How to find files that look similar to one you have

You have one file in front of you.

You have one file in front of you. A design mockup, a reference photo, a research paper. You know there are others like it scattered across your workspace: earlier drafts, related references, images from the same shoot, documents on the same topic. But you cannot describe what you are looking for in a neat search query, because the connection is not a keyword. It is a resemblance.

This is the gap that similar search fills. Instead of asking you to type a query, it lets you point at something you already have and say "find me more like this." No keywords, no guessing at filenames, no remembering when you saved something. You start from an example, and the tool does the rest.

Why keyword search falls short for this problem

Traditional search is built around a simple transaction: you type words, the system finds files whose names or contents match those words. But it assumes you can articulate what you are looking for, and that the thing you want was labelled in a way that matches your articulation.

When you are looking for files that resemble something, that assumption breaks down. Suppose you have a photograph of a sunlit courtyard and you want to find other photos with a similar mood. What would you type? "Warm lighting outdoor architecture"? That query might miss the image tagged as "valencia_afternoon_03.jpg" or the one buried in a folder called "client references." The connection between these images is visual, not verbal.

The same problem appears with documents. You might have a strategy memo and want to find every related brief, presentation, and set of notes on the same theme. A keyword search catches some of them if you guess the right terms, but it misses the ones that discuss the same ideas in different language. This is where the limitations of filename-based search become most visible. You need a tool that understands meaning and appearance, not just labels.

How semantic similarity works

Similar search is powered by the same class of technology behind semantic search, but it works in reverse. Instead of converting your typed query into a mathematical representation and finding files near it, similar search takes a file you select and finds other files whose representations are close to it in the same space.

Every file in your workspace gets converted into a dense numerical representation when it enters your library. This representation captures not the surface features of the file but its deeper characteristics: what a document is about, what an image depicts, what concepts a presentation covers. Files that are related in meaning or appearance end up near each other in this mathematical space. Because Fabric acts as a persistent indexing layer across all your connected tools, this representation covers files from every source, not just those stored locally.

When you select a file and trigger similar search, the system finds the nearest neighbours to that file's representation. The results are ranked by proximity, so the most closely related files appear first. No query is needed because the file itself is the query. A PDF about brand strategy might surface a related set of notes, a slide deck from the same project, and a competitor analysis saved as a bookmark. The system does not care that these are different file types. It cares that they occupy the same conceptual neighbourhood.

Visual similarity for images and creative work

For visual files, similarity takes on a more literal meaning. When you select an image, similar search considers composition, colour palette, subject matter, and visual style. A photo of a mountain lake at dawn will surface other landscape photographs with similar tones and framing. A flat-lay product shot will pull up other flat-lays. A pencil sketch will find other hand-drawn work.

This is transformative for creative professionals. Designers maintaining reference libraries, photographers culling shoots, art directors building mood boards: all of these roles involve finding visual connections across large collections, and all of them have historically relied on meticulous tagging or impeccable memory. Similar search replaces both with something more natural. You start from a single image and let the tool reveal the visual thread running through your library.

Visual similarity also catches things that text search never could. Two photographs might share a colour palette and compositional structure while depicting entirely different subjects. A fashion editorial and an architectural photograph might both feature stark geometry against warm light. If you work with large photo libraries, visual similarity search becomes one of the most powerful tools for navigating them.

Cross-format connections

One of the most useful properties of similar search is that it works across file types. This sounds like a technical detail, but it changes how you think about your workspace.

Consider a research project. You might have academic papers as PDFs, your own notes in a text document, bookmarked articles, a few diagrams saved as images, and a presentation summarising your findings. In a traditional file system, these live in different folders, different apps, possibly different cloud services. They are connected by topic but separated by format.

When you select one of those PDFs and run a similar search, the system does not limit itself to other PDFs. It finds your notes, your bookmarks, your presentation, and those diagrams. Searching across all your apps at once already removes the barrier of location. Similar search removes the barrier of format.

This cross-format capability is especially valuable for people whose work spans multiple media. A brand strategist might move between written briefs, visual assets, and client feedback stored as notes. A researcher might work across papers, data visualisations, and field photographs. In each case, the relationships between files cut across format boundaries, and a tool that respects those boundaries cannot surface the full picture.

Finding every version of something

One of the most practical uses of similar search is tracking down every version, variant, and derivative of a file. This is a problem that plagues anyone who works iteratively, which is to say, everyone.

Design work generates versions relentlessly. A logo concept might spawn dozens of variations across colour, layout, and typography. A campaign asset might exist in formats for print, web, social media, and email. The original source file might be an Illustrator file, the exports might be PNGs, and there might be a PDF proof sent to the client six weeks ago.

Similar search finds all of these. Select any one version, and the system surfaces the others. It does not matter that they have different filenames, different formats, or that they live in different folders. What matters is that they are visually and conceptually close to the file you started with. This is also invaluable for identifying which version of a file is the latest, because when you can see every related version at a glance, the question of currency becomes much easier to answer.

Catching duplicates and near-duplicates

Workspaces accumulate clutter. The same file gets downloaded twice, saved in two locations, or exported in two slightly different crops. Near-duplicates are even more insidious: files that are almost identical but differ in some small way, whether through a minor edit, a format conversion, or a re-export at a different resolution.

Similar search exposes these duplicates naturally. When you run it on a file and find multiple near-identical results, you have a clear signal that your workspace contains redundancy. This is useful both for general tidying and for more structured digital file organisation, because cleaning up duplicates is one of the fastest ways to make a large library more navigable. For teams managing shared asset libraries, periodic similarity checks help keep things lean, which is one of the practical benefits of building a thoughtful digital asset management practice.

How it improves as your workspace grows

Similar search gets more useful as your workspace grows. In a small library of fifty files, it is a convenience. In a library of five thousand or fifty thousand, it becomes essential. The more material you have, the richer the web of connections, and the more likely a similar search will surface something useful that you had forgotten about.

A larger library provides more data points in that mathematical space. Fabric's memory engine builds on this by remembering context about your files and how they relate, so the system's understanding of your workspace deepens with use. A reference image you saved two years ago might suddenly become relevant when you save a new design brief, and similar search surfaces that connection without you needing to remember the image exists. Smart organisation amplifies this further: as files are automatically tagged and categorised, the connections between files become more precise. The combination of automatic organisation and similarity search creates a workspace that becomes easier to navigate the more you put into it, which is the opposite of how most file systems behave.

Putting similar search into practice

Using similar search is simple. Select any file in your workspace and look for the option to find similar content. The system returns a ranked list of related files, with the closest matches first. From there, you can browse the results, refine by type or date, or use one of the results as a new starting point to explore further.

The most effective approach is iterative. Start from your initial file, scan the results, and if one of them is closer to what you are looking for, run a similar search on that file instead. Each step takes you deeper into the relevant cluster of material. This is particularly powerful in Fabric's explorer, which lays out your files spatially and lets you follow conceptual threads through your library.

For creative teams, a useful habit is to run a similar search at the start of any new project. Select a brief, a mood board, or a reference image and see what already exists in your library. This surfaces past work, forgotten references, and potential starting points that would otherwise require hours of manual searching. Fabric's AI assistant can help you work with these results, pulling insights from related documents or suggesting connections you might not have spotted.

Frequently asked questions

Does similar search work for documents as well as images?

Yes. When you select a document, the system finds other files that discuss related topics or belong to the same conceptual domain. This works across formats: a Word document might surface related PDFs, notes, bookmarks, and even images that are topically connected.

How is similar search different from typing a search query?

A search query requires you to articulate what you want in words. Similar search requires only that you have one example. This distinction matters most when the connection between files is difficult to verbalise, whether because it is visual, conceptual, or simply too fuzzy to pin down in a phrase. Selecting a file is also faster than composing a query when you are not sure which words to use.

Can I use similar search to find files across different apps and cloud services?

If your files are connected to Fabric, similar search works across all of them regardless of where they are stored. A file in Google Drive can surface similar content from your local storage, bookmarks, or notes.

Does visual similarity work across different image formats?

Yes, across all common formats including JPEG, PNG, HEIF, TIFF, and WebP. The analysis processes the image content itself, not the container format, so a JPEG and a PNG of the same subject will be recognised as similar. This is useful for agencies and teams that work with assets in multiple formats.

How accurate is it at finding related design files?

Near-duplicates, colour variants, and layout variations of the same design are reliably caught. Broader thematic connections, such as two designs that share a visual style, are surfaced with good accuracy but naturally involve more judgement about what counts as "similar." The results are ranked by closeness, so the strongest matches always appear first.

Will it find files that are topically related but visually different?

For documents, the system prioritises semantic similarity, so it will find topically related files regardless of their appearance. For images, the emphasis is on visual characteristics: composition, colour, and subject matter. Cross-format results appear when the semantic connection is strong.

Can I use this to clean up duplicate files in my workspace?

Similar search is one of the most effective ways to spot duplicates and near-duplicates. When you select a file and see multiple almost-identical results, you know you have redundancy. This works for both exact duplicates and files that differ only in format, resolution, or minor edits.

How does it handle files I saved a long time ago?

Every file is indexed when it enters your library, regardless of when that happens. A file you saved two years ago is just as searchable as one you added today. Similar search does not deprioritise older files; if an old file is the closest match, it appears at the top of the results.

Is there a limit to how many files similar search can work across?

There is no practical limit that would affect normal use. Performance remains consistent at scale because the system compares compressed mathematical representations rather than raw file data.

Can I combine similar search with other search methods?

Yes, and this is often the most effective approach. You might start with a semantic search to find an initial result, then use similar search to fan out from it. Or you might browse your library in the explorer, spot something interesting, and run a similar search to see what connects to it. Using these methods in sequence gets you to the right files faster than any single approach alone.