Global Consumer Electronics Brand

Every dedication had to produce a unique, real song — not a template variation, not a short clip. Lyrics that captured the specific relationship the user described. Music that matched the emotional tone. Album art generated to match. All in under a minute, at scale.

Three things made this technically demanding:

• Multi-modal orchestration at volume. Three separate AI systems — text generation, music composition, image generation — had to run in sequence, with each output feeding the next.
• Quality control at 700K outputs. No human review step when you're generating songs around the clock across six languages.
• Brand safety. Offensive or off-brand content on a flagship campaign is a PR problem, not just a quality issue.

We built a multi-modal AI pipeline that chains three specialised models:

• OpenAI — generates song lyrics from the user's dedication text, preserving specific details (names, memories, inside references) that make each song feel personal
• Suno.AI — composes original music from the lyrics, with genre and mood matching
• Adobe Firefly — generates album art aligned to the song's emotional tone

The pipeline runs asynchronously on AWS Lambda. Users receive progress updates rather than a loading screen. A content moderation layer screens every output before delivery. The platform launched across 6 languages with localised prompting for each.

The campaign drove sustained organic engagement — people shared songs, which brought new users, which generated more songs. The feedback loop ran without media spend amplifying it.

At 700K generations, the quality bar held. The moderation and prompt engineering work done upfront is what made that possible.