Visual Reference for Worldbuilding: Boards, Graphs, and Libraries

TL;DR: Writers and worldbuilders collect images alongside their prose, and most tools make you choose between a text PKM that treats images as afterthoughts and a cloud moodboard without a persistent searchable catalog. refern fills the gap: it is a local-first desktop app that organises character and setting images in a searchable library, lets you arrange them spatially on an infinite canvas, and maps their relationships in a visual graph. Think of it as the visual companion to Obsidian, not a replacement for it.

The worldbuilder's image problem

You are writing a novel. You have hundreds of saved images: concept art from games that nail the tone of your world, costume references for three different noble houses, architectural photography of the region that inspired your city, colour palettes for each faction, character face references you found across a dozen Pinterest boards and art blogs.

They live in a folder called "ref," or maybe eight folders with names like "characters-maybe" and "vibes-2." You know the image you need is in there somewhere. You have no idea how to find it.

Text-first tools like Obsidian are brilliant for prose, outlines, and research notes but treat images as attachments. Boards like Milanote and Are.na are designed for sharing with teams or curating small collections online, not for managing thousands of locally-owned images offline. The gap is a visual reference system that thinks the way a worldbuilder thinks: in images, in relationships, in spatial arrangements of ideas.

Before you start: what refern actually is

refern is a $35 one-time, local-first desktop visual reference manager with an infinite canvas and a relationship graph. It runs on Windows, macOS, and Linux. Normal workspace setup indexes existing source files in place. The SQLite database and generated thumbnails live alongside, or in a configured data location associated with, the workspace. Some licensing, update, model-download, feedback, web-download, and trusted-time actions use the network.

The 30-day trial is full-featured. After it expires, the app interface is gated by an activation screen, while normal on-disk files remain accessible. A $35 one-time license supports up to 3 devices and includes updates under the current terms.

One alpha user summarised the feel: "what if Obsidian had pictures instead of notes." That is the intended use case for writers.

How writers and worldbuilders actually use refern

The workflow has three layers that map directly to how worldbuilding research works.

Layer 1: The library (your reference collection)

Point refern at an existing folder and normal workspace setup indexes the existing source files in place. Explicit imports and file actions can create or change files. Practical capacity depends on hardware, storage, media types, and thumbnail state; this page does not publish a numerical benchmark.

Folders and tags for world structure. Create a folder hierarchy that mirrors your world: Characters / House Aldenmere, Locations / The Shattered Coast, Mood / Tension and Shadow. Within each folder you can add hierarchical tags (for example lighting > golden-hour, costume > military > cavalry) and bulk-apply them to selections. Tag macros let you type a shortcut and apply a whole cluster of tags at once.

Composable search. Full-text search supports 21 user-visible operator keywords. You can search tag:cavalry rating:>=4 in:Characters to narrow a query, use color: for dominant hue, or use visual similarity when files share no tags.

Metadata per image. Each image gets a description, notes, source URL, creator field, colour label, rating (1 to 5), and custom metadata. You can record where a piece of concept art came from and who made it, which matters for attribution if you are publishing a world guide. EXIF, IPTC, and XMP metadata embedded in image files is read automatically on import.

Browser extension for capture. The refern browser extension (Chrome, Firefox, Safari) supports hover-save, right-click save, and batch save. When you find concept art or photography that fits your world, you save it directly into the right folder without downloading to your desktop first.

Layer 2: The canvas (spatial boards for chapters, characters, and arcs)

The infinite canvas is where you move from library to composition. It is the layer that replaces a physical mood board pinned above your desk.

One canvas per story unit. A common pattern for worldbuilders: one canvas per major location, one per character arc, one per chapter. Images sit freely at any size. You can zoom out to see the shape of a scene and zoom in to examine a detail. Layers and groups let you separate levels of a board (the "feel" group, the "architecture" group, the "character" group) and toggle them on or off.

Shapes, text, and drawings alongside images. Add text labels, draw connective arrows, place a colour swatch next to an image to name the palette. Nine shape types and a freehand drawing tool cover annotation needs without switching to another app.

Non-destructive crop and filters. If an image has the right mood but the wrong crop, adjust it on the canvas without touching the original file. Image filters (brightness, contrast, saturation, grayscale) are applied non-destructively in the canvas view.

Pin on top of your writing software. The canvas window can be pinned on top of every other window at adjustable opacity with mouse click-through enabled. You keep your reference board visible while writing in Scrivener or Word without switching applications. This is the same overlay use case as PureRef, built into refern.

Find similar from the canvas. Right-click any image on the canvas and use "Find similar" to search the full library for visually similar images. If you placed a reference for the colour of your villain's throne room, you can find the thirty other images in your library that share that tone without leaving the canvas.

Layer 3: The graph (map of how your world connects)

The relationship graph view is refern's most unusual feature and the one that makes it feel closest to a text PKM. It renders every folder, image, canvas, tag, and group as a node, with typed edges between them.

Cross-references between images. You can create an explicit cross-reference link between two images that belong together conceptually, even if they live in different folders. The villain's costume reference and the heroine's costume reference exist in separate character folders but share a linked-fabrics relationship. The graph shows that edge.

Placed-in-canvas links. Every image you add to a canvas board gets a "placed in" link back to that canvas. The graph makes this visible: you can navigate from a character folder to every canvas board that image appears in, and from a canvas board to every image it uses.

Derived-from links. When you crop an image and save it as a new asset (for example, extracting a detail from a wide establishing shot), refern records a "derived from" link between the crop and the original. The graph preserves the visual provenance of every derived asset.

Navigable, not just decorative. Clicking a node in the graph navigates to that image or folder. The graph is a map of your world's visual logic, not a poster.

For a worldbuilder, the graph answers questions like: which images are central to my visual identity for the Merchant Quarter? Which character has the most visual connections to other characters? Where does my lighting palette for Act Three intersect with my location references?

Comparing the options

refern as the visual companion to Obsidian

If you already use Obsidian for notes, outlines, and research, refern fits as the image side of the same philosophy. Both are local-first, both keep data in normal files on your disk, both have a graph view, and both avoid cloud lock-in.

The important difference is that Obsidian was built for text. The Obsidian forum has a documented, open thread from 2021 titled "Creating an image gallery" with no satisfying built-in answer. A Medium article describes the "images as notes" workaround as impractical above 900 images, where each image requires a separate manually-created markdown note. Artists using Obsidian Canvas have submitted requests for grayscale toggle, nearest-neighbour rendering, and proper zoom for pixel art, with no resolution as of June 2026.

refern answers the pixel-art half of that request directly. Any file can be flagged Pixelated from the metadata sidebar, and flagged files then render with hard pixel edges instead of a blur wherever they appear: the grid, folder previews, the viewer, and the canvas, at any zoom. For a worldbuilder whose visual bible includes sprite work, map tiles, or old game screenshots, that is the difference between a legible reference and a smear. is:pixelated finds every file you flagged.

refern does not replace Obsidian for text. It replaces the plugin-heavy image workaround that Obsidian users build out of necessity. The stack that alpha users described as ideal: Obsidian for prose and text-based world notes, refern for every image that belongs to that world.

The relationship between the two tools is also reflected in how refern's graph view works. Just as Obsidian's graph maps connections between text notes, refern's graph maps connections between images, folders, canvases, and groups. The visual logic of a story world gets the same kind of navigable spatial map that writers already use for their prose.

What Milanote and Are.na do better (and when to choose them)

Honesty matters here, because the right tool depends on your workflow.

Milanote is genuinely beautiful and easy to use. Its templates library (100+ profession-specific options) gives a novelist or screenwriter a scaffolded board for a character bible or story structure in seconds. Its real-time collaboration and comment features make it the right choice if you are co-writing with a partner or sharing visual direction with an editor, cover designer, or publisher. If you want to paste a board link and have someone else annotate it without installing anything, Milanote does that. refern does not yet. The pricing comparison over time: Milanote at $9.99/mo individual costs $120/year; refern is $35 one-time.

Are.na has a genuinely distinctive community and a thoughtful "no algorithm" philosophy that resonates with writers who want slow, intentional curation without feeds or likes. Its "block in multiple channels" mechanic creates implicit cross-references that feel like research. It is a good tool for collecting and connecting ideas across a public or collaborative research space. Its limits for the solo offline worldbuilder: it requires an internet connection, stores everything in Milanote's cloud, has no spatial canvas, and offers no search inside block content.

Choose refern over both if you work solo, want to own your images on your own disk, plan to build a library of hundreds or thousands of files, and want a graph view that maps explicit typed relationships rather than implicit folder membership.

Common problems and fixes

"I pointed refern at my reference folder and it started scanning. How long does this take?" Scanning time depends on library size. A library of several thousand images typically indexes in a few minutes. The first thumbnail pass takes longer. A progress card shows the state. You can use the library immediately after scanning; thumbnails fill in as they are generated.

"My images are spread across five different folders in different locations." A single workspace can index several separate folders (content roots) at once, even when they live on different drives, so your whole world stays in one searchable library without moving anything. Add each folder as a content root and refern browses, searches, and tags across all of them together. (Adding extra content folders uses a separate custom data folder for the index.) If the folders are logically part of one world, you can also consolidate them under a single parent folder before pointing refern at it.

"I use Scrivener and want my reference board visible while I write." Open the refern canvas with the board you want, then use the pin-on-top option to keep the window above Scrivener. Set opacity to a comfortable level. Enable mouse click-through so clicks pass through the refern window to Scrivener without switching focus.

"I want to find images I saved months ago but I cannot remember what I called them." Use the colour search (click the colour picker in the search bar and sample or type a hex code). Use the visual similarity search (right-click any image and choose "Find similar"). Both search the full library by visual content rather than file names.

"I want to share my world's reference board with a beta reader or co-author." Shareable canvases are planned without a published date. Today, refern is single-user. Export a canvas to an image file for a handoff. For live collaborative access, Milanote is the better current choice.

Next steps

Once your reference library is set up in refern, several features extend the workflow further:

• Smart folders: define a saved search (for example tag:villain rating:>=4) and it becomes a permanent folder that updates automatically as new images match the criteria. Useful for a running "best references for the final act" view. Learn more about search and smart folders at refern.app.
• Timed study mode: open an image or a set of images on a timer. Originally designed for artists doing timed studies from reference, it works equally well for a writer who wants to spend five focused minutes inside a location reference before writing a scene set there.
• Import from Eagle: the dedicated importer has full parity across its supported transfer surface. It copies the first supported media placement into the refern workspace and represents later placements as linked database records. Password locks, saved smart-folder rules, arbitrary plugin-owned data, and unsupported media are excluded.
• Relationship graph exploration: start from any image in your library, open the graph view, and trace where that image connects. This is useful for finding visual threads you forgot you built, or for auditing whether your faction colour palettes are visually distinct.

For writers who already use Obsidian, see the refern relationship graph overview for a more detailed look at how typed entity links work compared to Obsidian's wikilinks.

Conclusion

Visual reference for worldbuilding is an underserved problem. Text PKMs treat images as attachments. Cloud moodboard tools lack search, metadata, and scale. Desktop reference managers built for concept artists do not have the relationship graph that worldbuilders need to see how their visual ideas connect.

refern sits at the intersection: an image-first library, a spatial canvas for building character and location boards, and a graph that maps the visual logic of a world. Core workflows run locally, normal indexing reads existing source files in place, and the license costs $35 one-time.

The ideal stack for a writer who takes their visual research seriously: Obsidian for prose and text notes, refern for every image that belongs to that world.