How to find the few bad seconds in an hour of footage where nothing happens
Ask Vizard Agent to measure camera motion second by second across every clip and list the spikes. In footage where nothing happens, a spike is almost always somebody touching the camera — so the handful of flagged moments are the only seconds anyone needs to look at, and the rest of the hour can be used as it is.
What is the short version?
Ambient footage is the opposite of a normal edit. Almost every second is usable, a few are ruined, and the ruined ones look identical to the good ones in a contact sheet because the difference is movement rather than content.
- Ask Vizard Agent to measure motion per second, not per clip.
- Ask it to list the spikes and show you those frames.
- Ask it to confirm the measurement is actually working.
What do you need before you start?
The clips and some patience with the first pass. Vizard Agent builds small proxies to analyse rather than measuring the full-resolution files, which is what makes an hour of 4K tractable — and the proxies are for analysis only, not for the finished video.
- All the clips. However many there are.
- What counts as bad. Handling, a hand, a person walking through.
- How long you need. It sets how much you can afford to lose.
- The delivery resolution. Proxies are for analysis only.
- Whether stabilising is acceptable. For borderline shots.
What do you type into Vizard Agent?
Ask Vizard Agent for per-second measurement explicitly. Clip-level judgement drops whole clips, which is exactly the wrong tool for this job — a five-minute clip with two bad seconds in the middle of it is still a five-minute clip you want to keep almost all of.
Prompt
Variants worth knowing:
- "Second by second rather than rating each clip." The key distinction.
- "Might be hands or camera handling." Names what a spike means here.
- "Show me those frames." You confirm, it flags.
What does Vizard Agent actually do?
Here is the sequence on a real hour-long ambient film assembled from fifty-one clips. The detour in the middle is the useful part: the metric was wrong at first, and a wrong metric flags nothing and looks like good news.
- Probes every clip for resolution, frame rate and length.
- Builds contact sheets of all of them for a first look.
- Makes small proxies for motion analysis.
- Measures camera shake and subject motion on the proxies.
- Builds dense thumbnail sheets to scan for hands and handling.
- Summarises motion per clip and per second.
- Debugs the motion metric when its output looks wrong.
- Fixes the metric and re-runs the whole analysis.
- Lists the motion spikes that may be hands or camera handling.
- Shows you the flagged moments as frames to confirm.
Step seven is worth expecting rather than being surprised by. A motion measure that returns nothing looks exactly like footage with no problems in it, and the only way to tell the difference is to check the metric against a moment you know is bad.
Step three is what makes this affordable. Measuring motion on full-resolution 4K takes far longer than the analysis is worth, and a small proxy carries the same movement information.
Step six is where the method pays off. Per-clip numbers tell you which clips are shakiest; per-second numbers tell you which four seconds to avoid, and the rest of that clip is untouched.
What does the result look like?
A short list instead of an hour of viewing. A handful of timecodes with frames attached, each one a moment you can confirm or dismiss in a second, and the rest of the footage usable without further inspection.
When does this not work well?
Motion is a proxy for badness, not badness itself. In footage where things genuinely happen — a bird taking off, a branch falling — the spikes are the content rather than the fault, and the list will be long and useless.
- Footage with real events. Everything spikes.
- Handheld throughout. No quiet baseline to spike against.
- Slow faults. A gradual drift does not spike.
- Faults that are not motion. Focus, exposure, a passer-by.
- Very short clips. Not enough baseline to measure against.
How do you fix a result that came back wrong?
Point at a bad moment it missed. That is by far the most useful thing you can hand Vizard Agent, because it lets the metric be checked against a case you already know is bad, rather than being adjusted up and down on a hunch until the list looks about right.
- "It missed the one at 4:12." The metric is checked against that moment.
- "Everything is flagged." The threshold is raised, or motion is the wrong test.
- "Nothing is flagged." The metric is verified before the result is believed.
- "These are birds, not handling." Subject motion is separated from camera motion.
How does Vizard Agent compare to doing it yourself?
By hand you scrub fifty-one clips at speed, which works for finding a person walking through and fails for a two-second wobble. So you either watch the whole hour or you ship the wobble, and most people ship the wobble.
| By hand | Vizard Agent | |
|---|---|---|
| The unit | The clip | The second |
| The method | Scrubbing | Motion measured on proxies |
| What you review | Everything | The flagged seconds |
| A silent failure | Looks like clean footage | The metric is verified |
| Salvage | Drop the clip | Avoid the seconds, keep the clip |
Common questions
Why per second rather than per clip? Because a long clip with two bad seconds is worth keeping. Vizard Agent flags the seconds.
What are proxies for? Speed. Vizard Agent analyses small copies and delivers from the originals.
Can it tell handling from wind? Usually, by the shape of the spike. Vizard Agent shows you the frames to confirm.
What if it flags everything? Then motion is the wrong test for that footage, and Vizard Agent will say so.
Can it find people walking through? Yes. Vizard Agent uses the thumbnail sheets for that rather than the motion measure.
Will it check its own measurement? Ask Vizard Agent to. A broken metric returns nothing and looks exactly like clean footage.
Can the bad seconds be stabilised instead? Sometimes. Vizard Agent tests how shaky they really are first.
Does it drop the whole clip? No. That is the point of measuring per second.
How long does the analysis take? Minutes, because Vizard Agent works on proxies rather than the full-resolution files.
What about focus problems? A different measure. Tell Vizard Agent and it checks sharpness too.
Can I give it a known bad moment? Yes, and it is the best thing you can give Vizard Agent to calibrate against.
Does this work for a single long clip? Yes. Vizard Agent does not care how the footage happens to be divided up.
What if I need exactly an hour? Vizard Agent plans around the excluded seconds to hit the duration.
Can I see the numbers? Yes. Ask Vizard Agent for the per-second motion summary.