How to add a figure to a shot while the camera is panning

Last updated 2026-09-22 · 8 min read

What is the short version?

You want something standing in a shot where the camera moves past it. The two things that break it are drift — the figure sliding against the floor as the camera pans — and the appearance, which reads as an effect rather than a discovery. Vizard Agent handles both before rendering the full shot.

1. Let Vizard Agent set the figure on a real mid-pan frame first.
2. Ask for the reveal to come from behind existing foreground.
3. Check any flagged defect on a real crop before fixing it.

What do you need before you start?

The shot, and a clear description of the figure — height, posture, where it is standing and what it is standing on. Scale is the detail people leave out and the one that gives an insert away, so say how tall it should be relative to something visible in frame.

You should also say what must not change. In the session this is drawn from the brief was explicit that the aspect ratio, grade, lighting and environment were all off limits, which rules out the easy cheats of re-grading the whole shot to hide a mismatch. Vizard Agent treats that as a constraint on the method, not just on the output.

What do you type into Vizard Agent?

Describe the figure and the moment. The moment matters as much as the figure, because a pan gives you a choice of when it comes into view, and that choice is most of the effect — the same figure revealed a second later is a different shot.

When the camera pans left, there should be a figure standing by the column — about two metres tall, completely still, lit by the same light as the rest of the courtyard. Do not change the grade or the framing.

Note the stillness. A figure that does not move is far easier to hold against a moving camera than one that does, and it is usually the more unsettling choice anyway. Vizard Agent will say when a described motion is going to fight the tracking.

What does Vizard Agent actually do?

It builds a still before it builds a shot. A frame from the middle of the pan gets pulled out, the figure is generated standing on the actual ground in that actual light, and that composite is reviewed for scale, grounding and lighting match. Only once the still is right does anything get applied across the moving frames.

The working order in that session was:

• Sampled the pan densely to see the camera move and the light in the scene.
• Inspected the composition at several points through the pan and picked a frame.
• Generated the figure onto that frame, standing on the real tiled ground.
• Checked that still for placement, scale, grounding and lighting before going further.
• Ran the integration through the moving shot with scene-matched tracking.
• Reviewed the result frame by frame for ground contact, motion blur and tracking continuity.

The still is cheap and the moving render is not, which is the practical reason for the order. The better reason is that a mismatch is much easier to see on one large frame than in motion.

What does the result look like?

The figure stays where it is in the world while the camera moves past it, keeps contact with the ground, and carries the same light and motion blur as everything else. Vizard Agent checks the first reveal frame, a mid-pan frame and a later frame rather than watching it once, because drift accumulates and is invisible at the start.

The reveal is the part you notice. In that session the first version had the figure simply appear, which read as a composite; the re-render brought it into view from behind a real foreground column, so the camera discovers it. That single change is the difference between an effect and a shot.

When does this not work well?

When there is nothing in the foreground to reveal from behind. A clean open pan gives you no occluder, and then the figure has to be in frame from the first moment or fade in, and fading in always looks like an effect. Vizard Agent will tell you what is available rather than inventing an occluder.

It also struggles on a fast pan with heavy motion blur. The blur on the added figure has to match the blur on the scene, and at high pan speeds a small mismatch is very visible. A slower move, or a shorter section of the move, gives a much better result.

And a figure that has to interact with the scene — casting a shadow onto a specific surface, or being partly behind something that itself moves — is a harder job than one standing still in open ground. It is doable and it takes more passes.

How do you fix a result that came back wrong?

Say which frame and what is wrong in it. "It floats at about four seconds" is actionable; "it does not look real" is not. Vizard Agent pulls that frame and the untouched source frame side by side, which usually settles what actually changed.

If a review flags something, check it is visible before accepting the fix. Late in that session a review reported the figure overlapping a barrier post, and Vizard Agent compared the source and final crops at that exact moment before doing anything, because the correction would have been a global change to a shot that was otherwise right. A flagged defect you cannot see on a real crop is not worth a risky fix.

If the reveal is the problem rather than the tracking, say so separately. Those are different renders, and Vizard Agent can change when and how the figure comes into view without redoing the tracking work.

How does Vizard Agent compare to doing it yourself?

This is compositing, and compositing by hand means tracking the shot, matching the light, matching the grain and building an occlusion mask for the foreground. The skill is real and so is the time. Most people attempting it without that background end up with the pop-in version, because the tracking is the part that is hardest to fake.

What Vizard Agent changes is the order and the checking. Establishing the match on a still first is what a compositor does too; reviewing three separate frames through the move for drift, and refusing to apply a global correction for a defect that is not visible in a crop, is the discipline that usually gets skipped when the shot already looks close enough.

Common questions

Does the figure have to be still? No, but still is much easier to hold against a moving camera, and Vizard Agent will say when described motion fights the pan.

Can I use my own image of the figure? Yes. Send it and Vizard Agent matches it to the scene's lighting rather than generating a new one.

How long a shot can this work on? A few seconds is the reliable range. Vizard Agent will tell you when a shot is long enough that tracking will drift.

Will the grade change? Not unless you ask. Vizard Agent matches the figure to the existing grade rather than regrading the shot.

What about the shadow? Ask for it explicitly. A figure with no shadow on a lit floor is the second most common giveaway after scale.

Can it stand behind something? Yes, and it usually should. Vizard Agent uses real foreground as the occluder.

Does the audio change? No. Vizard Agent leaves the original audio unless you ask for a cue.

How do I judge scale? Name something in frame. Vizard Agent sizes the figure relative to it rather than to an absolute number.

What if the ground is uneven? Say so. Vizard Agent places the contact point on the actual surface in the chosen frame.

Can I see the still before the full render? Yes, and you should. Vizard Agent builds it first for exactly that reason.

Does this work on a handheld shot? It is harder. Vizard Agent can stabilise first, though that changes the framing slightly.

What if it drifts only at the end? Say where. Vizard Agent re-tracks the later section rather than the whole shot.

Can I add two figures? Yes. Vizard Agent matches each to the scene separately, since they will be lit differently.

Will it survive compression? The match will. Fine grain detail is what a heavy encode softens first.