← All guides

Image reference workflows

How to Collect Visual References Across Search Results Without Losing Focus

  • Visual reference search
  • Design research
  • Source tracking

A broad image search can produce hundreds of results and very little progress. The useful task is usually narrower: find a few references that show different visual attributes, then stop and decide. This guide uses a tree-image search as an example. It shows how to make each result page answer a question, hold a small working set beside the browser, preserve original sources and avoid collecting duplicates that only make the board harder to use.

Recorded Linux demo from July 2026: tree results are browsed across a page while FloatBoard holds a mixed working set. Current controls may differ; the recording contains more images than the free plan permits per day.

Search for attributes, not just a noun

“Trees” is a starting category, not a research brief. Write down the attribute you need to study. An illustrator might need a bare winter silhouette, a dense green canopy and one isolated tree with open space around it. A designer might instead need muted color, a clear horizon and a subject that survives a narrow crop. The attribute list tells you why an image deserves attention and when the search can end.

Use separate queries for genuinely different needs rather than endlessly scrolling one generic result page. “Bare trees in fog” and “single tree open sky” are likely to show different compositions. The point is not to game search results; it is to reduce the chance that the first appealing image determines the whole visual direction. Keep the brief visible in a project note while you browse.

Give each result page a short first pass

Give each result page a quick pass with your attribute list in view. On page one, you might find a bare silhouette and a dense green canopy. Mark those two attributes as covered. On page two, look mainly for the missing isolated tree with clear space around it. This makes the next page answer a question instead of inviting another unsorted scroll. If a page has nothing new, do not add an image simply because you spent time looking at it.

Keep a tiny candidate queue in your project note: “A — winter branches, page link”; “B — overhead canopy, page link.” A source link is useful for returning to the exact candidate, but the label should describe its role in the search. When a new image serves B's role better, replace B instead of creating C for the same visual attribute. The queue stays readable even when browser tabs and search-result order change.

Keep only the active working set visible

FloatBoard can hold a small set of images where you can see it while browsing. Add candidates using the workflow in your installed version, then place the window so it leaves enough room to inspect the next source. The recording above shows a tree search in a browser with a FloatBoard window alongside it. That old session already contains mixed references from other searches; it illustrates the importance of clearing or separating unrelated working material before making a new decision.

Assign each candidate a role in your notes: “bare branches,” “colored canopy,” “single tree,” or “strong backlight.” Replace a weaker example when a better one serves the same role. The full recording includes more images than the current free plan's five images per day, so do not use its board count as a plan promise; review the current limits if your task needs a larger set.

Compare the attributes, not the prettiness

Ask the same question of each candidate: What shape is the tree? How much empty space surrounds it? Does the light clarify or hide the branches? Could the composition be cropped into the intended canvas? A visually striking image can still be a poor reference for the specific problem. Write one observation per candidate instead of a star rating; observations explain why an image will help later.

If you need to compare color mood, keep display conditions in mind. Browser thumbnails and floating previews may have been resized or compressed. Use them for broad direction, then open original files in your design workflow for precise color decisions. If you need a form study, crop a legitimate working copy around the silhouette while retaining the uncropped source. Do not let a thumbnail's attractive crop substitute for the composition of the actual source image.

Remove near-duplicates as the search expands

Two files can be different photographs yet contribute the same visual lesson. If both show a dark tree centered against pink sky, compare their role rather than their filenames. Keep the one with clearer branch structure or a composition that fits your canvas. This is a more useful form of deduplication than looking only for identical files. It also keeps one colorful category from crowding out the winter or isolated-tree example your brief still needs.

When an image appears on several result pages, follow the version whose page gives you reliable context and a way back to the original. Note the chosen page link in your candidate queue, then ignore the reposts. Do not let repeated exposure make the same image seem more representative. The board is a small sample of distinct visual attributes, not a popularity tally of what the search engine happens to repeat.

Use a stop rule before opening another page

Stop when each required attribute has a useful example, or when another page is unlikely to change the next design decision. For the tree brief, place one winter silhouette, one leafy canopy and one isolated tree beside a rough sketch. Try the actual line weight, background and crop you plan to use. If the sketch still lacks a form reference, the missing form gives you a precise next query. If it works, more browsing is probably delay disguised as research.

Write a short outcome for each selected image: “bare branches define negative space,” “canopy shows overlapping masses,” and “single tree tests focal placement.” This turns a mixed search into a coherent set of design questions. If the board also contains unrelated code or landscape pictures from an earlier task, mentally separate them or begin a fresh working set for this decision; otherwise the page can look full before your attribute list is actually complete.

Know what the board and search page cannot tell you

A compact board can make very different images look deceptively similar when every tile is small. Before you select an image for a fine-detail study, open it larger and check that the branch, leaf or lighting feature you need is actually visible. For a composition study, test the candidate in your intended crop. A square search thumbnail may hide the empty space that made an isolated tree useful, or create empty space that disappears in the original.

The board also has no knowledge of your categories unless you write them down. Keep the attribute list in the same order as the candidates, or label the candidates in your project note. If you rearrange the working set, update that order. You only need enough organization to answer the current visual question. A separate source-page check is still necessary before publishing an image, but the active search should be driven by the missing visual attributes, not by a growing pile of metadata.

Frequently asked questions

How many images should I take from one result page?

Often none or one. Add an image only when it covers an attribute that is missing or improves an existing candidate's role.

What if my board already has unrelated images?

Start a fresh working set for the new question, or keep a written list of which tiles belong to it. A full-looking board can hide gaps in the actual brief.

How do I know when to stop searching?

Use the selected images in a quick sketch or layout. If every attribute has an example and the test exposes no missing need, stop and make the design decision.