Bright white data centre corridor with matte black racks
Story · Data Centres

We turned a ten-frame brief into an unlimited, on-brand image engine.

Imagen lets a non-technical marketer log in, pick an approved look, and generate fresh, on-brief data centre imagery at 4K in under five minutes — without ever going off-brand.

Confidential
Client
Generative Content & Tooling
Discipline
2026
Year
Global · APAC-led
Region
The brief

Ten frames were never enough.

The client’s existing data centre library leaned on generic stock: dark server rooms, neon-blue and purple “tech” lighting, the same technician in every shot. They commissioned a refreshed visual language — bright white architectural shells, matte black racks, controlled reflections, diverse people genuinely at work — and we delivered ten approved hero frames. But ten frames is a moment, not a capability. Every post, deck, report and partner asset created fresh demand, and each new request risked drifting off-brief. The real problem wasn’t the images. It was the bottleneck.

The approach

A pipeline that stays on-brief.

Once the initial production locked the creative direction — the world rules, prompt framework, character library, lighting language and treatment guardrails — we realised that foundation was reusable infrastructure, not a one-off. So we built Imagen: a private, password-protected web app that turns the locked framework into a self-serve generation tool. A user picks one of three approved looks — Architectural realism, People at work, or Cinematic treatment — chooses from curated plain-English parameters, and the app composes a guard-railed prompt behind the scenes and calls the image model on the client’s behalf.

The guardrails are structural, not advisory. There is no free-text prompt box. The server-side composer is the only path to the model, it rejects anything outside the approved allow-list, and it injects the negative guards — no neon, no purple wash, no logos, no fantasy server rooms — into every generation regardless of what the user selects. That is what makes off-brief output genuinely hard to produce rather than merely discouraged.

A human stays in the loop where it counts. Every frame lands in a per-user gallery attributed by name, where the user reviews, favourites the keepers, regenerates or tweaks a parameter, then exports 4K or pushes to a shared folder. NWC owns the versioned prompt library, so the creative can be refreshed centrally without the client ever touching a prompt.

The outcome

From ten frames to a capability.

Imagen moved the client from a finite deliverable to an owned, brand-safe production capability. A marketer can now self-serve three on-brief options for a post in minutes instead of briefing a fresh agency round. Every frame ships at 4K 16:9, ready to drop into decks, social and reports, and because there is no raw-prompt surface, the published output stays on-framework by construction.

Selected stills

From the work.

Inside the tool

The Imagen app.

The AI stack used on this.

The tools, selected for the brief and orchestrated through our COLLIDE-AI pipeline.

Higgsfield
Image API (nano_banana_2, soul_2)
Next.js 16
App framework
Netlify
Hosting, Blobs, edge
Prompt composer
Server-side, guard-railed
shadcn/ui + Tailwind
Component layer

Credits.

Now We Collide / TheBird.AI
Creative & AI direction, prompt framework
NWC AI image studio
Image generation, iteration & finishing
NWC engineering
Imagen app design & build
NWC account & production
Strategy, programme & delivery
All projects NBA Rising Stars Invitational

Like what you see? Let’s build yours.