Work/
AI Celebrity Video

SHARE THE SCREEN WITH A STAR

Category

AI & Automation

Client

Global FMCG Brand

Delivered via

Media.Monks

Role

Automation & AI integration

A national WhatsApp campaign that puts an everyday fan into a video with a film celebrity. Not a voiceover, not a name typed onto a template - the star is on screen, introducing the fan by name and talking about them, while the fan stands right beside them, moving and reacting. All generated from a single selfie and a voice note, with no human in the loop.

Project Overview

The wow is simple to describe and hard to earn: you open WhatsApp, and moments later you are watching yourself in a video, standing next to a star you admire, as they speak about you. The celebrity is really there in the frame and talking; you are really there beside them, alive and reacting - not a static cut-out photo. Everything about that video is assembled by machine, at the scale of a national campaign.

The Challenge

  • 01 Two people, both alive - the star and the fan share one frame, and both have to move and react convincingly, not sit there as a static cut-out
  • 02 A believable placement - the fan has to sit beside the celebrity at the right scale and position so it reads as one shot, not a paste-up
  • 03 Personalised to the individual - the celebrity segment introduces and speaks about this specific fan, by name and from their own story
  • 04 Trust with no human in the loop - every selfie and voice note is user-generated, and one unsafe input inside a celebrity-branded video is a PR incident
  • 05 Scale and language - it had to run unattended, in more than one language, at national campaign volume

The Solution: How the Fan Ends Up Beside the Star

  1. 1 User messages the campaign number, picks a language, and shares their name, a short voice story and a selfie
  2. 2 An AI moderation gate validates the selfie and voice note before anything expensive runs - with retries and coaching if they fail
  3. 3 The selfie is animated into a short clip so the fan actually moves - nodding and reacting, not a frozen photo
  4. 4 Pose detection aligns the fan, and FFmpeg chroma-keys and composites them into the celebrity's footage so the two share one frame
  5. 5 The celebrity segment - the star on screen, introducing and speaking about the fan - is stitched with the personalised segment into one seamless video
  6. 6 The finished video is delivered back over WhatsApp with a share prompt

Technical Architecture

Backend & Infrastructure

  • n8n orchestration, self-hosted on Docker
  • Node.js + TypeScript API, Postgres via Prisma
  • QikChat WhatsApp Business API layer
  • Azure Blob storage with signed URLs
  • FFmpeg chroma-key + overlay compositing

AI Services

  • fal.ai image-to-video - animates the fan's selfie
  • TensorFlow PoseNet - places the fan beside the star
  • OpenAI vision - selfie moderation (structured JSON)
  • OpenAI text - voice-note content moderation

Key Results

On Screen

You + The Star

Human Review

Zero

Lessons Learned

The magic is not "a video was generated" - it is that you are standing next to someone you admire and they are talking about you. Selling that illusion depends on the fan being alive in the frame: animating the selfie so they nod and react is what turns a paste-up into a moment.

The hard engineering problem underneath a celebrity campaign is trust and failure handling. Gating every selfie and voice note with a structured AI check - before the star and the fan are ever composited - protects both the brand and the celebrity's name, with nobody watching.