Kemal Mermutlu
Broadcast-grade creative, built with AI.Spec work: ElevenLabs × Audi F1 - plus the full workflow, documented so any team could run it.Over a decade in creative production - BBC Radio 1 & Amazon Music - now shipping AI-first.
I built a 10-second, social-first trailer concept inspired by the ElevenLabs × Audi F1 partnership - premium, cinematic, campaign-ready. But the trailer is only half the story. The other half is below: the complete workflow, documented step by step, the way I'd hand it to a customer's production team.What this demonstrates:• End-to-end AI production: visuals → voice → SFX → mix → final export, one operator• A reusable workflow, not a one-off - structured JSON prompting, versioned and documented so the process transfers• Creative judgement - a defined quality bar (stable geometry, believable reflections, clean motion, no distorted logos) and outputs selected against it• Enterprise readiness - deliverables can then be cut for real campaign use: 16:9 master, 9:16 social, 1:1 optionalNote: This is speculative concept work created for demonstration purposes but showcases the type of work I've produced for clients for over a decade.
ElevenLabs x Audi F1 Concept
Push Play
The Workflow - documented as a reusable template
Every step written the way I'd leave it with a customer team: what was done, why, and the exact prompt so it can be re-run without me.1) Reference image of the carStep 1: Reference intake (grounding the creative)
Started with a strong reference image to lock the “world”: proportions, lighting mood, materials (carbon fibre, gloss paint), and the overall premium motorsport finish. This keeps the generation anchored to a specific aesthetic rather than “generic AI F1.”
2) Video software (nodes) with JSON prompt & outputStep 2: Visual generation (node workflow & structured prompting)
Built the visual sequence using a node-based workflow so I could iterate quickly and keep control. I used Sequencer for this prototype. I wrote the prompt in a structured JSON format to make changes deliberate (camera language, lighting, realism constraints) and to document versions cleanly.Quality bar I optimised for: stable geometry, believable reflections, clean motion and avoiding distorted text/logos.
Full JSON Visual Prompt:
{
"description": "A single, unbroken continuous camera shot with NO cuts. The camera performs an infinite macro zoom, starting wide on the Audi F1 car and flying physically INSIDE the materials of the car, dodging carbon-fibre-reinforced polymer (CFRP) and mist before bursting out the other side.",
"style": "High-end CGI, hyper-realistic macro cinematography, 4K, kinetic energy, cold atmosphere",
"camera": "Continuous forward dolly, infinite macro zoom, simulated probe lens, obstacle-dodging motion, no cuts",
"lighting": "Clean studio lighting transitioning to internal atmospheric cold blue light, ending with warm spotlight",
"environment": "Minimalist studio -> Internal Microscopic World -> Concrete Wall",
"elements": [
"Audi F1 car (black, grey and red paintwork)",
"Red synthetic carbon-fibre-reinforced polymer (CFRP) fibers (giant scale)",
"Cold white vapor/mist",
"titanium and aluminum alloys for suspension and engine components",
"Titanium, Zylon & Aramids (Kevlar) (compressed silver pebbles/beads)",
"ElevenLabs Logo"
],
"motion": "Fast, fluid, obstacle-dodging, plunging downward, bursting forward",
"text": "none",
"music": "Whoosh of air, icy wind sound, mechanical crunch of carbon, squeak of titanium alloys, heavy bass impact",
"sequences": [
{
"sequence": 1,
"timestamp": "00:00-00:10",
"action": "In one continuous, unbroken take with no cuts, the camera starts on a completely static profile of the Audi F1 car and accelerates rapidly forward, shrinking to microscopic scale to fly directly into the carbon fibre front panel. Inside the car, the camera weaves and dodges around giant red carbon fiber elements while cold white vapor gusts swirl through the gaps. The camera then dives sharply downward, skimming fast over the rigid, black cross-hatch texture of the Carbon Fiber Flyplate, before pushing aggressively through the titanium and aluminum alloys for suspension and engine components core, squeezing past tight clusters of inflated silver alloy beads. Finally, the camera bursts out the other side, pulling back to reveal the ElevenLabs logo painted on a textured concrete wall."
}
]
}
3) ElevenLabs voice creation for VOStep 3: ElevenLabs voice direction
Moved into ElevenLabs to build a voice that matched the tone: calm, controlled, premium sports broadcast tone. I treated the VO like a real trailer read with short copy, clear emphasis, and pacing designed to land key visual beats.Directing a synthetic voice is still voice direction: copy length, emphasis, pacing to picture - the same discipline as coaching live presenters at BBC Radio 1, applied to a new instrument.
4) Splice library searching SFXStep 4: Sound design palette (Splice)
Pulled a small, purposeful SFX palette: low-end rumble/engine bed, radio static and a one shot for momentum, plus one or two impacts to punctuate transitions. The goal wasn’t “lots of sound,” it was expensive restraint that supports the edit.
5) Ableton Live building audio to match visualsStep 5: Mix & timing (Ableton Live)
Built the final audio in Ableton: aligning VO and SFX to cuts, shaping dynamics and ensuring the reveal moment hits. VO sits forward; SFX support motion and scale.
6) Final videoFinal 10 second social-first trailer
Final export combines AI visuals + ElevenLabs voice + Splice SFX + Ableton mix. Built as 16:9 but can be reframed to work natively on social: fast hook, cinematic pacing and a clear partnership-feel finish.Deliverables: 16:9 master • 9:16 social cut (Reels/TikTok) • 1:1 optional
How this runs inside a creative organisation:The trailer is the artefact. The method is what transfers: find where the tools fit the pipeline a team already has, build the workflow, prove the quality bar, then leave behind the template and the training so adoption sticks.
Working across Creative, Design and Production, that looks like:Map before building. Sit with creative, design and production leads and find where generative tools remove real friction in the pipeline they already have.Prototype against a live brief. Reference content and working prototypes built against an actual piece of work, so the value is arguable in the room rather than theoretical.Set a quality bar and test against it. Every tool assessed on whether its output would survive a client review. The ones that only add speed don't ship. The bar for the film above: stable geometry, believable reflections, clean motion, no distorted logos.Document, then train. Every engagement written up as a reusable template and taught through hands-on workshops. Coaching teams and talent has been my daily discipline since BBC Radio 1.Keep a human where one is needed. A decade delivering to BBC Editorial Guidelines and Ofcom, focused on rights, clearance, consent, multilingual human-in-the-loop review. Responsible use isn't a policy slide; it's knowing which step can't be automated, and saying so early enough to matter.Bring it into the pitch. New capability is worth most when it's visible in the work being sold and demonstrations that show a client something they haven't been shown before.
Also in build: Tone Lab
A voice and tone-of-voice localisation tool with ElevenLabs TTS integration in progress. It productises a similar loop that I ran for years at the BBC's Asian Network - language specialists flagged content, producers owned the editorial call. AI dubbing at scale doesn't remove that judgement; it industrialises the demand for it.
Kemal Mermutlu
3x Sony/ARIAS Awards · 100M+ plays - Diplo & Friends, BBC Radio 1 · £1M+ enterprise creative investment at Amazon Music · Lovable Diamond Expert - Level 5 · Executive MBA (91.78%)