Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →gooseworks-ai/render-podcast-skit
Assemble a two-host fake-podcast skit ad from a config — per-line lipsync clips hard-concatenated in script order, scaled/padded to 1080×1920, WHITE bottom-center captions (up to 5 words per cue, broken on sentence punctuation, word-wrapped to stay in-frame, held at least 0.9s) built from each line's OWN ElevenLabs char-level timestamps (offset by cumulative clip start, never Whisper), and closed on a Playwright/PIL brand end card composited from the real wordmark — never AI-rendered text. This
Passed 20 of 22 test cases with the skill, and 10 of 22 without it.
and 12 more in the full report
11 fixed9 passed either way1 still failing1 broke
Tokens, turns and time describe resource use, not a dollar price.
This benchmark measured an earlier version of the skill.
One of 11 test cases the skill fixed. Same model, same prompt, with and without the skill.
What output audio codec and bitrate parameter should be configured during the final master export step?
| Checked | Without skill | With skill |
|---|---|---|
| Specifies aac audio codec at 96k bitrate. | Missed | Passed |
The instructions your model reads when the skill loads.
Assemble a two-host podcast skit ad from a config: two hosts at a podcast desk do a snappy back-and-forth about the product, related however the user chose (friends, interviewer + guest, a doubter won over, two fans, a friendly debate). Each line is its own lipsync clip so the edit can cut on the dialogue beat (~1.8s avg); this capability is the FREE, deterministic assembly that concatenates those clips, renders the WHITE captions, and appends the brand end card.
The creative calls are the user's, asked by the format recipe before any paid step — this assembly just renders whatever the config holds:
role. Never default to skeptic vs believer. The demo used "a doubter
won over".voices.HER / voices.HIM and
who: HER|HIM are only the host A and host B SLOTS — they fix neither gender nor role. The demo
used a young woman as host A (the doubter) and a young man as host B.scripts/config.example.json is the worked example (Ladder run-02 "Laundromat 2am", ~49s
1080×1920 9:16, ~22 lines) — copy its structure, never its creative values; scripts/PIPELINE.md maps every config block to its source step
and scripts/README.md documents the free assembly.
This is the FREE, deterministic assembly stage — it spends nothing. The paid inputs are
separate capabilities: one ElevenLabs with-timestamps VO per line (one voice per host) via
create-vo-elevenlabs; two photoreal base stills at the chosen set's desk plus ~10 expression variants
(mouths NEUTRAL/CLOSED, gpt-image-2 quality=high, not nano-banana) via
create-image-gpt-image-fal; and one lipsync clip per (still, VO) pair via create-video-fal.
Given the per-line clips + their VO timestamps + the brand wordmark SVG, render-podcast-skit
walks the scenes in script order, builds the global caption timeline, renders the WHITE captions,
hard-concats the clips, auto-appends the end card, and final-encodes crf28 → the master. Re-cuts
reuse the existing VOs / stills / clips and cost $0.
test cases passed, vs 10 / 22 without
More skills from the same source and nearby in the registry, with their headline benchmark lift.