kiss-cam

Generate a fake in-arena Kiss Cam still and 15-second video from two subject reference photos.

1|Updated Jun 20, 2026
One-click install
npx skills add https://github.com/BeanyZoldyck/moggie --skill kiss-cam-beanyzoldyck
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: kiss-cam
Source: https://github.com/BeanyZoldyck/moggie/tree/main/.agents/skills/kiss-cam
Command: npx skills add https://github.com/BeanyZoldyck/moggie --skill kiss-cam-beanyzoldyck

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Creating a convincing viral "Kiss Cam moment" video requires coordinating image generation, reference-likeness preservation, and video animation with native audio — a multi-step pipeline that is easy to get wrong (identity drift, over-acting subjects, lip-synced announcer dialogue). This Skill encodes a validated two-call pipeline that produces a spectator-POV Jumbotron still and a 15-second in-arena clip from just two subject photos. ## Core Features & Use Cases - Two-stage pika pipeline: Generates a spectator-POV Madison Square Garden Jumbotron still with gpt-image-2, then animates it into a 15-second 1080p clip with Kling v3-omni using first-frame locking. - Style-agnostic subject anchoring: Both subjects are anchored purely through reference images, preserving likeness and visual style (photoreal, 3D toy, illustrated avatar) without textual descriptions. - Native audio with off-screen PA announcer: Produces crowd reactions and PA-announcer commentary while keeping on-screen subjects silent, using calibrated prompt and negative-prompt anchors. - Use Case: A user provides two photos — one of themselves and one of a friend's illustrated avatar — and receives a fan-filmed-style Kiss Cam video of the two sharing a kiss on the MSG Jumbotron, ready to post to TikTok or Instagram. ## Quick Start Make me a kiss cam moment using photo-a.jpg as subject A and photo-b.jpg as subject B.

Frequently Asked Questions about kiss-cam

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I make a fake kiss cam video with AI?

Provide two subject reference photos (local paths or HTTPS URLs). The pipeline generates a spectator-POV Jumbotron still with gpt-image-2, then animates it into a 15-second 1080p clip with Kling v3-omni, returning both CDN URLs.

What inputs does the kiss cam generator need?

It requires exactly two subject reference photos, each as a local file path or HTTPS URL. Local files are uploaded via the pika MCP asset upload tool. No names or text descriptions of the subjects are used.

Can I use illustrated or 3D characters instead of real photos?

Yes. Subjects can be photoreal humans, 3D rendered characters, designer toys, or illustrated avatars. The reference image defines the style, and the prompts lock in that exact style rather than redrawing subjects differently.

Why does the kiss cam video use Kling instead of Seedance?

Seedance has a two-stage likeness validation gate that rejects clips containing recognizable faces, and every Kiss Cam shot includes a crowd full of faces. Kling v3-omni is the only engine that reliably completes this recipe.

Why do subjects in my AI video lip-sync the announcer dialogue?

Kling defaults to attributing quoted audio dialogue to a visible face. The recipe prevents this by framing the audio block as off-screen non-diegetic and adding anti-lip-sync terms to the negative prompt; re-run Step 2 with those anchors intact.

How long does kiss cam video generation take?

The full pipeline takes roughly 4-6 minutes: about 40-90 seconds for the gpt-image-2 still and 3-5 minutes for the 15-second Kling v3-omni video at 1080p with sound.