How to make an object change over the length of a clip

Last updated 2026-09-22 · 8 min read

What is the short version?

You want the cup on the table to be an apple, and you want that apple to be visibly rotten by the last frame. That is not a replacement job; it is a replacement plus a timeline, and Vizard Agent has to build both ends before it can animate anything.

1. Tell Vizard Agent what the object becomes at the start.
2. Tell it what that thing looks like by the end.
3. Say how the change should be paced across the clip.

What do you need before you start?

The clip and a description of both ends. Vizard Agent generates the two states itself, but the more specific you are about the end state — browning, shrivelled, collapsed inwards — the less it has to guess about how far the change is supposed to travel.

• The clip. Ideally with a still camera.
• The starting object. What replaces what.
• The end state. How far the change goes.
• The pace. Even, or mostly at the end.
• What stays. The table, the hands, the sound.

What do you type into Vizard Agent?

Describe the end state as a picture rather than as a process. "Rots" is a verb with an enormous range, whereas "brown, shrivelled and sunken by the final frame" tells Vizard Agent exactly where the animation has to arrive by the time the clip ends.

Prompt

Variants worth knowing:

• "Progressively." Across the clip, not in one cut.
• "Clearly brown and shrivelled by the end." The destination.
• "Keep the background exactly." Scope, stated.

What does Vizard Agent actually do?

Here is the Vizard Agent sequence on a clip where a coffee cup becomes an apple that rots as it plays. The two states are built as a pair rather than separately, and that single choice is what makes the animation between them believable.

1. Checks the clip's duration and dimensions and samples frames.
2. Looks at the cup, the background and the movement in those frames.
3. Extracts the first and last frames and inspects the cup area closely in each.
4. Opens a full frame to measure the cup's exact position.
5. Checks the display orientation and crops the cup region correctly.
6. Widens the crop so the whole cup and the surface under it are included.
7. Generates a fresh apple and a rotting apple in the same frame, in parallel.
8. Checks each version for fidelity to the original background.
9. Isolates both apples and prepares them as transparent layers.
10. Fixes the replacement mask so the cup's edges disappear without leaving a halo.
11. Builds an animation layer that moves gradually from fresh to rotten.
12. Composites it over the clip, keeping the original soundtrack.

Step seven is the decision the whole job rests on. Generating the two states separately gives you two apples with different lighting, different shadows and different positions; generating both in the same frame means the only difference between them is the thing that is supposed to change.

Step ten is the detail that makes it invisible. The old object has to be gone completely — a rim of the cup still visible behind the apple, or a bright halo where the mask ended, is the first thing a viewer sees.

Step six is the mistake worth avoiding. A crop tight to the object leaves the shadow and the surface behind, and a replacement that floats without its own shadow reads as a sticker regardless of how good it looks.

What does the result look like?

The same shot with a different object in it, changing as the clip runs: fresh at the start, unmistakably rotten at the end, sitting on the same surface with the same light and shadow, and the original audio untouched.

When does this not work well?

Some shots cannot hold a state change. If the object is handled, moves behind something, or the camera moves around it, the two-state approach has to become full tracking, and every frame between the states has to hold up on its own.

• Handled objects. Hands cross the replacement.
• Moving cameras. The layer needs tracking as well as animating.
• Very long clips. A slow change over minutes has to be built in stages.
• Reflective surfaces. The old object survives in the reflection.
• Shallow focus. The replacement's sharpness will not match.

How do you fix a result that came back wrong?

Say which end is wrong, or say that the problem is in the middle. Vizard Agent can regenerate one of the two states, change the pacing of the change between them, or rebuild the mask, and those are three genuinely different repairs.

• "It is not rotten enough at the end." The end state is regenerated further along.
• "It changes too early." The pacing is weighted towards the end.
• "There is a halo around it." The mask is corrected against the original edges.
• "It looks pasted on." The crop is widened to include the shadow and surface.

How does Vizard Agent compare to doing it yourself?

By hand you would either shoot a time-lapse of a real apple and match the lighting, or build two matte paintings and cross-dissolve. The dissolve is the usual answer, and it looks like a dissolve — both states visible at once in the middle, which is exactly when the viewer is looking hardest.

Common questions

Can it go the other way — rotten to fresh? Yes. Vizard Agent treats it as the same job with the states reversed.

What else can change this way? Anything with a clear before and after — Vizard Agent handles wilting, rusting or melting the same way.

Does the camera have to be still? It is much easier. A moving camera means tracking as well as animating.

Can the change be sudden? Yes. Name the moment and Vizard Agent times the whole change to land there.

Will the shadow change too? Vizard Agent changes it too when the object's shape changes much.

Can it keep the original audio? Yes. Vizard Agent composites the picture and leaves the sound alone.

How long can the clip be? Vizard Agent builds longer clips in stages so the middle states hold up.

Can two objects change at once? Yes, though each one is its own layer and its own mask.

What if someone picks the object up? Then it needs tracking, and Vizard Agent will say what that costs.

Can I choose the end state from a reference? Yes. Supply a picture and Vizard Agent aims at that.

Does it work on animation? Yes, and Vizard Agent usually finds it easier than live footage.

Will it look real? It looks as real as the two generated states do — which is why they are checked first.

Can I see the states before the animation? Yes, and that is the cheap moment to change your mind.

How fast should the change be? Slow enough not to notice a single frame, fast enough to be obvious by the end.