Airbnb Welcome Voiceover Generation

Problem

Welcome content for guest houses is often produced as a continuous promo video, but check-in guidance is easier to verify when one still frame maps to one factual line. This skill turns authorized portraits or room and facility stills, plus confirmed Wi-Fi, parking, check-in times, and written rules, into 2–15s welcome or facility voiceover clips without inventing fees, lockbox codes, rules, or unseen facilities.

Workflow

• Check inputs first: inspect each still’s MIME type, width, height, aspect ratio, byte size, and alpha channel; when a face is present, confirm likeness rights, upload the file only, and do not pass local paths to remote tools.
• Storyboard first: assign each still a 2–15s slot, the spoken line, and the matching frame in check-in order; the storyboard is a plan, not approval.
• Separate voice and video approvals: reuse confirmed audio when present; otherwise list text_to_speech capabilities and voices, show the voice production card, submit one slot at a time, and poll for actual mime_type, duration_seconds, and size_bytes.
• Drive video: list image_to_video capabilities and confirm the card accepts [image, driving_audio]; use the still as a strict first frame, default model to auto, and split overlong voiceover into consecutive in-range segments, each with a fresh client_request_id.
• Deliver uncut: return clips in check-in order with actual size, duration, usage, and billing.net_charged_credits; do not add concat or editing tools.

Boundaries

This is for authorized stills and confirmed stay facts. Use listing-room-video-pack for silent room annotations and talking-avatar-video for general narrator shorts. Stop before paid calls if likeness or voice rights are missing, or if media facts or formats are incompatible; for low balance, relay the official notice and retry only after the user confirms.