BREVITTA ROASTERS — AI PRODUCTION PIPELINE
Run Time
Prod Time
Tool Stack
Quality
The Objective
Produce a broadcast-ready luxury coffee commercial entirely through AI, eliminating the need for physical shoots, lighting crews, or macro lenses. Delivere the complete project, from initial brief to final picture and audio lock – in 16 hours.
Pipeline Architecture
- Concept & Scripting (ChatGPT / Gemini): Used LLMs to lock in the brand voice, shot list, and pacing before generating any visual assets. No generation began without a fully resolved brief.
- Static Asset Generation (Nano Banana 2): Built the core visual language against fixed technical parameters: high-contrast chiaroscuro lighting, 100mm macro lens simulation, and deep shadow ratios. These variables were locked and did not change across the asset set.
- Video Generation (Google Veo 3.1): Animated the shots using the same lighting and lens rules established in the static stage. 24fps cadence enforced across all generations.
- Score (Google Lyria 3): Generated a custom cinematic ambient track locked to 80 BPM. That tempo was structural — it was the timing reference every visual cut was built against.
- Assembly & QC (Adobe Premiere Pro): Multi-pass edit. Pass one: clip boundary cleanup, cutting on peak motion to absorb generation artifacts within natural movement. Pass two: audio-picture sync to the 80 BPM grid. Pass three: luxury typography application — wide-tracked serif, consistent hierarchy.
Production Metrics
| Stage | Tool | Time | Iterations | Rejection Rate |
| Concept & Scripting | ChatGPT / Gemini | 1.5 hrs | – | – |
| Static Assets | Nano Banana 2 | 2.5 hrs | ~15 | 60% |
| Video Generation | Veo 3.1 | 5 hrs | ~40 | 75% |
| Score | Lyria 3 | 2 hrs | ~10 | 70% |
| Assembly & Mix | Premiere Pro | 5 hrs | – | – |
| Total | 16 hrs |
Core Production Decisions
Why Veo 3.1: Fluid motion in extreme close-up is where most video models fall apart. Veo handled coffee pours and steam behaviour with consistent physical accuracy at macro focal lengths. Every other model I tested showed drift or deformation in the same shots. That made the decision straightforward.
Why Lyria 3: I needed strict BPM adherence, not approximate tempo. The entire edit was timed to music — a visual action hitting on a specific beat — so tempo drift wasn’t acceptable. Lyria held the 80 BPM structure reliably across all iterations. Other tools I evaluated didn’t.
The editing strategy: Current video models produce degraded frames at clip boundaries — it’s a known characteristic, not an exception. Rather than trying to prompt around it, I designed the edit to account for it. Every hard cut lands on a peak action moment, so the boundary frames are absorbed inside natural movement. The pipeline was built around the model’s behaviour, not against it.
Production Efficiency & ROI
| Metric | Traditional Production | This AI Pipeline |
| Infrastructure | Macro-studio, High-speed Bolt rig | Cloud-based (Veo / Lyria) |
| Crew | Director, DP, Gaffer, Food Stylist | Solo AI Director |
| Post-Prod | Colorist, Sound Designer, Editor | Integrated Editorial |
| Total Overhead | $15,000 – $25,000+ | $0 (excl. subscriptions) |
| Timeline | 2–4 Weeks | 16 Hours |
This is not a cost-cutting exercise. It is a structural shift in how broadcast-quality production gets resourced. The same output standard — previously dependent on a multi-disciplinary crew and dedicated studio infrastructure — now runs as a single-operator pipeline. The budget delta goes back to the client or into iteration cycles that a traditional production schedule cannot accommodate.
Redeployability
This pipeline can be redeployed for any premium brand brief in approximately 6 hours. The lighting logic, lens parameters, and BPM architecture stay fixed. A new campaign requires only new visual seed references and updated brand voice inputs. The system does the rest.
Leave a Reply