Podcast Testimonial Clips

Yes, if you want a 12-15 second clip that feels like a real moment from a podcast — two people actually talking about your product. Guest drops the insight, host reacts in real time (nods, laughs, quick acks, maybe a cut-in). Lands softer than a talk-show ad. Feels like someone sent it to you because it's interesting, not because it's an ad.

1. step 1 / 6

   Install the podcast-clip skill

   Click Generate install command below → copy → Terminal → enter. Skill lands in ~/.claude/skills/podcast-clip-prompting/.

2. step 2 / 6

   Write the conversation beat

   Seedance clips cap at 15 seconds. Write a tight CONVERSATION (not a monologue) — guest does most of the talking, host engaged throughout. Target 12-14 seconds of total dialogue.

   Structure:

   HOST: short setup question (1-2s) GUEST: the insight with the product drop (8-10s) HOST: quick reaction — 'no way', laugh, or a skeptical cut-in (1-2s)

   Example:

   HOST: "Wait so what actually changed?" GUEST: "Dude, I was posting 3 videos a week getting 200 views. Started using LYRC to template the whole thing and my last drop did 400k. I'm not saying it was the tool, but… yeah." HOST: (laughing) "Bro come on."

   Keep the guest line punchy — contractions, pauses, real-speech rhythm. Not a pitch.

3. step 3 / 6

   Pick the podcast vibe

   In Claude Code:

   Use podcast-clip-prompting to generate a 15s clip: [your conversation beat] — vibe should be like [JRE / Theo Von / Smartless / Call Her Daddy / H3 Podcast / generic two-person interview]
   

   The skill sets scene (mic setup, audio panels, lighting, guest vs host framing) based on the vibe you name.

4. step 4 / 6

   Lock the host reactions

   Host needs to be VISIBLY engaged throughout — nods on key phrases, laugh on the punchline, quick cut-in at the end. The skill writes host B-roll + verbal acks (yeah mhm no way). Review the shot list; tweak specific host reactions to match the emotional beat of each guest line. If host looks frozen or stiff, it reads as a monologue video, not a podcast.

5. step 5 / 6

   Send to Seedance

   Duration 15 (max). 720p. 9:16 for Reels/TikTok, 16:9 for YouTube. Generate. First try is almost never perfect — expect 2-3 regenerations dialing in the conversation rhythm.

6. step 6 / 6

   Ship without captions

   Controversial rule: ship podcast clips WITHOUT burned-in captions. Viewers scroll past captioned clips treating them as ads. Uncaptioned clips force audio-on engagement, which signals interest to the algo. If you need text on-screen, use a single title card at the start (2 seconds) — then let the dialogue breathe.

• Going for DEBATE format (the real engagement driver)? Flip the host from agreeing to pushing back: guest claims '[your thing] is the most underrated [category] right now,' host: 'come on man, that's not actually true,' guest counters with one specific. 10-12s total, max comment-bait.
• Selling to an age bracket? Match host age. Gen Z → Call Her Daddy / Barstool energy. Millennials → Theo Von / H3. Older → Smartless / Oprah vibe. Host demographic = implicit audience demographic.
• Want story-time energy? Guest's line gets ~10 seconds, host gives ONE short reaction at the end instead of mid-beat. Feels more like quiet listening, less like interview.
• Going multi-part? Generate 2-3 clips from the same conversation — different 15s slices (setup, reveal, punchline). Post as a series across consecutive days. Same two faces, different beats.

• Voices sound generic / like a stock AI host.

  You didn't give a voice reference. Pull 10+ seconds of clean continuous speech from whoever you want to sound like (podcast clip, interview, streamer VOD), feed as `voice_urls`. Under 10s = Seedance falls back to generic.

• Host reactions look stiff / don't match the guest's line.

  You skipped the reaction-locking step. Edit the prompt: for each guest beat, specify the host's expected reaction ('eyes widen on [key phrase]', 'small laugh on [punchline]', 'slight head shake on [claim]'). Without explicit cues, Seedance ships a resting AI-face and it reads as a monologue.

• Seedance rendered the podcast in a too-dark studio, looks like a commercial.

  Bright daytime + window light is the locked rule. The skill should set this but double-check the prompt has 'natural daylight, large window out of focus, NOT dim/tungsten.' Dark studios = infomercial vibes.

• Guest looks directly into the camera — kills the podcast illusion.

  Guest is talking to the HOST, not the viewer. Add explicit `seated at 3/4 angle, eyes tracking off-frame host, NEVER into lens.` Any direct-to-camera eye contact = it reads as a YouTube vlog, not a podcast.

• Captions appear even though I didn't ask for them.

  Seedance's default often burns them in. Add explicit negative: 'no subtitles, no burned-in text, no captions.' If the model ignores the negative, generate without captions and strip any text layer post-render.

• ★ required
  Content-at-scale for founders

  Ghostwrite + clip founders' content into a high-volume LinkedIn / X / podcast presence. Premium DFY play.

• ★ required
  Podcast clipping as a service

  Sell short-form clip output to podcasters who hate editing — $200-2K/mo retainers per show. Pipeline runs in your sleep.

• ★ required
  Skill-pack course around a proven pattern

  Take ONE pattern you've actually run. Package it as a $200-500 course. Sell the formula, not theory.

• AI animated series (the Basement Boys pattern)

  Build a recurring AI animated series with named cast and running gags. Episodes compound. The IP becomes the asset.

• Creator memberships / paid Telegram channels

  Lock your most useful daily output behind $20-50/mo. Telegram + Skool + Whop are the cheapest delivery rails.

• Faceless YouTube AI channel

  Niche faceless channel — AI script + AI voice + AI footage + scheduler. AdSense + niche affiliate stack on top.