Image Search With Operators: refern Guide (2026)

By refern | Last updated: June 2026

refern ships 14+ inline search operators you type directly in the search bar. Combine type:image, tag:anatomy, rating:>=4, color:#1a1a2e, is:duplicate, derived:, linked:, and more in one query. All search runs locally on SQLite FTS5, so there is no cloud latency and no data leaves your machine.

If you have ever stared at a library of a few thousand images trying to remember what you tagged something, you know how quickly keyword-only search breaks down. This guide covers every operator refern supports, how to compose them, and copyable examples you can paste in right now.

Before you start

You need a refern workspace open with at least a few indexed images. If you are just getting started, see how to set up your first reference workspace or how to organize reference images.

The search bar sits at the top of every folder view. Press Ctrl+K (Windows/Linux) or Cmd+K (macOS) to open the full-screen search overlay, which searches your entire workspace. The inline bar at the top of a folder scopes results to that folder and its subfolders.

All search in refern is local. No query leaves your machine. Results appear in milliseconds even on libraries with tens of thousands of images.

Step 1: Understand the two search layers

refern search has two parts that work together.

Full-text search (FTS5 BM25). When you type plain words, refern searches across file name, folder path, description, notes, source URL, creator field, and content tags using SQLite FTS5 with BM25 relevance ranking. The tokenizer uses Unicode normalization with diacritic removal, so "Cafe" finds "Cafe" and "Cafe" finds "Cafe." Typo correction is not included, so spelling must be correct.

Inline operators. Operators are keywords typed directly in the search bar in the form operator:value. They produce precise SQL filters that combine with full-text terms. A query like landscape type:image rating:>=3 runs the FTS5 match on "landscape" and the SQL filters for image files rated 3 or above simultaneously.

You do not need to choose between the two layers. Every query can mix both.

Step 2: Filter by file type

The type: operator narrows results to a specific file kind.

type:image is the most useful starting filter in a mixed workspace. Use it whenever you want to exclude canvases, documents, and creative source files from results.

Example. Find all canvas files that mention "character" in any metadata field:

character type:canvas

Step 3: Search by tag

The tag: operator filters by a tag applied in your library.

• tag:landscape matches images tagged with "landscape" exactly.
• tag:"character concept" wraps a multi-word tag in quotes for an exact match.
• tag:anatomy,gesture requires both tags (comma-separated tags are ANDed).

Because refern uses hierarchical tags, searching a parent tag surfaces images tagged with any of its children. If you have a parent tag "figure" with children "anatomy," "gesture," and "proportion," then tag:figure returns all of them.

Composed example. Find portrait references rated 4 or above that have both "lighting reference" and "face" tags:

type:image rating:>=4 tag:"lighting reference",face

Step 4: Filter by rating

The rating: operator filters by star rating (1 to 5) and supports comparison operators.

Composed example. Find your highest-rated color references before starting a new project:

type:image rating:>=4 tag:color

Step 5: Search by color

The color: operator is one of refern's distinctive features. Type any hex value and refern scores your library by visual color similarity, returning the closest matches ranked by a blend of dominant color, HSV histogram, color layout, and edge histogram.

Format. color:#rrggbb (standard CSS hex)

Only images that score at or above a quality threshold appear. Use sort:colorMatch to make the ranking explicit.

Examples.

color:#1a1a2e

Dark navy-toned references.

type:image color:#f5a623 sort:colorMatch rating:>=3

Well-rated images with warm amber tones, ranked by how close the color match is.

You can also open the color picker (the color chip icon in the search bar) to choose a color visually without typing a hex code.

Step 6: Find duplicate images

Type is:duplicate to surface images whose perceptual hash (pHash) is close to at least one other file in your library. pHash compares image content, not filenames, so files with different names but similar pixels appear together.

Useful combinations.

is:duplicate type:image

Limits the scan to images, excluding canvases and documents.

is:duplicate in:references tag:anatomy

Finds duplicate anatomy references inside a specific folder.

After surfacing duplicates, select them in the grid and use the sidebar to decide which to keep, then soft-delete the rest.

Step 7: Search by relationship status

refern tracks typed relationships between images, canvases, and groups. Three operators expose those relationships in search.

linked:true and linked:false

• linked:true returns images that have at least one cross-reference link to another image.
• linked:false returns images with no cross-reference links.

linked-to:<entity-id>

Finds all images directly linked to a specific image by its ID. In practice, right-clicking an image and choosing "Find linked" pre-fills this for you.

derived:true and derived:false

When you use refern's crop tool and choose "Save as new," the cropped image is linked to its source via a "derived-from" entity link.

• derived:true returns images that carry a derivation link.
• derived:false returns original, non-derived images only.

Composed example. Original, highly-rated source images with no derivatives:

type:image derived:false rating:>=4

Step 8: Scope search with in:

By default, the search overlay searches your entire workspace. The in: operator scopes results to a specific folder by name.

in:Characters tag:anatomy

Finds anatomy-tagged images inside a folder named "Characters."

in:"Environment Art" type:image rating:>=3

Finds rated images inside "Environment Art."

The in: operator does a case-insensitive prefix match on folder name. in:char would match "Characters," "Character Concepts," and "Character Art" if they all start with "char."

Step 9: Control sort order with sort:

The sort: operator changes how results are ordered. Sort runs at the SQL level and does not slow down on large libraries.

Example. All lighting references, best-rated first:

tag:lighting sort:rating

Step 10: Filter by other metadata fields

Several operators cover the remaining metadata fields.

Example. Find untagged images added this year that still need annotation:

type:image dateAdded:>=2026-01-01 hasDescription:false

Composing multi-operator queries: real workflow examples

All operators combine freely with AND semantics. Comma-separated tag values use OR within the tag filter but AND with everything else outside it.

Best anatomy references for a figure-drawing session:

type:image tag:anatomy rating:>=4 sort:rating

Warm-toned mood references with descriptions:

type:image color:#e07a5f rating:>=3 hasDescription:true

Clean up duplicates in a specific folder:

in:Collected is:duplicate type:image

Review your highest-quality favourited images:

favourite:true rating:>=4 type:image sort:rating

Find original character references that still need tags and notes:

type:image tag:character derived:false hasNotes:false

All images from ArtStation rated 4 or above:

source:artstation rating:>=4 type:image

How refern compares to other tools on search

TagStudio uses Boolean AND/OR/NOT with parenthesis grouping and glob path syntax for filename and path matching. That model is genuinely expressive for users comfortable with query languages. The trade-off is that TagStudio has no image-specific operators: no color search, no visual similarity, and no pHash duplicate detection. TagStudio is still in alpha, so performance on large libraries can be inconsistent.

Pixcall offers filtering by color palette (9-color index), file type, rating, and aspect ratio through UI panels as of 2026. Pixcall also has an AI smart-search feature, but it requires each file to be individually pre-analyzed before it becomes searchable. There is no inline operator language.

Milanote has basic keyword search across visible card text with no image-specific operators. Search is an acknowledged weak point: one user review described a recent update as limiting results to recently viewed content, making it harder to find older boards. Milanote is a board collaboration tool first, not a library manager.

Common problems and fixes

The operator returns no results even though the image exists.

Check spelling and formatting. Operators must be formatted without spaces around the colon: rating:>=3 not rating: >= 3. Tag names must match exactly. If the image was added very recently, wait a moment for indexing to complete or trigger a manual folder resync.

Color search returns images that look unrelated.

Color matching is perceptual rather than exact. A very common hex like #ffffff matches any light image, not just pure white. Try a more specific hue, or combine color: with tag: or rating: to narrow the result set.

is:duplicate is not catching a pair I know are duplicates.

pHash uses a similarity distance threshold. Very small thumbnails, heavily compressed images, or images with significant color-space differences may not cross it. For a broader match, use the visual similarity search instead: right-click any image and choose "Find similar."

tag: returns fewer results than expected.

Confirm the tag name matches exactly. Also check whether you mean a parent tag (which covers children) or a specific child tag. If you recently applied a tag in the sidebar, confirm the save bar was submitted.

Next steps

• How to tag reference images covers the hierarchical tag system that makes tag: searches powerful.
• How to find similar images in your local library covers visual similarity and is:duplicate workflows in detail.
• How to search images by color explains the color scoring model and how to use the color picker.
• How to visualize reference relationships as a graph shows the derived: and linked: operators from the relationship graph side.
• What is a reference manager? explains the tool category if you are still evaluating options.