Prepare a VFX Shot Turnover for a Commercial Pitch Test

A turnover is a set of claims about time, and every one of them has to be checkable by someone who isn't in the room. Which frames of which files become which frames of which shot, and what happens at the joins. Written down, because the compositor can't ask you at 11 p.m.

For a pitch test the material is small — often two elements and three seconds — which is exactly why it goes wrong. A folder of plausible clips looks finished. It isn't. What makes it finished is that a stranger can reconstruct the shot from your notes, including the parts you couldn't supply.

The example below is invented. No elements were shot, no package was sent, no comp came back, no timeline was opened. The frame numbers are kept consistent so you can follow the arithmetic, and the arithmetic is the part worth copying.

Fix the shot and reference edit being handed over

Before you collect anything, name two things.

The shot. A stable identity — SH040 — that survives renaming, re-cutting and re-versioning. Not "the can shot" and not "hero_comp_final2."

The reference edit. A revision you can point at: PITCH_REF_v04, 24 fps, with the shot sitting at a known position. In the invented example, SH040 occupies reference timecode 00:00:31:12 through 00:00:34:11 — frames 756 to 827 of the edit, seventy-two frames at 24 fps. State the arithmetic rather than the vibe: shot frame 1 is edit frame 756, shot frame 72 is edit frame 827. That single sentence is what lets a returned file be placed without a phone call.

Then declare your numbering convention. "Frame 22" means nothing until you say whether you count shot-relative frames from 1, reference-edit frames, or source timecode. Two of those will look identical in a filename and land thirty frames apart in the cut.

Include adjacent context, but only the context that explains the effect. What precedes the shot, what follows it, where it lands in the thirty-second spot, how long the hand stays in frame. A compositor needs to know why the shot exists and what it has to hand off to. That is a few sentences of editorial description, not the client deck. Sharing scope is a real constraint on a pitch: send the frames and the explanation the collaborator needs to solve this one problem, and leave the rest of the treatment where it is.

Keep the viewing material and the measuring material separate, and say which is which. The reference movie is flattened: it carries the temp grade, the temp scale on the hand, and a raster that isn't the source raster. It is the best available statement of intent. It is not a record of source ranges, and no one should ever derive a handle count from it.

And when the cut changes, change the name. A new reference revision with the shot at a new position is PITCH_REF_v05 and a revised turnover, not a quiet replacement of the file behind the old name. Someone will have the old one open.

Map each source element onto the shot timeline

Now the part that a folder cannot do. For each element, record four things: the actual file, the used source range, where that range sits on the shot timeline, and the frame that anchors it there.

In the invented example, SH040 is built from two elements. A product plate: a can on a surface, no hand. A separately filmed foreground action: a hand against greenscreen. They were shot at different times, on different cards, with no shared timecode. Their source clocks start an hour apart.

Look at the two mapping rules: minus eleven and plus twelve. Subtract one from the other and you get a twenty-three-frame difference between two files whose first frames are each labelled zero. If you had trusted the frame numbers to mean the same thing in both files, the hand would swing almost a second out of position.

The elements share a moment. It is the frame where the fingertips meet the can, and it is not recorded in either file's metadata — it is a decision. The contact sits at shot frame 22, which is plate source frame 33 and hand source frame 10. Three numbers, one instant. Write all three, because checking the alignment later means checking that instant, and the check only works if both sides of the correspondence are written down.

The plate's take was rolled well before the action, so its used range begins at the shot's first frame, with twelve frames of unused take (source 0–11) sitting in front of it. The hand's take was rolled late, and its used range doesn't begin until shot frame 16 — sixteen frames into the shot's own history. That asymmetry is normal and harmless as long as it's stated. What is not harmless is a filename that says hand_v3_final and communicates nothing about where the hand belongs. Filenames carry content; the turnover carries time.

Separate available handles from requested latitude

A handle is a frame that exists. Latitude is a frame somebody wishes existed. Keep them in different columns.

The used range is what the reference cut shows. Everything in the source file outside that range is available handle — real frames, already photographed, sitting on the card. Count them, then compare with what the compositor asked for.

In the example, the compositor asked for twelve frames of handle on each end of each element.

The plate meets it. Used range starts at source frame 12, so source frames 0–11 sit in front of it: twelve frames of head handle, exactly the request. Behind the used range, source frames 84–109 give twenty-six frames of tail handle.

The hand does not. Its used range starts at source frame 4, so only source frames 0–3 sit in front of it — four frames of head handle against a request for twelve. The take was rolled late and the hand was already moving. Subtract: eight frames short. At the tail the hand is generous, source frames 52–119 giving sixty-eight frames.

Eight missing frames is not a crisis. It becomes one only if it goes unmentioned, because the compositor will ask for twelve, receive a file with four, and either pad it or say nothing. Both outcomes are worse than the truth. Do not duplicate a frame to reach the count. Do not hold the first frame and present the hold as material. If the missing frames are later manufactured — a repeat, a re-time, a painted extension — that is a treatment, and treatments get declared, with the frame range and the method, in the note and on the return.

The same discipline applies to the reference's temporary work. In the example, PITCH_REF_v04 has a contrast-and-warm grade burned into it, the hand is scaled to 96% with a feathered edge, and the whole screener has been transcoded to 1920×1080 while both sources are 2048×1152. Every one of those is a decision someone made in the edit, not a property of the footage. Say so explicitly: the sources are untreated, the grade is not delivered, the 96% placement is one option and not a framing instruction, and the screener's raster is a viewing transform that will not match the return.

A truthful note for the hand element reads something like:

SH040_HAND_A002_C014.mov, 24 fps, 2048×1152, source first frame 02:00:00:00. Used range: source 02:00:00:04 – 02:00:02:03, mapped to shot frames 16–63. Contact pose at source frame 02:00:00:10 maps to shot frame 22 (shot = source + 12). Available handles: head 4 frames (02:00:00:00 – 02:00:00:03); tail 68 frames (02:00:02:04 – 02:00:04:23). Requested head handle of 12 frames is not available. Shortfall: 8 frames. No padding, duplication or retime applied. Any extension of the hand's entry is a treatment and must be labelled as one on return. Source is untreated. The reference grade, scale and screener raster do not describe this file.

That last line does more work than it looks like. It stops the compositor from matching a grade that was never meant to be matched and from inheriting a framing that was only ever a temp placement.

Agree on the return and inspect its reinsertion

Ask what the recipient wants before you tell them. Whether they want the plate and hand as separate files or the composite only; whether they want a movie or a numbered sequence; the raster they want to work at; the number of frames they intend to send back; and how they want the frame numbering expressed. On a pitch test there is no standard to appeal to, and a recipient who has asked for something specific has already answered the question you were about to guess at.

In the example, the agreed return is: shot frames 1–72 inclusive, seventy-two frames, 2048×1152, 24 fps, the first delivered frame being shot frame 1 and saying so. Named with the shot, the reference revision it answers, and a version. Any temporary treatment inside it identified in a short written note.

The first draft of that instruction said "send back the comp with twelve frames of handle either side, 1080p, starting at frame 1." It is corrected on three counts. It names a raster, 1080p, that is neither the source raster nor a stated viewing transform, and never says what the sources are. It does not say whether frame 1 means the shot's first frame or the file's first frame. And it asks for handle frames on the return, which the hand element cannot supply — you cannot request latitude back from a collaborator you failed to give it to. The return covers the shot; the shortfall stays in the note.

Then the check, which is the step most often skipped because the file opens and looks right.

Place the returned file at edit frame 756, which is 00:00:31:12, which is shot frame 1. Then step to the contact.

The contact belongs at edit frame 777 — 00:00:32:09, shot frame 22. Verify that specifically. If it lands at 778, 00:00:32:10, the return is one frame late, and the most likely reason is that its first frame was a head-handle frame rather than shot frame 1. When a file beginning with a handle frame is placed at the shot's start, the handle frame occupies shot frame 1's position and everything after it shifts one frame down the timeline. The file opened. It just doesn't fit where you put it.

Either offset the return by one frame deliberately and record the offset, or ask for a re-delivery whose first frame is shot frame 1. Do not nudge it until it "looks better" — that is how a one-frame error becomes an unexplained two-frame error at the tail.

Then the rest of the check: the shot's last frame at edit frame 827, 00:00:34:11, and the join against whatever follows it. The hand's entry at the head of its used range and its exit at the tail, for continuity. Sync against the audio. Whether anything in the returned frames is still temporary and still labelled as such. Whether the delivered range is what was agreed or has quietly grown.

None of that is difficult. All of it requires that the numbers were written down before the files left.

What the finished turnover looks like

Someone who has never spoken to you can state, from your notes alone, which frames of which files make SH040, where the contact sits in each source and in the edit, how many frames of real handle exist at each end of each element, what in the reference was temporary, and what is expected to come back and where it goes. They can also see the eight frames that aren't there. That is the point: the missing handle is more useful to your collaborator in writing than a folder that appears complete.

The separation of a shot from its elements, of in-cut duration from handles, and of returned versions from the material they were cut from is described in Evan Schiff's feature turnover guide, a first-person account of feature editorial workflow rather than a universal standard for commercials; the reading behind this article is from September 2026 and the linked reference images and native turnover files in it were not examined. Worth reading in full for the editorial-side discipline, with the scope limit kept in view.

The materials this example describes do not exist yet. Producing original or authorized sources, confirming the requested outputs with an actual compositor, and checking a real return against a real reference edit are the work that remains, and they are the work that would turn this method into evidence rather than a procedure. Until then the honest position is the same one the turnover itself takes: here is what is known, here is what is missing, and here is exactly where to look to find out whether it lines up.

Frequently asked questions

What should a VFX shot turnover let a stranger reconstruct?

It should state which frames of which files become which frames of which shot and what happens at the joins. That means a stable shot identity, a named reference edit revision with the shot at a known position, the arithmetic linking shot frames to edit frames, the numbering convention, each element's file, used source range, placement on the shot timeline, and anchor frame, the contact frame across sources, available handles versus requested latitude, temporary work in the reference, the agreed return, and where the return is reinserted. The point is that a compositor cannot ask you at 11 p.m., so the notes must be checkable without you.

Why keep the reference movie separate from the measuring material?

The reference movie is flattened and carries temporary decisions: in the invented example it has a contrast-and-warm grade burned in, the hand is scaled to 96% with a feathered edge, and the screener is transcoded to 1920×1080 while both sources are 2048×1152. It is the best available statement of intent, but it is not a record of source ranges, and no one should derive a handle count from it. The sources are untreated, the grade is not delivered, the 96% placement is one option rather than a framing instruction, and the screener raster will not match the return.

How do available handles differ from requested latitude?

A handle is a frame that exists in the source file outside the used range. Latitude is a frame somebody wishes existed. Count the real frames, then compare with what the compositor asked for. In the invented example the plate has twelve frames of head handle and meets a request for twelve; the hand has only four frames of head handle against the same request, an eight-frame shortfall. Do not duplicate a frame to reach the count, do not hold the first frame and present the hold as material, and declare any manufactured extension as a treatment with its frame range and method in the note and on the return.

Why write the contact frame in all three numbering systems?

The contact is the moment the fingertips meet the can, and it is not recorded in either file's metadata; it is a decision. In the invented example it sits at shot frame 22, plate source frame 33, and hand source frame 10. Writing all three lets someone check alignment later, and the check only works if both sides of the correspondence are written down. It also exposes why frame numbers do not mean the same thing across files: the mapping rules are shot = source minus eleven for the plate and shot = source plus twelve for the hand, a twenty-three-frame difference between two files whose first frames are each labelled zero.

What does reinsertion checking look like after a return?

Place the returned file at edit frame 756, which is 00:00:31:12, which is shot frame 1, then step to the contact. The contact belongs at edit frame 777, 00:00:32:09, shot frame 22. If it lands at 778, 00:00:32:10, the return is one frame late, most likely because its first frame was a head-handle frame rather than shot frame 1. Either offset the return by one frame deliberately and record the offset, or ask for a re-delivery whose first frame is shot frame 1; do not nudge it until it looks better. Then check the shot's last frame at edit frame 827, 00:00:34:11, the join against what follows, the hand's entry and exit, sync against audio, whether anything is still temporary and labelled, and whether the delivered range is what was agreed.

Further reading

• Move a Pitch-Reel Edit to Another Editor Without Losing Its Timing — Read as a parallel before handing a shot across: a handoff can open cleanly and still be wrong, so verification compares source boundaries rather than duration.
• Send a Smaller Editable Pitch Reel: Trim the Media or Keep Complete Sources? — Compare when handle expectations meet an editing package: it explains retained source extents and why original masters stay separate from any trimmed deliverable.