Skip to content
Isaac Nsubuga

Evidence lens

The movement behind the creative artifact.

Recognized pain
Challenge — create a faceless 15-second product ad that protects the founder's identity, preserves the real garment and its logos exactly as photographed, and avoids unnecessary ElevenLabs credit use.
Existing value
Real product photography, approved identity assets, and an existing voiceover were already available.
Constraint
Success meant a finished vertical master where the product is unmistakably the real product, no face or body is generated, and every paid generation is deliberate rather than exploratory.
Decision
Strategic decision — generate only the one shot where motion actually added value. Everything else stays source pixels: the original product photograph, the angular-G identity asset, and the GRAPHITE wordmark are used as scenes, not as prompts.
Activation
What I built — a 15-second vertical product ad; identity-safe prompt guardrails; an audio system using the approved Joshua voiceover, a cinematic heartbeat bed, fabric SFX, and an end-card stinger; a credit-governance workflow; and reusable production documentation.
Evidence
Outcome — a portfolio-ready 15-second master and a repeatable workflow that can produce the next product ad without re-solving identity, audio, or credit questions.
Economic consequence
Exploratory generation risked unnecessary credit spend and inaccurate reconstruction of the product or founder.
What remains unproven
Audience response, conversion performance, and reuse across future garment drops have not yet been measured.
← All evidence

Portfolio case study · AI creative strategy

GraphiteBuilt different

How I turned a real family streetwear project into an identity-safe, multi-layer AI campaign system—and learned where automation creates leverage, where it creates risk, and where human judgment must remain in control.

  • Creative strategy
  • AI production
  • Brand systems
  • Workflow design
  • Identity protection

The brief

Build proof before scale.

GRAPHITE is a real family streetwear project, so the first client was the hardest kind: people whose product, name, and likeness actually carry the risk. Serving family first made this a genuine validation environment rather than a portfolio exercise.

That turned “make an ad” into something larger — a test of product storytelling, identity-safe content, audio direction, and repeatable operations. If the method held here, it would hold for the next drop.

Start with product truth.

GRAPHITE streetwear garment photographed for the campaign, logo and stitching detail visible

Non-negotiable constraints

  • Preserve garment details and logos exactly as photographed
  • Founder remains faceless
  • No reconstructed face, beard, scar, limbs, or unseen anatomy
  • Concept art kept separate from physical merchandise
  • Typography stays editable

Diagnose → Clarify → Design → Leverage

01

Diagnose

Name the real constraint before touching a tool: identity exposure, product fidelity, and unbounded generation cost.

02

Clarify

Define what a finished master must be true of — faceless, real garment, deliberate spend — so approval is objective.

03

Design

Storyboard first. Decide which single shot earns motion and which scenes stay source pixels.

04

Leverage

Turn the one-off build into guardrails, an audio bed, and documentation the next drop can reuse.

One controlled motion shot. Three source-preserving scenes.

GRAPHITE — 15-second vertical product ad

Play the completed 15-second portfolio cut.

Why only one generation

Motion was generated only where it added value. Everything else in the cut preserves source pixels — the original product photograph, the angular-G identity asset, and the GRAPHITE wordmark are used as scenes, not as prompts.

Static holds, text, transitions, and mixing are handled locally, so the garment on screen is the garment that was photographed.

Identity evolution

GRAPHITE angular-G logo mark with textured graphite fill and EST. 2024 lockup
The angular-G identity asset — used as a source-pixel scene, never regenerated.
GRAPHITE wordmark lockup reading Built Different. Made to stand out. GRAPHITE Streetwear, EST. 2024
The GRAPHITE wordmark lockup that closes the cut on the end card.

Campaign architecture

  1. 0–3s

    Hook

    Controlled motion opens the piece — the only generated movement in the cut.

  2. 3–7s

    Product truth

    The real garment photograph holds the frame. Logos and stitching unaltered.

  3. 7–11s

    Identity

    The angular-G asset carries the brand without generating a person.

  4. 11–15s

    End card

    GRAPHITE wordmark and stinger. Music rises after 9.3 seconds, beneath the voice until then.

Operational lesson

Automation does not create clarity.

It scales whatever the workflow already contains—including ambiguity, retries and waste.

System redesign

Preflight first

Assets, prompt, and intent reviewed before any paid generation runs.

One corrective retry

A failed generation gets one fix attempt, then the approach changes — not the budget.

Human judgment boundary

Identity, anatomy, and product fidelity are never delegated to a model.

Asset reuse

Approved voice, music bed, and identity assets are recovered, not regenerated.

Render gate

Nothing renders to master until QA on identity, product, and audio passes.

Credit governance

Generate once. Review once. Reuse many times.

Outcome

A master, and a method.

A portfolio-ready 15-second vertical master, plus a repeatable workflow that can produce the next product ad without re-solving identity, audio, or credit questions.

Portfolio truth statement: the approved motion and audio originated through the ElevenLabs and Runway Gen-4 Turbo workflow. The final 15-second master was assembled locally from those approved assets — it was not rendered inside ElevenLabs.

  • ElevenLabs
  • Runway Gen-4 Turbo
  • Local video assembly
  • ChatGPT / Codex
  • Lovable
Strategic AI creative production case study by Isaac Nsubuga · GRAPHITE “The Return”