How to direct a camera move as something a person did rather than an effect
Describe the move to Vizard Agent as a human action with a reason — "someone grabs the phone up quickly because they realise they have made a mistake" — and say where it ends up pointing. It then finds a believable end frame in the wider shot, builds the anticipation before the move, and matches the shake to that action rather than applying a preset.
What is the short version?
Camera moves described in parameters come out as parameters. Speed, direction and shake amount produce a technically correct movement that reads as an edit, because real camera movement has a cause and a bit of hesitation before it.
- Tell Vizard Agent who moved the camera and why.
- Say what the last frame should be pointing at.
- Ask for the lead-in as well as the move.
What do you need before you start?
The clip and a picture in your head of the person holding the camera. Vizard Agent works from that description, so the more concrete the action — snatched up, put down guiltily, lowered slowly — the less the result looks like a transition preset.
- The clip. With the frames before the move.
- The action. What the person physically did.
- The reason. It changes the speed and the hesitation.
- The end frame. Ceiling, floor, table, away.
- Whether the audio continues. Usually it should.
What do you type into Vizard Agent?
Write a sentence about a person. If it reads like a stage direction rather than a settings list, it is right — and if you find yourself naming a number of degrees, you have slipped back into describing an effect.
Prompt
Variants worth knowing:
- "As if they realised." The reason shapes the speed.
- "The final frame should be the ceiling." Gives it a destination.
- "Everything before that untouched." Ring-fences the clip.
What does Vizard Agent actually do?
Here is the sequence on a real clip that needed a snatched-away ending. Notice steps four and five: before building the move, Vizard Agent goes looking for something in the original that the camera could believably end up pointing at.
- Reframes the shot to what should stay visible, timing and audio unchanged.
- Tightens the framing further until only a glimpse of the detail remains.
- Adds a brief handheld motion to the final moment only.
- Reviews the movement and its lead-in as separate things.
- Rebuilds it as a realistic upward grab toward the ceiling.
- Inspects the wider original framing for a believable ceiling area.
- Views that wider frame before building the new ending.
- Prepares the final seconds as a focused retake rather than a whole re-render.
- Creates the mistake-take swing with its anticipation and its overshoot.
- Verifies the motion is smooth and the audio stays aligned.
Step six is what stops the ending looking invented. A camera swinging up has to arrive somewhere, and if the ceiling it arrives at was never in the original footage, the last frame gives the whole thing away.
Step four separates the two things people conflate. The move is the easy part; the lead-in — the quarter second where the hand tightens before it lifts — is what makes it read as a person and not a keyframe.
Step eight keeps the cost sensible. Only the final seconds are rebuilt, so the approved part of the clip is not re-rendered and cannot drift.
What does the result look like?
An ending that reads as an event rather than as a transition. The camera hesitates for a moment, swings, overshoots slightly and settles on something that was really there in the room, and Vizard Agent confirms the audio stayed in step with the picture throughout the rebuilt section.
When does this not work well?
A move has to have somewhere to go. If the original framing is tight, there is no wider view to swing into, and the camera arrives at something invented — which is exactly the frame the viewer will look at hardest.
- Very tight original framing. Nowhere believable to land.
- Tripod-locked footage. No handheld character to build on.
- Moves that need new geometry. A pan into an unseen wall.
- Long moves. The further it travels, the more is invented.
- Audio that implies stillness. The sound contradicts the motion.
How do you fix a result that came back wrong?
Describe the action again, differently. Vizard Agent is responding to the human description rather than to a set of numbers, so a note like "less panicked, more like they simply put it down" changes the speed, the path and the hesitation all together rather than one at a time.
- "Too smooth to be a person." More anticipation and overshoot.
- "It looks like a transition." The lead-in is rebuilt.
- "The ceiling looks wrong." A real area from the wider frame is used.
- "Too violent." The reason is re-stated and the speed follows.
How does Vizard Agent compare to doing it yourself?
By hand you keyframe a rotation and add a shake preset, and it looks like a rotation with a shake preset. Making it look human means hand-animating the anticipation and the overshoot, which is a specialised skill and an hour of work for five seconds.
| By hand | Vizard Agent | |
|---|---|---|
| The instruction | Degrees, duration, shake amount | A person and a reason |
| Anticipation | Hand-animated, if at all | Built as part of the action |
| The end frame | Whatever the move lands on | Chosen from the real footage |
| The rest of the clip | Re-rendered | Left alone |
| Changing it | Re-keyframing | Describing the action again |
Common questions
Why does describing the reason help? Because it sets the speed and the hesitation. Vizard Agent uses both.
Can it do a slow, deliberate move too? Yes. Say the person set it down carefully and Vizard Agent paces it that way.
What if the ceiling is not in my footage? Vizard Agent looks in the wider frame first and tells you if there is nothing usable.
Will the audio still line up? Yes. Vizard Agent keeps the original audio and checks it after the rebuild.
Does the whole clip get re-rendered? No. Vizard Agent rebuilds only the seconds that change.
Can it add a move at the start instead? Yes. A pick-up at the start works the same way as a put-down at the end.
How do I ask for shake? By naming the action. Vizard Agent derives the shake from what the person did.
What if it looks too smooth? Tell Vizard Agent it reads as an effect and ask for more anticipation.
Can it match a move in a reference clip? Yes. Send the reference and Vizard Agent reads the motion from it.
Does this work on generated footage? Yes, though the frame it swings into is generated too, so check it.
Can I keep a detail visible during the move? Yes. Say what must stay in shot and Vizard Agent frames for it.
What about a move that ends off-frame entirely? Fine, and often easier — there is nothing to land on believably.
Will it change the length? No, unless you ask. The move happens inside the existing seconds.
Can I get two versions to compare? Yes. Ask Vizard Agent for a gentler and a sharper take on the same action.