kiss-cam

This skill generates a stylized, fabricated "in-arena Kiss Cam moment" of any two subjects using the Pika MCP: a spectator-POV phone shot of an arena Jumbotron with a retro Kiss Cam graphic and scoreboard (via gpt-image-2), followed by a 15-second video clip with PA-announcer commentary and crowd reaction (via kling-v3-omni), first-frame-locked to the still. It is a meme/trend content generator: subjects can be in any visual style (photoreal human, 3D toy, illustrated avatar), the two reference images are passed straight through as the identity anchors, and the skill states that no names appear anywhere in the graphic.

The workflow is a disciplined two-call pipeline with strong operational guardrails. Before any paid call it runs a cost-transparency gate that surfaces the account balance and burn rate and requires the user to reply "proceed" (estimated ~$15-$35) or "cancel." It enforces a pre-generation wall-clock guard (the first paid image call must fire within five minutes), caps prompt-iteration passes, prints stage checkpoints, and uses a tight task_status polling loop with a 15-minute cap and automatic cancel for long-running generations. It also handles intake, uploading local images to presigned URLs, and re-rolling on moderation blocks rather than swapping providers.

The intended users are social/meme creators who want a polished, on-trend synthetic clip from two photos. The notable consideration is that the output is explicitly a realistic-looking fake ("fan-filmed phone shot") of two subjects who may be real people, so the media is synthetic and potentially misleading if presented as genuine, even though it uses no names and enforces cost gates rather than any content-authenticity controls.