presenter-videolisted

# Presenter Video Make a verified presenter video from a script or topic and one photo, with fal.ai models. One rule organizes everything: **the locked narration is the master clock.** The presenter is generated from it, captions are timed to it, cuts land on its pauses, and the delivered video lasts exactly as long as it does. Scripts live in `scripts/` inside this skill's base directory (shown when the skill loads). Below, `$SKILL` stands for that directory and `$JOB` for the job folder. Run scripts with `python3 "$SKILL/scripts/<name>.py"`. Every script prints JSON and exits non-zero on failure. ## Non-negotiables - **Consent before upload.** The user confirms they hold the rights to the photo, that the presenter is an adult, and that uploading to fal.ai is fine. Cloning a voice also needs the voice owner's authorization. Never guess a real person's voice from a photo, and never make a real person say something they did not approve. - **Quote every paid call.** Before the first paid call, show a quote built from live prices: files uploaded, endpoint, units, unit price, subtotal, pilot size, retry ceiling. Wait for a clear yes. `fal.py run` refuses to start without `--approved`, so pass that flag only for the plan the user approved, and quote again whenever the model, duration or retry count changes. - **Pilot first.** Generate the shortest useful segment, review it, then run the rest. - **Resume instead of paying twice.** Every paid call is logged to `$JOB/qa/requests