless of what folder they're in or what they're called.
Locating a reference image you saved
You saved a screenshot of a design reference months ago somewhere in your Downloads or Desktop. You remember what it looked like. Drop anything visually close to it and let reverscan scan your index.
Checking if a client asset is already in your library
A client sends you a new brand photo. Before you save it, drop it in — you might already have it, or a similar version, from a previous project.
Auditing visual consistency across a project
Index your project folder, then drop a hero image. Instantly see all the visuals in that project that share a similar look — useful for consistency reviews and handoffs.
Finding images across disconnected drives
Index your archive drives once. Even when they're unplugged, reverscan can tell you visually similar images exist on them — so you know which drive to pull out of the drawer.
How Similar Is "Similar"?
reverscan uses a 0.50 cosine similarity threshold. Here's what that tends to look like in practice:
Similarity score | Likely relationship |
|---|---|
0.90 – 1.00 | Same image, different export or crop |
0.75 – 0.90 | Very close variant — different colour grade, slight crop |
0.60 – 0.75 | Visually related — similar subject, layout, or palette |
0.50 – 0.60 | Loosely similar — shared visual theme or composition |
Results below 0.50 are filtered out entirely. If you're getting too many loose matches, that's a signal to look at the top results first — the highest-scoring matches are almost always meaningful.
Tips for Better Results
Index your external drives. If your project archives live on external drives, add them in the Drives tab. They stay in the visual index even when unplugged, so visual search always covers your full library.
Use a high-quality image to search. A blurry or heavily compressed thumbnail produces a weaker embedding. If you have a clean full-resolution version of what you're looking for, use that.
Combine with content search for richer results. If you remember anything about what was in the file — a project name, a product, a client name — use that as a text search first, then use visual search to find image variants.
Frequently Asked Questions
Do I need internet access for visual search?
No. Apple Vision runs entirely on-device. Visual search works offline.
How many images can it handle before slowing down?
At 7,000 indexed images, a visual search takes roughly 5ms — imperceptible. The search uses parallel processing via rayon, so it scales well with larger libraries.
Can I find visually similar PSD or Sketch files?
The visual search works on JPG and PNG files specifically. However, if you've run a Deep Scan, reverscan has rendered your design files to images and OCR'd them — those renderings don't feed into visual search directly, but the text content becomes searchable. For finding design files by visual appearance, export a PNG from the design tool first and drop that in.
What if I drop an image and get zero results?
Either nothing in your index scores above 0.50 (the image is visually unique or your library doesn't contain similar content), or your images haven't been indexed yet. Check the Drives tab and run a sync if needed.
Does this detect duplicate images?
Yes — images that are identical or near-identical will score very high (0.90+) and appear at the top of results. It's not a dedicated duplicate detector, but dropping any image immediately surfaces its near-duplicates.
What happens to visual search when I add new images?
reverscan updates the visual index automatically as part of its regular sync. New images get embeddings generated during the sync pass.
Your image library is findable by what images look like, not just what they're called. Download reverscan free and see what's hiding in your own file system.