Vizard Agent

How to make a speed ramp peak on the action instead of the dead space

Last updated 2026-09-10 · 8 min read

Tell Vizard Agent which moment the ramp is for. It samples the source across its timeline to find where the action genuinely starts and finishes, checks a finer set of frames around that window, then locks the ramp's slow point to the take-off and the landing — instead of applying a curve to the whole clip and hoping the peak lands somewhere useful.

What is the short version?

A speed ramp is only ever about one moment. Applied across a clip it peaks wherever the maths says, which is usually the run-up, and the actual jump goes past at full speed while the viewer is still catching up.

  1. Go to Vizard Agent with the clip.
  2. Say which moment the ramp exists for.
  3. Ask it to find that moment before setting any ramp.

What do you need before you start?

The clip and a sentence about what happens in it. Vizard Agent can find action on its own, but naming the beat — the jump, the catch, the reveal, the impact — is what separates a ramp that lands from a ramp that is technically present.

What do you type into Vizard Agent?

Describe the beat rather than the effect. "Ramp this clip" is an instruction about a curve; "the ramp is for the jump" is an instruction about a frame, and the frame is what the whole thing hangs on.

Prompt

Make a ten-second cut from this clip with a speed ramp. The ramp is for the jump and the landing — find where those actually are in the footage first, set the slow point on them, and do not let it peak on the run-up.

Variants worth knowing:

What does Vizard Agent actually do?

Here is the order on a real ten-second sports cut. Notice that four steps happen before any speed change is applied at all, and that all four of them are about finding the moment rather than shaping the curve.

  1. Checks what the source actually contains.
  2. Samples frames across its whole timeline.
  3. Looks at those frames to confirm the action is the right one.
  4. Refines the action window so the ramp will peak on the jump and the landing.
  5. Samples that window more finely before committing.
  6. Locks the ramp points against those frames.
  7. Builds the ramp around the locked points.
  8. Adds an impact sound on the peak, if you want one.
  9. Renders and pulls a contact sheet of the result.
  10. Checks the pivot visually rather than trusting the curve.
  11. Has the finished cut watched back for how the ramp reads.

Step four is the one that makes the difference, and it is the step a curve-first approach skips entirely. Deciding where the action window sits is a decision about the footage; everything after it is arithmetic that Vizard Agent can do once the window is right.

Step five exists because a coarse sample is enough to find the jump and not enough to place a ramp on it. Vizard Agent goes back at a finer interval around the moment, because being half a second out is the whole failure.

Step ten is why this does not come back wrong. A ramp is judged by watching it, not by reading its parameters, so Vizard Agent looks at the rendered pivot before calling it done.

What does the result look like?

A cut where the slowest frames are exactly the ones you would have paused on anyway. The build-up runs quickly, the moment itself holds long enough to register, and the speed picks up again on the way out of it.

Vizard Agent also tells you which timestamps it locked the ramp to, so a note like "half a beat later" becomes a precise adjustment rather than another round of nudging.

When does this not work well?

A ramp needs frames to stretch, and Vizard Agent cannot invent them. Footage shot at an ordinary frame rate has a hard limit to how far it can be slowed before the motion turns into a slideshow, and no amount of correct placement fixes a source that simply does not have the frames.

How do you fix a result that came back wrong?

Talk to Vizard Agent about the moment rather than about the curve. Because the ramp was locked to specific frames in the first place, a note like "it peaks too early" becomes a shift of those frames rather than a redesign of the whole effect from scratch.

How does Vizard Agent compare to doing it yourself?

By hand a ramp usually gets applied to the clip and then nudged by eye until it feels right, which works when the clip is short and fails as soon as there is a run-up. Vizard Agent inverts the order: find the frames first, then build the curve.

By hand Vizard Agent
Starting point A curve on the clip The frames of the action
Finding the moment Scrubbing Sampled coarse, then fine
Placing the peak Nudged until it feels right Locked to take-off and landing
Verification Watch it back Contact sheet plus a watch-back
Adjusting later Re-nudge everything Move the window

Common questions

How slow should the peak be? As slow as the frame rate allows. Vizard Agent will say what that is.

Can it ramp into a reverse? Yes. Ask Vizard Agent to pivot at the peak.

Does the audio slow down too? By default. Tell Vizard Agent to keep the sound at speed if you prefer.

Can it add an impact sound? Yes, timed to the locked peak rather than to the middle.

What if there are two moments? Vizard Agent will ask which one, or cut them apart.

Does it need high frame rate footage? It helps a lot. Vizard Agent will warn you when it does not have it.

Will it interpolate frames? Only if you ask. It has trade-offs worth seeing first.

Can it match a reference edit? Yes. Give Vizard Agent the reference and it copies the shape.

How does it find the action? By sampling frames and looking at them, then sampling finer.

Can I set the points myself? Yes. Give Vizard Agent timestamps and it uses them.

Will the ramp fit a ten-second cut? Vizard Agent sizes it against the length you asked for.

Does it work on generated footage? Yes, and the frame rate limits are the same.

Can it ramp a whole sequence? It can, but one ramp per idea reads better.

Why did mine peak early? Because the curve was applied to the clip rather than to the moment.