How to break a reference video into its actual scenes
Ask Vizard Agent to cut the reference into its real scenes and give them to you as separate files. The boundaries that matter are the ones the video already uses: the fades it passes through, where its sentences end, the pause before the music drops. Equal intervals will not find any of them.
What do you type into Vizard Agent?
Ask for the pieces, not a summary. The instinct is to ask what the style is, but a list of scenes with real durations tells you more than any adjective. The prompt's last sentence tells Vizard Agent what the precision is for: a decomposition you intend to build from needs exact durations.
Analyse this in as much detail as possible. Break it into its actual scenes and give me each one as a separate file I can watch, with its duration. Cut on the boundaries the video itself uses, its fades to black and where the sentences end, not on fixed intervals. I want to rebuild in the same style and format.
What does Vizard Agent actually do?
It measures the reference from several directions before cutting, because structure is carried by different things in different videos: sometimes the edit, sometimes the speech, sometimes the music. Then it cuts twice. Phrase boundaries give the fine grain of how the thing is written, and the blackouts give its sections. In the session behind this article it:
- Built an edit map, a full transcript and a contact sheet of the reference, and checked the motion, transitions, sync and audio delivery.
- Examined the title animation frame by frame, and went through the first and last twenty-three seconds the same way.
- Looked at the waveform's shape and density, and checked the pause and drop around thirty seconds.
- Cut the video into twenty-two scenes, verified each duration, trimmed the audio tails and delivered them as separate files.
- Re-cut on exact spoken-phrase boundaries using word timings.
- Located the exact fade-to-black intervals and cut eleven larger scenes on those.
When does this not work well?
When the reference has no structural devices to find. A continuous single-take video with no fades and no clear sections does not decompose, and cutting on arbitrary boundaries would invent a structure the original never had. Copying the scene lengths too literally also misleads, because they suit the reference's own content. And its footage, music and graphics aren't yours to reuse.
Common questions
Why not just describe the style? Descriptions lose the durations, and Vizard Agent works from durations.
Why two passes? Phrases give Vizard Agent the fine grain; blackouts give the sections.
What if there are no fades? Then say which device to cut on. Vizard Agent will not invent one.