How to find the moments worth using in a two-hour film

Last updated 2026-09-17 · 8 min read

What is the short version?

A feature film has no transcript worth searching and nothing in it that tells you where the good parts are. Watching the whole thing costs you two hours, and at the end of it you still will not remember which minute the cavalry charge was in.

1. Ask Vizard Agent for a timestamped map of the whole film.
2. Read the map and point at the windows worth having.
3. Ask for dense frames in only those windows.

What do you need before you start?

The film and a rough idea of what you are hunting for. Vizard Agent can map anything, but the map is only useful if you know what you are scanning it for — a mood, an action, a face, a landscape.

• The film. Whole, not a cutdown.
• What you are looking for. Even loosely.
• The finished piece. A trailer, a montage, a teaser.
• Any house style. A previous piece to match.
• Anything off limits. Endings, spoilers, faces.

What do you type into Vizard Agent?

Ask for the map as a deliverable in its own right. If you ask for a trailer straight away you get a trailer built from whatever it found first, and no way to tell what it passed over.

Prompt

Variants worth knowing:

• "Before cutting anything." The map comes first.
• "Timestamped contact sheets." So you can point at a minute.
• "Then sample those densely." The second pass, stated up front.

What does Vizard Agent actually do?

Here is the Vizard Agent sequence on a real feature film cut down into a sub-sixty-second theatrical trailer. The whole method comes down to two passes at two different resolutions: one cheap and complete, then one expensive and deliberately narrow.

1. Probes the film and extracts its opening frames.
2. Builds contact sheets of the opening and reads them.
3. Pins the title-card timing frame by frame.
4. Samples a previous piece to learn the house style.
5. Samples the film every forty-five seconds for a visual map.
6. Builds timestamped contact sheets of the whole film.
7. Reads the map in thirty-minute passes.
8. Extracts dense frames from the windows the map flagged.
9. Builds contact sheets for each candidate window.
10. Reviews each candidate by name — cannon-hauling, river crossing, night battle.

Step five is the whole trick. Forty-five seconds is coarse enough that a two-hour film becomes a hundred and sixty frames, and fine enough that no sequence lasting more than a minute can hide from it.

Step seven matters more than it looks. A hundred and sixty frames in one image is unreadable; split into three passes of half an hour each, the same frames are something you can actually scan.

Step ten is what makes the rest of the job possible. Once each candidate has a name — "the oxen and cannon", "the firelight romance" — you and Vizard Agent can discuss the edit without going back to timecodes every time.

What does the result look like?

A shortlist with pictures attached. Six or eight named windows, each with frames you have already looked at, and the rest of the film accounted for rather than merely unwatched — so when Vizard Agent starts cutting, you know what it is choosing between.

When does this not work well?

A coarse map misses short things. A three-second reaction, a single line of dialogue, a brief insert — none of those are reliably caught at one frame every forty-five seconds, and if that is what you need the interval has to come down.

• Very short moments. They fall between samples.
• Dialogue-led films. A transcript serves better than frames.
• Visually uniform films. Every sample looks the same.
• Long films at a fine interval. The map stops being cheap.
• Spoiler-sensitive work. The map shows you the ending.

How do you fix a result that came back wrong?

Point at the map itself. Because every frame on it carries its own timestamp, a note like "there is something at 1:12:30 I want to see properly" is already a complete instruction, and Vizard Agent samples that one window densely rather than scanning the whole film again.

• "Sample around 1:12:30." That window gets dense frames.
• "Nothing here is what I meant." The interval drops or the brief is restated.
• "Too much to read." The sheets are split into shorter passes.
• "You missed a sequence." The interval was too coarse for it.

How does Vizard Agent compare to doing it yourself?

By hand you scrub the timeline looking for something that catches your eye, which works and takes an afternoon, and you will be confident you saw everything when you did not. The alternative is watching it, which takes two hours and leaves you with impressions rather than timecodes.

Common questions

What interval should it use? Forty-five seconds suits a feature. Vizard Agent drops it for shorter material.

Can I change the interval afterwards? Yes. Vizard Agent normally drops it only for the windows you care about.

Does it need to watch the film? Not at first. Vizard Agent reads frames, which is far cheaper.

What about dialogue? Ask for a transcript alongside. Frames and words answer different questions.

Can it match a previous piece? Yes. Give Vizard Agent an earlier trailer and it reads the house style from it.

How many windows should I pick? Six or eight is workable, and Vizard Agent samples each one of them densely.

Will it spoil the ending for me? The map covers the whole film. Say if you want the last reel excluded.

Can it find a specific actor? Yes, if you describe them. Vizard Agent flags the windows they appear in.

What if the film is longer than two hours? Vizard Agent uses the same method with a coarser interval and more passes.

Does it work for a lecture recording? Partly. For talking-head material a transcript is the better index.

Can it pull the clips as well? Yes, once you have chosen the windows.

How do I refer to a moment later? By its name. Vizard Agent labels each candidate window.

What if two windows look identical? Vizard Agent samples both densely so you can tell them apart.

Is the map worth keeping? Yes. It is the cheapest index of the film you will ever have, and Vizard Agent can reuse it next time.