How to pick the usable clips out of a pile of amateur footage

Last updated 2026-08-25 · 7 min read

What is the short version?

When twenty-odd people send you clips from the same day, a good half of them are unusable — shaky, soft, or the same moment sent twice. Deciding which ones survive is the real work, and it should be done on measurement rather than on a quick look.

1. Go to Vizard Agent and upload every clip you were sent.
2. Say to keep only the steady, sharp ones and discard the rest.
3. Say what order they should run in.

What do you need before you start?

All the clips and a rule for the order. Vizard Agent measures the quality itself, so pre-screening is a waste of your evening, but it cannot guess whether the clips should run in filename order, in time order, or in the order the event actually happened.

• Every clip. Including the ones you suspect are bad.
• The ordering rule. Numbered filenames, or chronological.
• The music. A track, or a description to generate one.
• The length. Or say "as long as the song".
• The occasion. It sets the tone of the cut.

What do you type into Vizard Agent?

Say the quality bar and the ordering rule in the same sentence. Those two instructions do most of the work, because they turn a vague pile of clips into a filtered, sequenced set that Vizard Agent can then cut to music without further guidance from you.

Prompt

Variants worth knowing:

• Length matched to the track. The music decides the runtime.
• Chronological order. Rebuilt from the file timestamps.
• A generated theme. When no track fits the occasion.

What does Vizard Agent actually do?

Here is the order Vizard Agent worked in on a real 23-clip event video shot by several people on phones. Two steps do the filtering that everything else depends on: an md5 hash pass to catch duplicates, and a measured sharpness and stability read on every clip.

1. Loads the event recap playbook and reads the metadata of every uploaded clip.
2. Downloads all 23 clips and analyses duration, resolution and timestamps.
3. Extracts sample frames from every clip and builds contact sheets in batches.
4. Reviews all three contact sheets by eye — clips one to eight, nine to sixteen, seventeen to twenty-three.
5. Hashes the files to catch duplicates, then checks the input order and durations.
6. Searches the music library, checks each clip's own audio, and generates a themed track.
7. Analyses the generated track — tempo, beat structure, dynamics and the final climax.
8. Measures sharpness and camera stability on every clip to decide which ones survive.
9. Assembles the surviving clips in the requested order, cut to the track's beats with smooth transitions, then reviews the result.

Step 8 is what makes the selection defensible. "That one looks shaky" is an impression that changes depending on how tired you are; a measured stability figure per clip gives the same answer every time and can be applied consistently across twenty-three files.

What does the result look like?

A short event film in which every clip is one somebody would be happy to see, running in the order you asked for, cut to a generated theme of roughly seventy seconds so the video ends when the music does rather than fading out mid-phrase.

The clips that failed the measurement are simply absent. Nothing is stabilised into usability here — the instruction was to discard rather than rescue, which is usually the right call when there are twenty-three clips and only ten good ones.

When does this not work well?

Triage only helps when there is something worth keeping in the pile to begin with. These are the cases where Vizard Agent's measurement pass correctly identifies a problem and then hands it straight back to you, because the fix is editorial rather than technical.

• If most clips fail, the video is short. Measurement cannot invent coverage.
• Discarding is not rescuing. Ask for stabilisation instead if you need those shots.
• Numbered order can be wrong. Filenames from several phones do not interleave.
• Everyone filmed appears. Community events include people who did not agree.
• Sharpness is not composition. A steady, sharp, badly framed shot still passes.

How do you fix a result that came back wrong?

Name the clip. Vizard Agent keeps the contact sheets, the duplicate hashes, the per-clip measurements and the beat map, so a rejected clip can be forced back in or an accepted one dropped without recutting the whole film.

• "Clip 14 should be in." Reinstated despite the measurement.
• "Two shots are the same moment." The duplicate hash pass already flagged them.
• "Wrong order." Rebuilt chronologically from the timestamps instead.

How does Vizard Agent compare to doing it yourself?

By hand this means opening twenty-three clips, deciding about each one, noticing halfway through that two are the same file under different names, and then discovering the music is longer than the footage. It is an evening, and the judgement drifts as it goes.

Common questions

How many clips can Vizard Agent screen? Twenty-three here. The measurement pass scales to far more than that.

How does Vizard Agent decide a clip is shaky? It measures camera stability and image sharpness per clip rather than judging by eye.

Will it stabilise the bad ones instead? Only if you ask. Vizard Agent discards by default when you say discard.

Can it catch duplicate files? Yes. Vizard Agent hashes every file, which finds the same clip sent twice under two names.

Can I keep a clip it rejected? Yes. Name it and Vizard Agent reinstates it in the cut.

Why measure instead of just looking? Because judgement drifts across twenty-three files and a borderline clip gets a different verdict at the start of the evening than at the end. A measured threshold applies the same standard to the first clip and the last.

Can Vizard Agent generate the music? Yes. It generated a themed track here, then analysed its tempo and structure before cutting.

Can the video match the song's length? Yes. Say so and Vizard Agent builds the cut to end with the track.

Does the numbered order always work? Not when several phones are involved. Ask for chronological order from the timestamps instead.

What if the clips are different resolutions? That is normal when several phones are involved. Vizard Agent reads each clip's resolution during the analysis pass and normalises them into one frame rather than letting the video change shape mid-cut.

Does Vizard Agent check the finished film? Yes. It reviews the assembly and the music transitions before delivery.