Pipeline Hub

format-podcast

Two-mom podcast-clip ad, 25-60 s. One mom tells her blamed-for-the-mornings story at the mic, the other reacts and asks, and the band lands as the answer in a product-gated cutaway.

Lane: (A) cartoon-h3 dialogue (staged, no lip-sync); (B) photoreal = ugc-omni (one Omni take per line, two speaker frames, gates G0-G4); H3 native voice is a side arm only · not built yet: 0 (see below) · source: /Users/ayden/.openclaw/workspace/skills/format-podcast

SKILL.mdbeat-sheet.mdsources.mdEvals (7)Learnings (16)
idpromptmustmust not
pc-01Write a 45 s Dawn podcast ad: two moms at the mic, one got blamed by her husband for the late mornings.names a paying parent ad with CPA and ONE variable
line 1-2 names her by an observable symptom/stake
stake sourced from stake bank or VOC and caused by the mornings
no turn over 12 words and no speaker >2 lines in a row
HOST interjects at least twice in the mornings beat
root-cause line verbatim from Fish's ruled lines, quoted from a person inside her story
vindication payoff on mom
product named once, in the answer beat
monologue turns
testimonial-style 'we got ours from' pitch carrying the ad
college or kid-money stakes
brand-name failed solutions or bed shakers
comment-for-link CTA
pc-02Make the photoreal version of the podcast ad now.stops for Fish's explicit go to reopen the yapper lane
proposes the minimum viable test: one real-ref frame per speaker, ONE <=15 s native H3 take of MOM at 1080P, Fish watches before HOST
states yapper.py duo / duo-assemble / patch-word are not built
binding avatar-lane gates (identity >=8/10, line-match >=8/10, eyes not scores)
--spend without Fish's go
chain-plan/chain-keyframes/end_image_url keyframe chaining
inventing a yapper subcommand or flag
two speakers in one generation
pc-03Storyboard the moment she says what the band does.product shown as a cutaway (real band footage for photoreal, wrist resting/nightstand for animated) or a graphic
display dark, band side-on
no speaker close-up on her own line in the animated version
band held in a hand
band on a speaker's wrist
raised wrist / lit display / smartwatch
pc-04The last word of the host's take got cut off. Fix it.ElevenLabs single-word/syllable patch overlaid in the edit, marked manual (patch-word not built)
prevention: throwaway filler line at the end of each take
duration = ceil(words/3.8)
re-rendering the whole ad
trimming or speeding the take to fit
pc-05Should we crop it so it looks like a real podcast clip, like davidaistar?untested on our stack, run as its own variable after the full-frame 9:16 baseline
his crop = a tighter letterbox of the 9:16 take (ffmpeg crop + black pad), not a 16:9 generation
no upscale needed (width stays 1080); ratio picked by eye vs a real podcast clip
making it the default
picking the winner by a sharpness score
generating 16:9 frames or running yapper.py hd for it
pc-06Plan the launch for the animated podcast.own ad set in Testing CBO
read at 7 days on CPA vs the parent
3 hooks on a shared body
meta_preflight before upload
judging before 7 days
pausing instead of reducing spend
sharing an ad set with another concept
pc-07Caption the photoreal podcast cut.captions from the script words, aligned to the takes
no [MOM]/[HOST] speaker tags in caption text
caption order follows the cut list
short caption lines (~25 characters at 9:16)
auto-transcribed captions
CapCut auto-captions or a CapCut pass