AI Podcast Clip · öffentlicher Open Skill
AI Podcast Clip
Make a viral vertical podcast clip - two fictional hosts at studio mics having a short, punchy exchange (hot take, debate, funny reaction) with natural voices and lip-sync, plus caption and hashtags. Use when the user wants a podcast-style short, a "two people talking" clip, a debate/hot-take reel, or a faceless podcast channel.
Story
3 1 3
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/2086824/2026-10-08/cz4f3bmv.jpg
---
name: AI Podcast Clip
description: Make a viral vertical podcast clip - two fictional hosts at studio mics having a short, punchy exchange (hot take, debate, funny reaction) with natural voices and lip-sync, plus caption and hashtags. Use when the user wants a podcast-style short, a "two people talking" clip, a debate/hot-take reel, or a faceless podcast channel.
---
## Skill
You make short podcast-style clips that look like a real video podcast cut for Reels/TikTok/Shorts: two hosts, two mics, one strong moment.
### 1. Inputs (ask once, then go)
- **Topic or hot take** (e.g. "AI will replace dating apps", "breakfast is overrated", a niche: finance, fitness, tech, relationships, true crime).
- **Hosts**: user's own uploaded characters, or create two fictional hosts (age, vibe, outfit).
- **Tone**: funny, heated debate, or "mind-blown" fact. **Language** (default English).
If only a topic is given, choose a tone and two contrasting hosts and go.
### 2. Safety
- Fictional hosts only, or the user's own photos. Never imitate a real, identifiable podcaster or celebrity (face or voice).
- No medical/financial advice presented as fact, no slurs, nothing sexual. Feed-safe.
### 3. Write the exchange (the most important step)
- 2 lines total, max ~25 words together, 8-12 seconds spoken.
- Line 1 (Host A): a bold, specific hook claim. Line 2 (Host B): the reaction that pays it off (funny twist, sharp counter, or "wait, what?").
- Must make sense with zero context to someone scrolling. Show the user 3 options, pick the strongest.
Example: A: "I let AI plan my whole week." B: "And?" A: "It scheduled a nap during my own wedding." (keep it this short)
### 4. Base image (one image, both hosts)
Vertical 9:16 (or 4:5), medium two-shot of both hosts sitting side by side or angled at a table, each with a podcast microphone on a boom arm, headphones optional, cozy studio: warm practical lamps, acoustic panels or bookshelf, soft bokeh, realistic skin texture, natural colors. Looks like a real video podcast frame, not a movie still. No text, no logos.
### 5. Video (one pass, image-to-video with native audio)
9:16, 10-15 s, audio ON, both lines inside the prompt, with clear speaker attribution so the right mouth moves:
"Realistic video podcast footage, static camera, two hosts at microphones. The person on the LEFT says: \"<LINE A>\". Then the person on the RIGHT reacts and says: \"<LINE B>\". Only the speaking person moves their lips; the other listens with natural reactions (nods, smirk, raised eyebrow). Natural blinks, small hand gestures, room tone of a small studio, no music, no subtitles, no text on screen."
If the model mixes up who speaks: regenerate once with the speaker described by appearance ("the woman in the green sweater on the left") instead of position. If still wrong, make two single-speaker clips (close-up of each host, same studio) and deliver them in order.
### 6. Check before delivering (mandatory, use recognize_video)
Always inspect the finished clip with recognize_video before showing it. Reject it if you see any of:
- blocky/pixelated color squares, glitchy patches or flicker on faces or background,
- words missing or garbled, the wrong host speaking a line, lips out of sync,
- morphing hands or mics, extra people, garbled text.
If rejected: regenerate once (add "clean, stable image, no compression artifacts"; use a different video model if available). Never deliver a glitchy clip without telling the user.
Trim to end ~0.4 s after the last word.
### 7. Deliver
- The clip.
- Suggested on-screen hook text for the first 2 s (the user adds it in the editor): 3-6 words, e.g. "AI planned my week 💀".
- Caption: one line + emoji, a comment-bait question ("Team A or Team B? 👇"), 5-6 hashtags (#podcast #podcastclips + niche tags).
- Remind the user to switch on the platform's AI label.
- Offer 3 next episode ideas on the same theme for a series.