How to pick clips for people who already follow you
Tell Vizard Agent that the clips are for a channel of people who already chose to be there. That rules out the usual selection standard — the most striking, most emotional, most clippable moment — and replaces it with something harder: the passages where the speaker is most herself and the thought stands on its own.
What is the short version?
Clips for discovery are chosen to stop a stranger scrolling past. Clips for your own subscribers are chosen to be worth their time instead, and in practice the two standards rarely point Vizard Agent at the same minute of a recording.
- Tell Vizard Agent the clips are for your own channel.
- Say explicitly that the most viral moment is not the target.
- Say what a viewer should come away with instead.
What do you need before you start?
The recording and a description of the speaker at their best. Vizard Agent can rank passages for presence, but it needs to know what presence means here — warmth, authority, playfulness — because that is the quality being selected for.
- The full recording. Not a pre-cut highlight.
- Who is watching. Followers, members, a mailing list.
- What "at their best" means. In your words.
- How many clips. Six is a common ask.
- What must not be lost. Context that makes a thought make sense.
What do you type into Vizard Agent?
Say what you are not looking for as well as what you are. Naming the anti-goal is the fastest way to move Vizard Agent's selection away from the obvious, because "most engaging" is the default standard in every clip tool and every editor's instinct.
Prompt
Variants worth knowing:
- "Not the most viral or emotional." The anti-goal.
- "Experience her at her best." The actual criterion.
- "Standalone." Each clip works without the rest.
What does Vizard Agent actually do?
Here is the Vizard Agent sequence on a masterclass of two hours and twenty minutes, cut into six clips for a subscriber channel. The selection is most of the work here, and it is carried out against the whole recording rather than a sample of it.
- Watches the whole masterclass, paying attention to the speaker's presence.
- Captures the wording, the structure and the framing together so selection uses all of it.
- Checks the actual camera image and whether gesture and face survive a mobile crop.
- Splits the recording, since two hours twenty is too large to analyse in one pass.
- Builds smaller analysis sections with the audio and content unchanged.
- Prepares a gapless review in ten-minute sections so nothing is skipped.
- Compares the strongest standalone thoughts across the whole recording.
- Ranks the candidates by presence and standalone value.
- Reads each statement in context and tests possible openings against the full passage.
- Checks the substantial passages for a complete idea rather than a striking line.
- Looks for examples that show character without needing context that is not in the clip.
- Includes the later sections and the closing questions, so no stronger moment is missed.
Step six is what makes this selection honest. A clip tool samples; a gapless review in sections means the sixth-best moment in the last twenty minutes is considered against the third-best in the first ten.
Step nine is the test that distinguishes a clip from a quote. A line can be excellent and unusable, because the sentence that made it land was thirty seconds earlier — so each candidate is read against the passage it came from.
Step three matters for a talk recorded on a webcam. If the gesture and the face do not survive the vertical crop, a passage that reads beautifully in the transcript is not a clip.
What does the result look like?
Six clips that each carry one complete thought, in which the speaker sounds like herself rather than like a highlight, that make sense to someone who was not there, and that nobody would describe as the most dramatic moments of the recording.
When does this not work well?
Some recordings are not built for it. If the material is structured as a build-up to one revelation, if the speaker is reading a script, or if every passage depends on a slide you cannot show, standalone clips will misrepresent it.
- A single build-up. Nothing stands alone.
- Scripted delivery. Presence is what you are selecting for.
- Slide-dependent passages. The words need the picture.
- Very short recordings. Six clips from twenty minutes is thin.
- Discovery goals. If you want reach, the viral standard is the right one.
How do you fix a result that came back wrong?
Say what was wrong with the choice rather than with the cut itself. Vizard Agent can re-rank the candidates against a different quality — more warmth, less teaching, more personal story — and it is that ranking which produced the six clips in the first place.
- "These are the obvious moments." The viral test is excluded and the ranking redone.
- "This one needs context." It is extended backwards or replaced.
- "She sounds like she is performing." The ranking favours the quieter passages.
- "Two clips say the same thing." One is swapped for a different theme.
How does Vizard Agent compare to doing it yourself?
By hand you scrub for the moments you remember, and what you remember is the drama. That is the same instinct a clip tool has, which is why both produce the same six clips — and why a subscriber channel full of them starts to feel like an advert.
| By hand | Vizard Agent | |
|---|---|---|
| Coverage | The parts you remember | Every section, gaplessly |
| Standard | Most striking | Presence and standalone value |
| Context | Assumed | Each candidate read in its passage |
| Crop | Applied after | Checked before selection |
| Result | The obvious six | Six that reward being there |
Common questions
Is this different from viral clipping? Completely — tell Vizard Agent which you want, because the selections barely overlap.
How many clips from a long talk? Six from two hours is realistic; Vizard Agent will say what it found.
Can it do both sets? Yes. Ask Vizard Agent for a discovery set and a subscriber set separately.
What if the recording is very long? Vizard Agent splits it into sections so nothing gets sampled past.
Will the clips have captions? Yes if you want them; on a subscriber channel they are optional.
Can it keep the speaker's slower moments? Yes, and Vizard Agent often finds those are the ones that belong here.
What about vertical crops? Vizard Agent checks the gesture survives the crop before selecting a passage.
Can it rank rather than choose? Yes. Ask Vizard Agent for the ranked list and pick the six yourself.
Will it tell me why it chose each one? Yes. Vizard Agent gives a sentence of reasoning per clip.
What if my channel is on Telegram or a members' area? It makes no difference to the selection; only the export format changes.
Can it avoid repeating a theme? Yes. Vizard Agent deliberately spreads the six clips across different ideas.
Should each clip end on a call to action? For subscribers, usually not — it is already their channel.
How long should they be? As long as the thought takes, which is usually longer than a discovery clip.
Can it reuse the same clips later? Yes, and a subscriber audience minds repetition far less than a feed does.