How to make a teaser from a long podcast episode
Upload the episode and ask Vizard Agent for a teaser. Vizard Agent transcribes the whole thing, measures where each host sits in frame, cuts to the strongest lines, and then checks whose mouth is actually moving at every cut so the picture matches the voice.
What is the short version?
A teaser has to make someone want the episode without giving them the answer. That means picking the question rather than the conclusion, and in a multi-host show it means being careful about which face is on screen while each line is spoken.
- Go to Vizard Agent and upload the episode.
- Ask for a teaser and say the length.
- Give it your logo and brand colours for the graphics.
What do you need before you start?
The episode and your branding. Vizard Agent reads the whole transcript and chooses the lines itself, so nothing needs marking beforehand. Supplying the logo early is worth doing, because the end card and the captions both take their colours from it.
- The episode. Any length; Vizard Agent maps the whole thing.
- The length. Thirty to forty-five seconds for a teaser.
- Your logo and colours. For the end card and caption styling.
- The hook, if you know it. A specific question or claim from the episode.
- The shape. Vertical for feeds.
What do you type into Vizard Agent?
Ask for a teaser explicitly rather than a highlight. Vizard Agent treats the word as a structural instruction, selecting for lines that open a question instead of lines that answer one, which is a different search through the same transcript.
Prompt
Variants worth knowing:
- Open on a specific question. If you know which moment will pull people in.
- Brand-coloured captions and end card. Vizard Agent samples the exact values from your logo.
- A close-up ending. It lands better than fading out on a wide two-shot.
What does Vizard Agent actually do?
Here is the order Vizard Agent worked in on a real podcast teaser. The section worth reading closely is near the end, where it went back and verified, cut by cut, that the person on screen was the person speaking.
- Probes the episode, samples frames and transcribes the whole thing.
- Reads the full transcript in several passes and maps the shots against it.
- Measures where each host sits in frame using a grid overlay on the wide shot.
- Pins the exact word boundaries for every chosen cut point.
- Tests the vertical crop window on each chosen moment and checks the framing.
- Zooms in on a brand detail in the room and measures the close-up framing.
- Generates the music, the headline overlay and the end card, then checks the headline at full size.
- Samples the exact colours out of the logo and rebuilds the end card, headline and subtitles in the brand palette.
- Watches back and checks whose lips are moving in the opening shot, compares the mouths at every cut, then rebuilds the edit with the wrong-speaker shot corrected and a new close-up ending.
Step 9 is the check almost nobody runs. In a two-host episode the cameras cut on their own schedule, and a clip lifted from the middle of a conversation will often show the listener rather than the speaker, which reads as an error to anyone watching.
What does the result look like?
From the run this page is written from, probed on the delivered file: 1080x1920, H.264, 30fps, 43.43 seconds, AAC audio. Vertical, forty-three seconds, six segments reframed onto whoever is speaking, with a headline overlay, brand-coloured subtitles and an end card built from the real logo.
Forty-three seconds is at the top of the teaser range and earned by the material. Vizard Agent cut it to where the question is fully posed, because trimming a hook to hit thirty seconds usually means cutting the half of it that makes people curious.
When does this not work well?
A teaser depends entirely on the episode containing at least one moment that works when lifted out of context. Vizard Agent reads the whole transcript and finds the best candidates in it, and some conversations simply do not have a self-contained hook anywhere in them.
- Not every episode has a hook. A good conversation and a good teaser are different things.
- Wide two-shots crop badly. Two hosts in a widescreen frame will not both fit vertically.
- Cutting mid-argument misleads. A line pulled from context can say something the guest did not mean.
- Studio graphics come along. Lower thirds from the original recording end up in the crop.
- Guests should see it. A teaser attributed to someone is worth showing them first.
How do you fix a result that came back wrong?
Name the moment. Vizard Agent keeps the full transcript, the word boundaries, the seat measurements, the crop tests and the sampled brand colours, so re-cutting a segment or changing the opening is a re-render against work it already did on the whole episode.
- "Start with the myth question instead." Re-cut from that point, as this run did.
- "That shot shows the wrong host." Re-checked mouth by mouth and replaced.
- "The captions are the wrong red." Re-sampled from the logo and re-rendered.
How does Vizard Agent compare to doing it yourself?
By hand this means scrubbing an hour for the line you remember, cropping it vertically and moving on. The wrong-speaker problem is the one that gets missed, because on a small timeline everyone looks like they might be talking, and it only becomes obvious once the teaser is posted.
| By hand | Vizard Agent | |
|---|---|---|
| Finding the hook | Scrub for what you recall | Read from the whole transcript |
| The vertical crop | Centre on the wide shot | Each host's seat measured, crops tested |
| Who is speaking | Assume | Mouths compared at every cut |
| Brand colours | Eyeball from a screenshot | Sampled from the logo file |
Common questions
Does this work for any long-form video? Yes. A webinar, a panel or an interview all teaser the same way: Vizard Agent reads the transcript for the moment that opens a question rather than the moment that answers one.
How long should a podcast teaser be? Thirty to forty-five seconds. Long enough to pose the question, short enough to withhold the answer.
How is a teaser different from a clip? A clip delivers a point. A teaser opens one and stops, which means Vizard Agent searches the transcript differently.
Will it use our branding? Yes. Vizard Agent samples the exact colours from your logo for the captions and the end card.
Can it handle a multi-host show? Yes, and that is where the lip check matters. Vizard Agent measures each host's position and verifies who is speaking.
Why check the mouths at all? Because a multi-camera podcast cuts on its own rhythm, not on who is talking. Lift a segment out and you will frequently be showing the person listening, which every viewer notices even if they cannot say why.
Can I choose the opening line? Yes. Name it and Vizard Agent rebuilds the cut from there.
Does it add subtitles? Yes, in your brand colours, built from word-level timings.
Does Vizard Agent check the finished teaser? Yes. It reviews the frames, measures the mix and had this cut watched back three times before it was right.