← Selected workAI Video Pipeline / Shipped 2025

AI Video PipelineThirteen spots. Ten brands. One pipeline.

Thirteen video advertisements across ten invented brand identities. Every script, clip, and voiceover produced through a multi-model pipeline, with Gemini, Veo 3, and Eleven Labs working in sequence under human creative direction.

Sizzle reelEighteen seconds across the thirteen spots.

01 / The problem

Show ad capability at scale, without a production budget.

Tempest needed real, varied video advertisements across industries and brand voices to show clients and investors. The traditional route means copywriters, videographers, voice actors, and editors for each brand. Multiply by ten brands and thirteen ads.

The work was framed as a proof of concept: could one person produce broadcast-quality spots across ten distinct identities with AI tools, fast enough and cheap enough to be worth taking seriously?

The rule

“AI generates. A person directs.”

Between every stage I reviewed output, adjusted prompts, and made the editorial call. The tools are the production team. The creative director is still a person.

  • Gemini
  • Veo 3
  • Eleven Labs
  • DaVinci Resolve

02 / The pipeline

Each model does what it does best.
01 / Script

Gemini

Scripts calibrated to each brand’s positioning, then prompt refinement for the video stage. This is where creative direction happens.

02 / Video

Veo 3

Footage from specific, brand-aligned prompts with shot descriptions, mood, and pacing. Vague prompts produce generic video.

03 / Audio

Eleven Labs

Ten voice profiles with their own pacing, warmth, and authority, so each brand has a consistent audio identity.

04 / Assembly

DaVinci Resolve

Edit, grade, and mix. Every spot cut to :15 with the brand’s logo, positioning line, and end card.

03 / Selected spots

Six of the thirteen.
Out WestSpot 01 · :15
Out WestSpot 02 · :15
Out WestSpot 03 · :15
Hungry HawkSpot 04 · :15
Hungry HawkSpot 05 · :15
Liquid ThunderSpot 03 · :15

04 / The scope

Ten brands, each built from scratch.

Every placeholder company received a name, logo, positioning statement, visual language, and voice profile that informed each downstream decision. The thirteen spots span formats and tones to show varied output, not variations of one template.

05 / What this proves

AI production scales when the pipeline is right.

Anyone can prompt a tool and get a video. The difference is doing it repeatably, across distinct brands, at consistent quality, with a documented process someone else could follow. The pipeline produced thirteen ads for ten brands. It could produce 130.

From spots to a product

A brief in, fifteen formats out.

Explore the AI Ad Generator

Scroll to explore. Press Escape to close.