Find Similar Images in Your Local Library (2026)

By refern | Last updated: June 2026

refern includes a built-in local visual similarity search. Select any image, choose Find Similar, and it surfaces related images from your library in seconds, with no internet connection, no plugin, and no API calls. The search compares a 512-byte feature descriptor that captures color distribution, composition, and edge patterns across every image you have indexed.

This guide explains how to use that feature, when it is the right tool versus color search or tag search, and how it compares to what other desktop tools offer.

What you need before you start

You need a refern workspace that has finished indexing. The streaming indexer builds feature descriptors for every image as part of the thumbnail pipeline. If you just added a large folder, wait for the progress card to finish before running similarity searches.

To check: look at the bottom status bar. When no pipeline is active, you are ready.

You do not need an account, an internet connection, or any plugin installed. Visual similarity is a built-in feature available in all refern workspaces.

Step 1: Open an image and trigger Find Similar

There are three entry points:

1. Right-click on any image in the grid and choose "Find Similar" from the context menu.
2. Open the image in the sidebar (single-click) and click the Find Similar button in the metadata panel.
3. On the infinite canvas, select an image, open the radial wheel menu (right-click or hold on the image), and choose Find Similar.

All three methods launch the same search. refern computes a similarity score against every indexed image and returns a ranked list with the most similar images first.

Step 2: Read the results

Results appear as a grid sorted by similarity score. The most similar images are at the top left.

What to expect:

• Images with a closely matching subject and lighting appear near the top.
• Images that share general mood or composition but differ in subject appear further down.
• Images that only share one color family with no compositional similarity appear at the bottom or not at all.

The search is matching on four weighted signals combined into a single score: the dominant color clusters (what the overall palette is), an HSV color histogram (how color tones are distributed), a color layout map (roughly where colors sit in the frame), and an edge histogram (where the major lines and transitions are). An image of a figure lit from the right against a dark background will surface other similarly lit figure studies, even if they depict different subjects.

Threshold note. The search returns images above a minimum similarity threshold. Low-confidence matches are excluded automatically, so the results you see are genuinely related rather than being the entire library ranked weakly.

Step 3: Act on the results

From the similarity results grid, you have the same options as any other grid in refern:

• Select multiple results and drag them onto a canvas to build a mood board.
• Tag them in bulk using the tag editor if you notice they share a theme you have not named yet.
• Favorite or color-label the ones you want to keep close.
• Open one result as the new source for another round of Find Similar, effectively walking a chain of related images through your library.

This is useful when you realize mid-project that a reference you saved months ago is exactly what you need, but you saved it under a different topic and cannot remember what it was called.

When to use visual similarity versus other search methods

refern gives you several ways to find images. Choosing the right one saves time.

Visual similarity and color search overlap, but are not the same. Color search is precise: it matches images that contain a specific hue above a threshold. Similarity search is holistic: it weighs hue distribution, composition, and structure together. Use color search when you have a hex value. Use Find Similar when you have an image and want its aesthetic neighbors.

Step 4: Use search operators to narrow similarity results (advanced)

The Find Similar results appear in a dedicated view, but you can combine similarity search with tag and folder filters to narrow the scope:

• Scope the search to a specific folder by navigating into that folder first, then triggering Find Similar. Results are drawn from the current scope.
• After the results appear, use the filter bar to add additional constraints such as rating, color label, or file type.
• To find images similar to a source AND tagged with a specific concept, run Find Similar first, then filter by tag in the results.

You can also use the is:duplicate operator in the main search bar independently of Find Similar. Duplicate detection uses pHash (perceptual hash), a different algorithm optimized for identifying near-identical copies of the same image at different sizes or compression levels. Find Similar uses the 512-byte visual descriptor, which is optimized for compositional and chromatic resemblance across genuinely different images. Both are local and offline.

How this compares to other tools

Eagle

Eagle is a capable local asset manager for Windows and macOS. As of March 2026, Eagle added local visual similarity via its AI Search plugin for Eagle 4.0, which is available free in the Eagle Plugin Center. Eagle AI Search plugin, March 2026

That is a real addition to Eagle. The key differences from refern:

• The Eagle AI Search plugin requires a manual installation step from the Plugin Center. refern's similarity search is built in with no plugin required.
• Eagle's approach requires users to know the plugin exists and install it. Many Eagle users who have not updated recently still have no local similarity search.
• Eagle does not run on Linux. refern runs on Windows, macOS, and Linux.

Eagle has genuine strengths refern does not match: it previews 99 to 108 file formats natively (covering fonts, audio, 3D, and more), has a mature plugin ecosystem, and has been around longer with a larger community. For format breadth, Eagle is the stronger tool.

Pixcall

Pixcall is a cloud-synced asset manager popular in the Chinese market. It has color-palette filtering and keyword search, but as of 2026 it has no image-to-image visual similarity search documented in its feature pages or help center. Pixcall features, pixcall.com/features/

Pixcall's core strength is its native cloud sync with end-to-end encryption across Windows, macOS, iOS, and Android, which refern does not have yet (cloud sync is on the refern roadmap for Phase 2). Pixcall also has AI auto-tagging shipped since October 2025. For users whose top priority is multi-device sync or mobile access, Pixcall addresses that need.

For visual similarity search, refern is the option with local matching built in.

Common problems and fixes

"Find Similar returns very few results."

This usually means the workspace is still indexing. Feature descriptors are built during the thumbnail pipeline. If you recently added a large batch of images, wait for the indexer to finish, then try again.

"The results don't seem related to the source image."

Try a different source image that has a clearer dominant subject. Images that are very abstract, very low-contrast, or mostly flat color produce less distinctive descriptors and may return less targeted results.

"I want similarity search limited to one folder."

Navigate into that folder in the sidebar first, then trigger Find Similar. The result scope follows the current view.

"I want to find exact duplicates, not just similar images."

Use the is:duplicate search operator instead of Find Similar. The duplicate detector uses pHash, which is tuned for catching identical-or-near-identical copies rather than compositional neighbors.

Next steps

Once you are comfortable with visual similarity, these guides cover related capabilities:

• How to search your library with operators covers the full set of 14 search operators including is:duplicate, derived:, linked:, and color search.
• How to search images by color explains hex-based color search and when to use it over Find Similar.
• How to visualize reference relationships with the graph view shows how to use refern's Obsidian-style link graph to navigate connections between images, canvases, and folders.
• refern vs Eagle covers the full feature comparison if you are evaluating both tools.

Conclusion

Visual similarity search in refern works locally, requires no internet, and is available on every image and canvas element without installing anything extra. It is one of several search tools in refern alongside full-text FTS5 search, color search, and duplicate detection, each designed for a specific kind of retrieval task.

Try refern free for 30 days, then $30 one-time at refern.app. ($30 is launch pricing, going to $35 about two months after launch. One license, up to 3 devices, commercial use included.)