The one comparison where they win our best argument

Almost every page on this site eventually reaches the same claim: a board here is a plain file in your own folder, which is something you can check rather than something you have to trust. Against Obsidian Canvas that claim is not a differentiator, it is a point of agreement — and theirs is the stronger version of it. So take the recommendation first. If you already live in Obsidian and you want to lay out a few dozen images beside your notes, use Canvas; it is free, it is on every platform, and its file format is a published open standard. Come here when the problem is not the layout but the twenty thousand images the layout is supposed to be chosen from.

What Obsidian costs, checked today

From their pricing page on 3 September 2026, which is unusually plain-spoken and worth reading in full:

• The app is free without limits, and needs no sign-up. Canvas is part of it, not an add-on.
• Sync — $4 per user a month billed annually, $5 billed monthly.
• Publish — $8 per site a month billed annually, $10 billed monthly.
• Catalyst — $25 once, a supporter licence.
• Commercial — $50 per user a year, and their own FAQ says you are not required to buy it, only encouraged to if your organisation uses the app.

Coii Ref: $19 USD once, three Macs, updates included, thirty-day trial with every feature and no card and no account, macOS 11 Big Sur or later.

On price this is not a contest. Free with no sign-up beats nineteen dollars, and pretending otherwise would be the kind of sleight this page exists to avoid.

The format argument, conceded properly

Obsidian's canvas page says a canvas is an infinite space stored locally in the open JSON Canvas format, designed to work with other tools. That format lives at jsoncanvas.org, uses a .canvas extension, and is released under an MIT licence — originally built for Obsidian and then handed to anyone who wants to implement it. The 1.0 spec defines four node types: text, file, link and group, with a file node carrying a path and an optional subpath into the file.

Read that list again, because it is the thing we usually say about ourselves and they have done it more thoroughly. A .coii-canvas is pretty-printed JSON in a folder you chose, which opens in a text editor, diffs in git and syncs with whatever already syncs — all true, all checkable with cat. But there is no published specification for it, and no second application implements it. Theirs is an open standard; ours is an open file. Those are not the same word.

The practical difference is smaller than it sounds — a JSON file you can read is a JSON file you can read, and the failure mode both of us are guarding against is the same one — but the honest ordering is theirs first.

Where the two products actually stop overlapping

A canvas is a document. A library is an index. Everything below follows from that.

Obsidian works from a vault, which their documentation describes as a folder on your local file system where Obsidian stores your notes. It is a place you write, and Canvas is the spatial view of that writing — with images, video, PDFs, audio and even interactive web pages placed alongside, which is a wider range of things on a board than we will ever put on one.

What it is not is a way to interrogate a collection you did not curate. There is no index over the four folders and two external drives where your reference images actually are, no way to ask for every image over 3,000 pixels added since June that nobody tagged, no ratings on pictures, no proxies so that a board of video clips scrolls.

That is not a criticism. It is a note-taking application with an excellent spatial view, and asking it to be a digital asset manager is asking the wrong thing of it. But if you arrived at this page by searching how to use Obsidian Canvas for images, that gap is what you are about to run into at around the two hundredth image.

The part Coii Ref is actually for

Point it at a parent folder anywhere on disk — internal, external, a drive that is unplugged half the week — and it reads what is in there without copying, moving or renaming a single original.

Then one line of search over all of it, before you have tagged anything, because size, format, folder, dates and ratings are facts the disk already holds:

The keys are type ext tag rating is has no in width height added modified. AND across keys, OR inside a comma list, no parentheses, and any word the parser does not recognise falls through to free text rather than erroring, so the grid keeps answering while you type. Saved views store the text of a search and re-parse it on open, which means renaming a tag updates every view that mentioned it instead of leaving a stale list behind.

And then the boards, which is the part that looks superficially like Canvas:

The difference from a Canvas is not the layout. It is that the plates arrived from a search over twenty thousand files rather than from a folder you had already narrowed by hand.

What is actually inside each file

Both formats are readable, and reading them is the fastest way to understand what each product thinks a board is.

JSON Canvas 1.0 describes a canvas as nodes and edges. Its four node types are text, file, link and group; a file node carries a path to something in the vault and an optional subpath pointing at a section within it. Edges connect nodes, which is what makes it as good for a diagram or an argument as for a wall of pictures. It is a general spatial document format, and the generality is deliberate.

A .coii-canvas is narrower on purpose. It names the image and video files it refers to, where each one sits, and how big each is drawn. There are no edges, because it is not trying to be a diagram; it is trying to be a wall you can look at while you work. What it does carry that a general format does not is the connection back to an index — a plate on the board is a file the application also knows the rating, tags, dimensions and notes of, so a board is a view onto a library rather than a folder of copies.

Neither of those is better in the abstract. One is a document format that other tools can implement; the other is a saved arrangement of an indexed collection. The reason to know the difference is that it predicts exactly which one runs out first for your kind of work.

At twenty thousand files

The number is where the two products stop being interchangeable, and it is worth being concrete about why.

Placing images on a canvas is a manual act: you decide what goes on, and you decide by having already found it. That works beautifully at forty images and tolerably at four hundred. It does not scale to a collection, because the bottleneck moves from arrangement to retrieval — you cannot lay out the nine right references if you cannot locate them among four years of files nobody tagged.

An index changes what the question can be made of. Instead of remembering which folder something is in, you describe the file: roughly how big, what kind, roughly when it arrived, whether you rated it, whether you wrote on it. Those facts are already on the disk, so the first day of an index is already useful, before a single tag has been typed.

type:image width:>=3500 rating:>=4                the big good ones
in:Scans tag:any:paper,texture                     either tag, one folder
type:video added:>2026-01 has:notes                clips you wrote on
ext:png -is:favourite no:tags                      unstarred, unfiled PNGs

Tagging then becomes an optional thin layer for the things a file cannot tell you itself: mood, project, client, why you kept it. Everything else was searchable from the start.

Video, and why it is not a footnote

A canvas in Obsidian will hold a video; their own page says so, and that is more than several dedicated image tools manage. What differs is what happens around it.

Here, a clip goes through the same index pass as a still. It carries the same tags and the same rating, answers the same query line — type:video ext:mov rating:>=4 is an ordinary sentence rather than a separate mode — and gets a proxy written at canvas resolution, so a board of nine clips scrolls the way a board of nine photographs does instead of stuttering while several source files decode at once. If a real part of what you study is motion, that is the difference between keeping one library and keeping two.

The table

Count the bold and it is roughly even, which is unusual for these pages and is the reason this one is worth writing carefully. The rows are not really in competition; they are two products that happen to share three of them.

One more thing that catches people who arrive from a note-taking habit: there is nothing to write in. No note beside a board, no long-form text on the canvas, no linking one board to another. What there is, is a notes field on each file — a line or two attached to the picture itself, searchable with has:notes, and carried along when the file is renamed or moved in Finder. That is deliberate scope rather than a missing feature, but if you expected to write around your references, the writing stays in Obsidian, where it was always going to be better anyway.

Using both, which is easy here

There is no conflict to manage, because neither application changes the other's files.

Keep the vault where it is. Notes, briefs, research, the writing — that is Obsidian's job and nothing here does it. Use Canvas for the small spatial things: a page of thumbnails beside a note, a diagram, a sequence you are reasoning about in words as much as in pictures.

Point Coii Ref at the folders where the actual image mass lives — the shoots, the scans, the four years of downloads, the external drive. Those folders can sit beside the vault, or inside it, or on another disk entirely; the index is written into whichever folder you pointed at and nothing else is touched. Search there, rate there, build the reference canvases there, and drop a link into a note when a board matters to a piece of writing.

The one habit worth adopting: keep folder names that will still mean something in three years. Both tools reward it and neither can invent it for you.

What Coii Ref does not have

Flatly, so nothing in the trial is a surprise. No notes, no writing, no markdown, no links between documents and no graph of any kind. No plugin system. No sync add-on and no publishing — if a board needs to be somewhere else, you move the file yourself. No accounts, no cloud, no collaboration, no mobile app. No browser extension. No PDFs, audio, documents or 3D formats — images and video only, which is narrower than what a Canvas will hold. No automatic tagging and no search by meaning: the query line is literal. No search by colour, though colour is read into the index. And macOS only, version 11 or later.

The network list is two items long: a licence check when you press Activate, and one request at launch read only for its Date header, so a wound-back clock cannot mint free trial days. No analytics, no crash reporting, no update check. That is a small surface, but it is not zero and this page is not going to round it down.

What the trial does and does not stop

Thirty days, every feature, no card, no account. On day thirty-one, if you have not bought, one category of action stops — the one that grows the library: registering a new folder, running the importer, starting an index pass by hand. Everything that reads carries on permanently: search, tags, ratings, saved views, canvases and thumbnails. The folder watcher keeps running too, turning filesystem events into the smallest re-scan that makes the index true again, so an unpaid copy does not sit there letting your index go stale. It just stops taking on more.

Who should pick which

Obsidian Canvas if the centre of your work is writing and the images are in service of it; if you want it free, on Windows or Linux or a phone as well as a Mac; if a published open format matters to you more than an inspectable one; or if the number of images involved is small enough that finding them was never the problem.

Coii Ref if the images are the work and there are far too many of them: twenty thousand files across folders and drives, unfiled, with video mixed in, and the daily problem is retrieval rather than layout. $19 once, indexed in place, one query line, boards that save as files beside the pictures.

Both, in the division above, which costs $19 and no ongoing arrangement.

Neither, if the collection is a couple of thousand images and Finder still copes. The moment to buy anything is the first time you cannot find a picture you are certain you saved.

If a board floating over your drawing app is closer to what you want, PureRef is the tool for that and is free for personal use; Milanote is where a board goes when it needs writing on it and a client looking at it; Miro is the same idea at team scale; and if the material you collect is really links and articles rather than image files, Raindrop is the better half of that argument. Finder tags is worth understanding first either way, since it is already on the machine.

Otherwise: the library has the rest, and the trial runs thirty days with every feature, no card and no account.