How to cut out every crowd shot in a multicam recording
Tell Vizard Agent the rule is absolute and includes spectators in the background. A multicam edit cuts to the crowd constantly, and the shots that survive a quick pass are the wide ones where the audience is small but present, and the tail of each section where the camera drifts off the stage.
What is the short version?
You have a recording of a performance that keeps cutting to the audience, and you want clips of the act alone. The rule is simple to state and it has to be checked at a much finer resolution than you would expect. Vizard Agent maps the stage-only stretches before cutting.
- Tell Vizard Agent to exclude every audience shot, including distant ones.
- Expect a second-by-second map rather than a quick selection.
- Say what the subject actually is, so a close-up does not ruin a clip.
What do you need before you start?
The recording, and the rule stated at its strictest. "No reaction shots" leaves room for a wide angle with a visible front row; "not even a distant shot of spectators" does not, and the second version is what you meant.
Say what the act is. It sounds like context and it is a selection criterion — for a magic act the hands are the content, so a shot that crops a hand is unusable however good it looks, and Vizard Agent cannot infer that from the footage.
What do you type into Vizard Agent?
Give Vizard Agent the exclusion and the subject in the same brief. The exclusion tells it which shots to drop outright; the subject tells it which of the remaining shots are genuinely usable, and those are two different filters doing two different jobs.
Make three vertical clips of the performance. Exclude all audience reaction shots, crowd shots, applause and cutaways to spectators. Start and end on complete moments.
Then add the part that brief is missing: what must stay in frame. In that session the hands turned out to be the real constraint, and it cost several rounds to establish.
What does Vizard Agent actually do?
It escalates from sampling to a manual map. A contact sheet finds the obvious stage sections quickly, and then the awkward truth appears — the wide shots contain the audience too, and so do the last seconds of most sections.
In that session the passes ran:
- Built an overview to understand the staging, then pulled frames from the stage sections only.
- Checked the final block second by second to exclude even distant shots of spectators.
- Resolved a disputed wide angle in the finale by stepping through it frame by frame.
- Swept the whole source second by second to build a reliable map of stage-only intervals by hand, reviewing it in four parts.
- Found the source carried its own burnt-in text and interface elements, and rebuilt clean versions without them.
- Removed two remaining tails where a reaction shot had survived at the end of a section.
Then two separate problems: the audio jumped at the visual joins, so the clips were rebuilt on continuous source music; and the performer's hands kept leaving frame, so every close-up and borderline shot was dropped until all three clips held both hands.
What does the result look like?
Clips of the act itself, with nobody in the audience visible at any point in any of them, and the original performance audio running continuously underneath rather than jumping at every visual join the way it did in Vizard Agent's first pass.
They also do not overlap. Once the stage map existed it was straightforward to make each clip about one routine and confirm no footage appeared twice — which is a different requirement from the exclusion rule and worth asking for by name.
When does this not work well?
When the stage and the audience share a frame throughout. A recording shot from the back of the room has the crowd in every wide shot, and the only stage-only material is whatever close work exists. Vizard Agent will tell you how little that leaves.
The rule also fights the coverage. Excluding the crowd removes the shots an editor would use to breathe between moments, so the clips end up denser than a normal performance edit — which is usually what was wanted, and is worth knowing.
And a tight rule plus a tight subject constraint can leave nothing. If every wide shot has spectators and every close-up crops a hand, there is no valid shot, and the honest answer is which rule to relax.
How do you fix a result that came back wrong?
If a spectator survived somewhere, give Vizard Agent the timestamp rather than a general note. It re-checks that one window at a finer interval instead of re-selecting the whole clip, because a miss like this is nearly always a boundary problem rather than a badly chosen shot.
If it is at the end of a section, say so. Tails are the characteristic failure — the camera leaves the stage before the cut does — and trimming the out-point is the fix.
If a shot is technically legal and still unusable, name what is missing from frame. That is how the hands constraint got established in this session, and it changed the selection more than the exclusion rule did.
How does Vizard Agent compare to doing it yourself?
By hand this is scrubbing, and it is the kind of scrubbing that produces mistakes: you check the middle of each section, the clips look right, and a front row appears three frames before a cut in the delivered file.
The escalation to a hand-built map is what makes it reliable, and it is boring enough that nobody does it voluntarily. The useful part is that once the map exists it is reusable — the non-overlapping clips, the thematic split and the continuous audio all came out of the same map rather than from three more passes.
Common questions
Why does a coarse pass miss shots? Because the audience appears in wide shots and at the tails. Vizard Agent sweeps finely for that reason.
How fine does the sweep need to be? Second by second, and Vizard Agent goes to half-second where a boundary is disputed.
Can it keep the applause audio? Yes. Vizard Agent can exclude the picture and keep the sound continuous.
Will the clips overlap? Not if you ask. Vizard Agent confirms no footage appears twice.
What about burnt-in text in the source? Vizard Agent found and removed it here by re-framing the clean area.
Why did the audio jump? Because each visual join cut the music. Vizard Agent rebuilds on a continuous track.
Can it tell which act is which? Yes. Vizard Agent maps the routines so each clip covers one.
What if a close-up crops the subject? Say what must stay in frame and Vizard Agent drops those shots.
How many clips can it make? As many as the stage-only material supports, which Vizard Agent will say is usually fewer than you hoped.
Does it need the whole recording? Yes, so Vizard Agent can build the map. Sending only the good bits defeats the purpose.
Can I get the map itself? Yes. Ask Vizard Agent for the stage-only intervals as a list.
What about a single-camera recording? Easier for Vizard Agent, since there are no cutaways to exclude.
Will captions be added? Only where the speech is clear. Vizard Agent checks before adding them.
Can it reframe to crop the audience out? Sometimes, at the cost of the stage width. Vizard Agent will show you the trade.