Inside a multimodal AI campaign, building Pulsar Underground

The brief was to become a cultural engine, not a broadcast campaign

More than eight in ten young Indians already describe themselves as creators in some form, and that audience has a fairly reliable way of telling a brand it does not matter to them, they simply scroll past the advertisement. Pulsar wanted to be relevant to that exact audience, and a conventional broadcast campaign, however well produced, would only have talked at people who have made it clear they want to be handed tools, not messages.

The brief we agreed on with Pulsar was deliberately open ended, build something that behaves like a cultural engine rather than a campaign with a start and end date, an always on place where fans create, share and compete, with the motorcycle woven naturally into whatever they make rather than pasted on top of it.

What we actually built

We built Pulsar Underground, a mobile first microsite that treats the bike as a creative partner rather than a product to be photographed. Creators choose a challenge across music, dance, art, style or stunts, then build their entry three different ways, fully automated AI generation for anyone who just wants a result fast, a hybrid mode that blends AI assistance with their own creative choices, or a fully manual upload for anyone who wants complete control.

The AI toolkit itself, powered by Google Gemini, includes a stunt video generator that drops a rider into visual styles like Manga and Urban Samurai, an album cover generator, and a mashup engine that blends cultural references into branded visuals. A live leaderboard and direct Instagram integration turned every entry into shareable proof rather than a private submission nobody else would see, which mattered more to participation than the generation technology itself.

What surprised us

We expected the manual upload option to be a fallback for people who distrusted the AI tools. It was not. Across the six week window, close to nineteen hundred of the roughly twenty two hundred total entries were made using the AI toolkit, meaning creators largely treated it as the actual creative instrument rather than a shortcut around making something themselves.

The other surprise was how much the cloud based prompt libraries mattered operationally. Being able to refresh a visual style, add a new creative challenge or retire one that was underperforming without shipping new code meant the platform could react to what was and was not resonating in near real time, something a traditional campaign build would never have allowed for.

What we would carry into the next one

The lesson that stuck hardest is that creators do not want fewer choices because AI is available, they want more of them. Giving people a genuine choice between automated, assisted and manual creation, rather than forcing everyone down one path, is what made the platform feel like a tool rather than a gimmick, and it is the structure we now default to on every creator platform we build.