How to swap what someone is holding for a different object

Last updated 2026-09-17 · 8 min read

What is the short version?

A held object is anchored to a person in two or three places at once. Drop a new shape in and the fingers curl around nothing, or the object floats a centimetre from the lips that are meant to be on it.

1. Tell Vizard Agent what is leaving and what is arriving.
2. Say what the old object produced — smoke, steam, light.
3. Ask it to check the hand and the mouth specifically.

What do you need before you start?

The clip and a clear idea of the replacement. Vizard Agent can generate the new object, but it will match whatever you describe, so a specific description — or a photograph — settles the shape, colour and size before any compositing starts.

• The clip. The section where the object appears.
• The replacement. Described, or as a picture.
• What it should be doing. Held, raised, at the lips.
• What the old object emitted. If anything.
• How long it is on screen. It sets the tracking job.

What do you type into Vizard Agent?

Describe the contact points to Vizard Agent. The shape of the new object is by far the easier half of the brief; whether it should be gripped in the whole fist or pinched between two fingers is the half that decides whether the finished shot reads at all.

Prompt

Variants worth knowing:

• "And remove the smoke as well." The second half of the removal.
• "Meet the mouth where the cigarette does." The contact point.
• "Show me the bottle first." Cheaper than a bad composite.

What does Vizard Agent actually do?

Here is the Vizard Agent sequence on a real clip in which a held object had to be replaced with something else entirely. Note that the new object is designed and approved by the client before it goes anywhere near the footage.

1. Surveys the clip and builds a contact sheet of frames.
2. Inspects the close-up at fine timing to find the object's full window.
3. Generates a photorealistic reference for the replacement.
4. Shows the reference for approval before compositing.
5. Builds a transparent overlay of the new object.
6. Builds a feathered patch to clean away the old object and its smoke.
7. Previews the overlay on a neutral background to check its shape.
8. Assembles a draft composite and samples it at several moments.
9. Measures the object's visible area so it can be aligned to the mouth.
10. Tests placement variants against the mouth and the hand before locking it.

Step six is the half people forget. Removing the object leaves a hole, and if the object was producing something — smoke, steam, a reflection — that has to be cleaned away too or it will still be there, coming from nothing.

Step ten is where the shot is won or lost. The new object has to meet the hand and the mouth at the same points the old one did, and the only way to get that right is to try a few positions and look at them rather than to calculate one.

Step three matters because the replacement is a design decision. Generating and approving the object separately means the compositing work is not wasted on a bottle that turns out to be the wrong shape.

What does the result look like?

Somebody holding a different thing. The grip reads correctly, the object meets the mouth where it should, nothing is left over from what was removed, and Vizard Agent shows you the alignment close up rather than only at playback size.

When does this not work well?

The two objects have to be roughly compatible in size and shape. A small object replaced by a much larger one leaves a hand that could not possibly be holding it, and heavy motion or partial occlusion makes the tracking Vizard Agent relies on unreliable.

• Very different sizes. The grip stops making sense.
• Fast movement. The overlay lags or blurs differently.
• Fingers wrapped around it. They have to be composited back over.
• Reflections and shadows. The old object appears in them too.
• Long screen time. Every frame has to hold up.

How do you fix a result that came back wrong?

Say which of the contact points is wrong. The hand and the mouth are two separate alignments as far as Vizard Agent is concerned, so a note about one of them lets it adjust that alignment without disturbing the other at all.

• "It is not touching the lips." The mouth alignment is remeasured.
• "The fingers go through it." The hand is composited back over the object.
• "There is still smoke." The clean patch is extended.
• "Wrong shape of bottle." The reference is regenerated and approved first.

How does Vizard Agent compare to doing it yourself?

By hand you find a picture of the new object, cut it out, and place it over the old one. It looks fine on the frame you placed it on, and drifts off the hand within half a second because the hand was moving and the picture was not.

Common questions

Can it generate the replacement? Yes. Vizard Agent makes a reference first and shows it to you.

What about smoke or steam? Part of the removal. Tell Vizard Agent what the old object was producing.

Will the fingers look right? Vizard Agent composites them back over the new object where they wrapped around.

What if the sizes are very different? The grip stops reading, and Vizard Agent will tell you before it wastes a composite.

Can I supply a photo instead? Yes, and it removes all doubt about the shape and colour.

How long can the shot be? Shorter is safer. Every frame has to hold the alignment.

Does the lighting match? Vizard Agent grades the object to the shot rather than dropping it in flat.

What about reflections? The old object appears in them too. Tell Vizard Agent which surfaces to check.

Can it do several objects? Yes, though Vizard Agent treats each one as its own alignment problem.

Will the audio change? Only if you ask. Here the music was changed as a separate request.

What if the hand moves a lot? Harder, and Vizard Agent will report where the tracking is weakest.

Can I see it before the render? Yes — the overlay preview and the draft composite both exist for that.

Does this work on animation? Yes, and it is usually easier than on filmed footage.

Why approve the object separately? So the compositing is not wasted on the wrong shape.