How to Make a Reference Contact Sheet From Video Stills

A contact sheet earns its keep when a reviewer can point at a frame and you can hand back the moment it came from. That round trip is the whole point. Without it you have a collage — pleasant to look at, useless to act on, and guaranteed to produce an email asking you to send "the one with the blue thing again."

So build the return path first, and the grid second. Export clean frames from clips you have identified and are allowed to use. Give each frame a short, stable ID. Keep a map from that ID to the clip, the position, the still file and the credit context. Then arrange the tiles at a size that lets the comparison you are asking about actually be seen.

Name the question before you choose any frames

A contact sheet is an argument about moments. If you cannot finish the sentence "I want the reviewer to decide between…", you are making a mood board, and mood boards invite opinions about taste rather than decisions about a specific relationship.

Name the relationship. Not "good frames from the river shoot" but "where the subject sits in the frame at the start and the end of the move in question, across these three clips." That question tells you how many frames you need — two per clip, one at each end of the move — and it tells you what has to stay visible in each tile. The frame of the subject's shoulders against the top edge is the evidence. The colour of the water is not.

When the question is about motion rather than position, the contact sheet cannot answer it. Speed, ease, how long a hand takes to enter frame — none of that survives a still. What you can do is mark the surrounding interval on the map so the reviewer can go and watch it. The sheet is an index to the clip, not a replacement for it.

One distinction worth holding onto from the start: you are choosing frames from clips you already have and can use. Finding the footage and clearing the rights are separate jobs with separate risks. This job starts once a clip is identified and permitted.

The example used through the rest of this piece

The walkthrough here is invented to show the decisions. No exports were run to produce it, no sheet was generated, and no reviewer responded. Treat the clip names, positions and numbers as declared facts of a scenario, not as measured results.

Three clips, all your own footage or explicitly permitted, each with two useful frames:

Two frames per clip, at the beginning and end of the move in question. Six tiles. The comparison is headroom: how much space sits above the subject at each end of each move.

Export the frame, not the window

The route you want is the one built for it. Adobe's Premiere Pro documentation for exporting a still image, last updated 7 January 2026, describes exporting the frame currently shown in the Source or Program Monitor, with a chosen name, image format and destination. That gives you a file sitting in a folder, named by you, at the shape and size the monitor is showing rather than the shape and size of a window on your desktop. If you want the reference itself: helpx.adobe.com — Export a still image.

The habit to break is the system screenshot. It is fast, and it quietly ruins the sheet. A screenshot carries the playhead, the transport bar, sometimes a cursor and a tab. Its resolution is whatever your monitor happened to be. Its filename announces the exact minute you were working and nothing about what is in the image. And because every screenshot looks alike, a folder of them gives you no way to tell which clip the third one came from.

Whichever route you use, four things belong on the map for every frame:

Clip identity. The actual file or asset name, not a nickname you will forget by Thursday.

Position. A timecode or a frame number, and — this matters more than people expect — which clock that position is on. A still exported from a clip's own media refers to the source. A still exported from the edited timeline refers to the sequence. Those two numbers can diverge badly once a clip is slowed, and someone asking about duration will get the wrong answer if you conflate them.

Rate and retiming. Note the source rate and whether the sequence changed it. In the example, the dye-tank clip was recorded at 50 fps and plays at half speed in a 25 fps sequence. A still taken from that timeline is not a still of the original recording at that instant.

Still type. Was this exported from the clip, or from the timeline? A timeline still can carry sequence-level things — an overlay, a reframe, the timing change — that are not in the source. Check what is actually in the image before you label it as a clean reference.

If you cannot identify where a clip came from, you cannot record a recoverable position, and you should say so on the map rather than guess a timecode that looks precise. "Source unknown, position unknown" is honest and still useful for an illustration. A fabricated timecode is worse than a blank, because someone will trust it.

Finally, keep the full-size export separate from the review thumbnail. They serve different jobs and one should never be regenerated from the other.

Build a sheet whose labels survive

Two layout decisions determine whether the sheet works, and both follow from the question you named at the start.

Tile size. Set it by the finest difference your reviewer needs to see. If the comparison is a composition — where the top edge sits relative to a shoulder — modest tiles are fine. If it is a shadow edge, a texture or a colour shift, small tiles will erase exactly what you are asking about, and the reviewer will answer a different question than the one you asked.

Aspect ratio. Preserve it, or declare the crop. Silently cropping every frame to one shape is the most common failure in reference sheets: the reviewer ends up judging your crop instead of the composition you meant to show. The example has three different shapes, and that costs something real. Keep every tile the same height and the vertical frames become narrow. Keep them the same width and the vertical frames make the sheet very tall — a 9:16 frame drawn three inches wide is more than five inches tall. Either the vertical pair or the horizontal pair ends up smaller than the other. That is fine if you decide which pair your question depends on, and it is a genuine problem if you pretend both pairs are equally legible. When it matters, split the sheet by shape.

For the grid itself, the tooling is the easy part. Capture One's documentation for Contact Sheets, updated 23 April 2026, describes a review grid of selected images with adjustable layout, optional filenames and PDF or JPEG export. It also warns that long filenames can be truncated. That warning deserves more attention than it usually gets, because truncation does not simply shorten a label — it can make two different frames carry identical-looking labels.

Look at the dye-tank file in the example. Printed in full it is long enough to crowd any tile. Truncated, both of its frames may show the same string: 2026-03-14_studio_dyetank_hand-entry_seg-03_camB.... The label has not become shorter and less helpful. It has become confidently wrong, which is worse than no label at all, because it looks like information.

The fix is not a better filename. It is to stop letting the filename carry the identity. Use a short ID — RIV-01, DYE-01, CT-01 — and keep the real names in the map, which has room for all of them. In the running example the map looks like this:

The export files can simply be named after the IDs: RIV-01.png, DYE-01.png. Nothing in that column truncates, because nothing in that column is trying to hold the whole story.

Then verify. Look at the exported sheet itself, at the size your reviewer will actually see it, rather than trusting that the application's preview matches what it wrote out. Check that every label is the one you meant, that the IDs appear in the order the map lists them, that no tile has been quietly cropped, and that the difference you are asking about is still visible after the file was saved. If the export is a JPEG, that last check is not paranoia — compression is designed to throw away small differences, and a subtle edge is a small difference. Reviewers on phones are also a real case: a PDF that reads perfectly on your monitor can be illegible at a glance on a small screen.

Send it out so the answer comes back usable

The sheet's whole purpose is to produce a selection you can act on, so ask for the selection in the form you can act on. Two sentences of instruction cover it: reply with the IDs, and say why. The rationale is not decoration. It tells you whether the reviewer chose the frame for the composition, the subject's position or something incidental — and those lead to different next steps.

When a response is ambiguous, resolve it before you touch anything. "The third one" is not an ID. It might mean the third tile in reading order, the third clip, or a number from a list the reviewer has open in another window. Turn every informal reference into an ID in writing and get it confirmed. The alternative — quietly substituting what you believed they meant — is how a review turns into an unrequested re-edit.

When you hand the selection back, hand back the original exported still. Do not send an enlargement cropped out of the contact sheet. The sheet's tiles were sized to be reviewed, and their pixels were spent on that job; the still is the thing that was sized to be used.

And when a selection raises a question about timing rather than appearance, the still is the wrong instrument. Go back to the interval marked on the map and look at the clip.

Keep the pick set separate from permission

A selection is a decision about which reference to work from. It is not clearance, and it is not an approval, and the artefacts should not blur the difference. The map is where that distinction lives: a credit and rights column, filled in from the start, travels with the sheet and with the stills, so nobody downstream has to reconstruct where a frame came from or what it may be used for.

Store the map, the sheet and the full-size stills together, as one set. A sheet without its map is a collage again.

One exception is worth naming. If a filename or a tile label carries something private — a client name, an unreleased product, a person who should not be identified — replace it only through an approved, reversible aliasing scheme, and keep the key wherever the map lives. Do not blank out provenance to make the layout tidier. The tidiness is worth nothing if it costs you the return path.

Follow one ID all the way back

Here is the return trip, in the invented example, from a reviewer's sentence to a usable asset.

The message comes back: "Third one's the hand position we want — can we keep that? And how long is the hand in frame there?"

You resolve "third one" to an ID and confirm it. It is DYE-01. You open the map and read the row:

• Clip: the long-named dye-tank file.
• Position: timeline 00:01:02:10.
• Notes: this is a timeline still, not a source frame; the clip was recorded at 50 fps and is running at half speed in a 25 fps sequence.
• Credit: your own footage, no external clearance involved.

So you send back three things: the sheet the reviewer already has, the map row so the ID is traceable, and the full-size exported still file for DYE-01 — not a crop lifted out of the sheet.

The second question is the one the sheet cannot answer, and it is worth being exact about why. Nothing on the map records when the hand leaves. DYE-01 and DYE-02 bracket the move, not the hand's whole stay in view — the hand may be on screen well before the first frame and after the second. So the interval the map hands you is the move: 00:01:02:10 to 00:01:22:10 on the timeline, twenty seconds. And that figure still has to be converted. The clip runs at half speed in a 25 fps sequence, so twenty seconds of timeline is ten seconds of what the camera actually recorded. Say "twenty seconds" without naming the clock and you are wrong by a factor of two, confidently, because the number came straight off the sheet. Call it the hand's time in frame and you have answered a question this pair does not cover. For that, the still is the wrong instrument: open the clip at the interval the map marks and watch where the hand goes.

That is the demonstration worth building the whole method around. The grid is what people look at. The map is what turns a look into a file, a moment and an interval someone can actually work from — and it is the part you find out you needed only after the attractive version has already been sent.

Frequently asked questions

What should I build first for a reference contact sheet, and why?

Build the return path first, and the grid second. A contact sheet earns its keep when a reviewer can point at a frame and you can hand back the moment it came from. Without that, you have a collage that is pleasant to look at but useless to act on, and it invites an email asking for the one with the blue thing again. Export clean frames from clips you have identified and are allowed to use, give each frame a short stable ID, and keep a map from that ID to the clip, position, still file and credit context. Before choosing frames, name the question: finish the sentence, I want the reviewer to decide between... If you cannot, you are making a mood board.

How should I export a still, and what should the map record for each frame?

Use the route built for it: Premiere Pro's documentation describes exporting the frame currently shown in the Source or Program Monitor, with a chosen name, image format and destination. Avoid the system screenshot, which carries the playhead, transport bar, cursor or tab, takes its resolution from your monitor, and has a filename that says nothing about the image. For each frame, the map should record clip identity as the actual file or asset name, position as a timecode or frame number and which clock it is on, rate and retiming such as a 50 fps recording played at half speed in a 25 fps sequence, and still type, meaning whether it came from the clip or the timeline. If you cannot identify where a clip came from, say source unknown and position unknown rather than guess a timecode. Keep the full-size export separate from the review thumbnail.

Why use short IDs instead of filenames as labels on the sheet?

Capture One's Contact Sheets documentation warns that long filenames can be truncated. Truncation does not simply shorten a label; it can make two different frames carry identical-looking labels. In the example, both dye-tank frames may show the same truncated string, so the label becomes confidently wrong, which is worse than no label at all. The fix is not a better filename. Use a short ID such as RIV-01, DYE-01 or CT-01, and keep the real names in the map, which has room for all of them. Export files can be named after the IDs. Then verify the exported sheet at the size your reviewer will see it, checking that labels are the ones you meant, IDs appear in map order, no tile was quietly cropped, and the difference you are asking about is still visible after saving.

How should I handle mixed aspect ratios and tile size?

Set tile size by the finest difference your reviewer needs to see. If the comparison is a composition, such as where the top edge sits relative to a shoulder, modest tiles are fine. If it is a shadow edge, texture or colour shift, small tiles will erase exactly what you are asking about. Preserve aspect ratio or declare the crop; silently cropping every frame to one shape makes the reviewer judge your crop instead of the composition. Mixed shapes cost something real: keep every tile the same height and the vertical frames become narrow; keep them the same width and the vertical frames make the sheet very tall. Either the vertical pair or the horizontal pair ends up smaller. That is fine if you decide which pair your question depends on, and a problem if you pretend both pairs are equally legible. When it matters, split the sheet by shape.

What should I send with a selection, and how does an ID trace back to a usable asset?

Ask for the selection in the form you can act on: reply with the IDs and say why. The rationale tells you whether the reviewer chose the frame for composition, subject position or something incidental. Resolve ambiguous references before touching anything; the third one is not an ID and might mean a tile, a clip or something in another window. When you hand the selection back, send the original exported still, not an enlargement cropped out of the contact sheet. If the question is about timing rather than appearance, the still is the wrong instrument; go back to the interval marked on the map and watch the clip. Keep the map, sheet and full-size stills together. In the example, DYE-01 resolves to a timeline still at 00:01:02:10, source 50 fps at half speed, own footage; the move brackets 00:01:02:10 to 00:01:22:10 on the timeline, twenty seconds, which is ten seconds of what the camera actually recorded. Saying twenty seconds without naming the clock is wrong by a factor of two.

Further reading

• Mixed Frame Rates in a Pitch Reel: Preserve the Event's Duration—or Deliberately Change It? — Go here when a slowed clip makes a duration ambiguous: it separates preserving the event's recorded length from deliberately changing its speed.