How to add something to your footage that passes behind the person

Last updated 2026-09-17 · 8 min read

What is the short version?

You want two sharks in the pool behind the swimmer. Generating them is the easy half of the job, and Vizard Agent does it in one step; the half that decides whether anyone believes the shot is which of the two layers ends up in front of the other.

1. Tell Vizard Agent what to add and roughly when.
2. Say what it must pass behind — the swimmer, a rock, a rail.
3. Ask to see one frame at the moment they overlap.

What do you need before you start?

The clip, and a sense of where in it the addition belongs. Vizard Agent will find the stretch with the most open space around the subject, but if you have a moment in mind — just as they turn, just before they surface — that is worth saying.

• The clip. The original file, with its audio.
• What to add. One creature, two, an object.
• Where in the clip. A range of seconds, or "wherever fits".
• What it goes behind. The person, the foreground, both.
• The look. Realistic, stylised, comic.

What do you type into Vizard Agent?

Say the depth relationship out loud, in the same sentence. "Add two sharks" gets you two sharks from Vizard Agent, correctly lit and convincingly animated, gliding across the front of the swimmer — which is not what anybody means by adding sharks to a pool.

Prompt

Variants worth knowing:

• "Behind him, not in front." The instruction that creates the mask.
• "Where there is space around the swimmer." Lets it pick the stretch.
• "Keep the framing." Stops the clip being re-rendered around the addition.

What does Vizard Agent actually do?

Here is the Vizard Agent sequence on a swimming clip with two generated sharks added to it. The first version is finished, reviewed and rejected on one point — depth — and the second half of the list is the repair.

1. Extracts representative frames and checks the available editing routes.
2. Generates two realistic sharks with lighting that matches underwater footage.
3. Extracts each one as a silhouette with transparency, and verifies the alpha is clean.
4. Finds the stretch with the most space around the swimmer.
5. Prepares separately scalable versions of the two animals.
6. Diagnoses and corrects the cutouts where the larger silhouette was clipped.
7. Renders the section with the sharks composited in.
8. Compares the rendered frame with the original at the exact overlay point.
9. Reassembles the corrected section into the clip, keeping the original audio.
10. Switches to a direct render when concatenating the whole video times out.
11. Builds an animated mask of the swimmer to carry the depth.
12. Tests the swimmer layer composited over the sharks and checks the alpha format.
13. Renders with the swimmer in front, then applies the mask to the full clip.
14. Confirms the depth on a mid-section frame and re-checks the audio sync.

Step eleven is the difference between a video and a sticker. The swimmer is masked frame by frame as his own layer, the sharks are rendered beneath him, and the moment where his arm crosses one of them is what sells the whole shot.

Step eight is the habit worth copying. Vizard Agent does not ask whether the sharks look good; it puts the rendered frame next to the original frame at the same timestamp and checks what actually changed.

Step ten is a practical failure. Concatenating the edited section back into the full video ran over the time limit, so Vizard Agent switched to rendering directly rather than retrying the same route.

What does the result look like?

The same clip, the same framing, the same audio, with two animals moving through the water behind the swimmer. They enter and leave the shot rather than appearing, they are lit like the rest of the water, and the swimmer passes in front of them where they overlap.

When does this not work well?

Some subjects cannot be masked cleanly enough to carry depth. Splashing water, hair, spray and motion blur all produce edges that no mask follows exactly, and a mask that flickers at the edge is more distracting than a composite with no depth at all.

• Heavy splashing. The subject's outline changes every frame.
• Hair and spray. Fine edges that a mask cannot hold.
• Low contrast. The subject and the background are the same tone.
• Fast camera movement. The addition has to be tracked as well as masked.
• Very short overlaps. Not enough frames to be worth the mask.

How do you fix a result that came back wrong?

Say what is wrong with the placement rather than with the creature itself. Vizard Agent can change the layer order, the scale or the path of travel without regenerating the animal, and regenerating is both the slow part and the part that gives you a different-looking animal.

• "It is in front of him." The subject is masked as a foreground layer.
• "They look too big." Each one is rescaled independently.
• "They appear out of nowhere." The entrance is moved off the edge of frame.
• "The edges glow." The cutout's transparency is rebuilt and re-checked.

How does Vizard Agent compare to doing it yourself?

By hand the composite takes ten minutes and the mask takes the rest of the day. Most people do the first part, look at the result, decide that the depth does not matter that much, and publish something that reads instantly as an overlay.

Common questions

Can it add more than one thing? Yes. Vizard Agent prepares each one separately so they can be scaled and placed independently.

Does it have to be realistic? No. Say cartoon, and Vizard Agent generates and composites in that style.

Will it match the lighting? It is generated to match the scene, and you can ask Vizard Agent to grade it further.

Can it pass behind more than one thing? Yes, with a mask for each. Each layer adds work, so name what matters.

What about reflections in the water? Ask for them. They are a separate pass, not part of the composite.

Does the clip get re-rendered? Only the section that changes. Vizard Agent leaves the rest of the clip as it was.

What if the camera moves? Then the addition is tracked as well as masked, which Vizard Agent will do.

Can I choose where it enters the frame? Yes. Say which side, and it enters from there.

Will the audio change? No, unless you ask for a sound to go with it.

How do I check the depth quickly? Ask Vizard Agent for the one frame where the subject and the addition overlap.

Can it go behind a foreground object instead? Yes. The same mask approach works for a rail, a rock or a doorway.

What if the mask flickers? Tell Vizard Agent where, and the mask is rebuilt over that range.

Can it be added for the whole clip? Yes, though a brief appearance is usually more convincing.

Is a still image enough to work from? For the addition, yes — Vizard Agent can animate it into the footage.