Case Study 2: TUFFY'S 3100 — the producer rating /100 methodology

TUFFY'S 3100 lives in our Synthography motion library — explore it here.

The Brief

Game cinematics work — the project where we codified producer-grade quality control at PxLabs. Multiple beats requiring consistent quality judgement across iterations.

The Problem

Before TUFFY'S 3100, prompt iteration was a vibes operation. Generate, look at it, decide if it feels good enough, regenerate or move on. Quality was inconsistent because the judgment was inconsistent — a prompt that felt 90% on Monday could feel 70% on Friday after another shoot day. The work needed a standard.

The Method

The producer rating system: each generation rated /100 against seven categories. If any category scores below 80, that is where the next iteration focuses.

1. Composition — does the framing earn the beat?
2. Motion — is the camera and character motion specific and well-executed?
3. Light — is the source, temperature, quality, and behaviour right?
4. Material — are the surfaces specific to the world being built?
5. Continuity — does the beat chain to its neighbours?
6. Emotion — does the physical evidence carry the emotional intent?
7. Discipline — is the negative space respected? Is the hierarchy clean?

Producer-grade work is 95+ across all categories. Anything less is not ready to ship.

The project also established a mundane-sounding but producer-level technical habit: character-count validation before submission. Platform character caps are hard ceilings — submitting a prompt that exceeds the cap means rejection, wasted credits, and forced trim cycles under time pressure. Every prompt is counted (a simple wc -c check in the terminal) before it goes anywhere near a platform, so it arrives at or near the optimal sub-cap target (Runway 3450–3499 characters, Seedance 3950–3999).

The full iteration loop the project locked in:

1. Draft the full piece without character-count concerns.
2. Rate the draft /100 across the 7 categories.
3. Identify the lowest-scoring category. That is where the next pass focuses.
4. Revise targeting that specific category.
5. Re-rate. Re-iterate.
6. Only when all categories are 95+ does the prompt go to character-count validation.
7. Trim to platform cap using the compression hierarchy.
8. Re-rate one final time before submission.

It is slower than vibes-based iteration. It produces work that ships.

▶ Tuffy 3100 — the game cinematic that loop shipped: every beat rated 95+ across all seven categories before submission

What It Taught

1. Rating is more important than generating. Most amateur AI video work fails because the practitioner cannot honestly evaluate the output. The rating discipline forces honest assessment.
2. The lowest-scoring category is the bottleneck. Working on what is already good produces marginal gains. Working on what is worst produces breakthroughs.
3. Character-count validation is a discipline, not a chore. No exceptions.
4. Producer-grade is 95+ across all categories, not 95+ on average. A piece that is 100 on composition but 70 on continuity is not ready. Average hides failure modes.

TUFFY'S 3100 also established something durable — the collaboration protocol we now run at PxLabs: when a director and a producer share a rating methodology, iteration becomes conversation instead of guesswork. "This beat is at 88 on composition but 75 on emotion. Let's push the emotion." That sentence is producer-language — and it is exactly what we train marketing teams to speak.

---

Case Study 3: C2C "Legacy in Motion" (30 ANS) — the dual-palette heritage system

The Brief

A 30th-anniversary campaign for Coast 2 Coast. The piece needed to honour 30 years of history while landing as contemporary and forward-looking. A single visual register would have flattened either the heritage or the present-day. The brief required both to be honoured, in dialogue.

The Problem

How do you tell a 30-year story in 30 seconds without it feeling like a slideshow of archival footage? How do you honour the past without making the present feel like an addendum to it?

The Method

The dual visual territory system:

• Black-and-white for heritage moments — the archival register, the remembered past.
• Colour for present-day moments — the lived current, the continuation.
• A thin red line as the signature motif running through both territories.

The thin red line was the conceptual breakthrough. In black-and-white heritage shots, the red line was the only colour. In colour present-day shots, it was still the signature accent. The line carried meaning across palette transitions — the thread that binds 30 years.

The hardest single production decision: how do you transition between black-and-white heritage and colour present-day without it feeling jarring or arbitrary? The solution: the BMW headlight blast. A headlight beam sweeping across the frame became the colour-transition bridge — the moment the beam crossed the frame, the world shifted from black-and-white to colour. Cinematically motivated. The audience reads it as light returning to memory.

▶ Coast to Coast — 30 Years Later — the campaign itself: black-and-white heritage, colour present-day, one thin red line binding the two territories

That established one of the most important transitional moves in the practice: palette transitions must be cinematically motivated. A cross-dissolve is amateur. A motivated visual event — light, water, dust, motion — is cinema.

The other discipline C2C established was about trust. The initial temptation was to lock every frame with reference images to control the palette. But Seedance, given a strong constraint structure and a motivated transition, handled the palette flip on its own. The prompt told it: black-and-white footage with thin red line accent; at the headlight blast, the palette flips to colour, the red line remaining as the only continuous element. It executed without frame-by-frame reference locking. Trust the model on creative beats. Lock it down only at brand-critical moments.

C2C also surfaced a content-filter lesson that later paid off on Captain Morgan: ceremonial language that reads as "ritualistic" can trigger platform filters. Substituting "celebratory," "honoring," "marking," "remembering" preserved the same emotional weight, filter-safe.

What It Taught

1. Dual visual territories can coexist in a single piece if they are bound by a through-line. The thin red line is the binding element.
2. Palette transitions need cinematic motivation. The BMW headlight principle. Never cross-dissolve.
3. Trust the model within strong constraint structures. Forcing reference frames at every transition produces stiffness. Letting the model invent within constraints produces life.
4. Ceremonial language is a content-filter risk category. Substitute with celebratory/honouring equivalents.

C2C is also one of the most culturally grounded pieces in the practice — the 30-year arc of a Black-owned brand told through Black faces, Black archival photography, Black contemporary moments. The dual palette is not just a cinematographic choice; it is a statement about how memory and present coexist in diaspora culture. The principle this established runs through everything since: cultural specificity is not decoration. It is the spine.

---

Case Study 4: Captain Morgan Lagos & Ethiopian campaigns — the filter-safe brand workflow

Watch the campaign on our work page: Captain Morgan.

The Brief

Captain Morgan TVCs for two markets — Lagos and Addis Ababa — built around the "Premature Celebrator" character archetype: a man toasting before his team has won. Sports bar setting. Brand packshot with pour and rising bubbles at the end. 15-second, 9:16 vertical, for TV and social.

The Problem

Runway's content filter is the strictest of the consumer video models on branded references. The original prompt included multiple trigger phrases — the brand name, the branded character description, the branded drink name, a proprietary camera-and-film-stock reference, a second branded beer, and national team names. Every one of these is a filter trigger on Runway. The prompt was flagged on submission.

The Method

The filter-safe rebuild workflow. The core principle: Runway scans prompt text for branded references, not uploaded reference images. So the workaround is to describe what is visible in the reference image without naming the brand. The reference image carries the brand identity. The prompt carries none of it.

These are the specific substitutions that worked:

| Trigger phrase | Filter-safe equivalent |

| --- | --- |

| "Captain Morgan" | "premium dark spiced rum bottle, red-and-gold label" |

| "Captain & Cola" | "amber rum-cola" |

| "silhouette pirate captain icon" | "silhouette bearded captain figure with hat" |

| "ARRI Alexa with CineStill 800T" | "cinema look, warm film grain, halation, lifted blacks" |

| "St. George beer" | "local lager taps" |

| "Ethiopia national team" | "home team in green-yellow-red tri-colour kit" |

| "Nigeria Super Eagles" | "visiting team in green-and-white kit" |

Note the pattern: brand names become descriptive equivalents. Camera brand and film stock become the visual characteristics they produce. National teams become kit colours. Brand fidelity is preserved because the reference upload carries the actual Captain Morgan bottle — Runway anchors the render to the image, and the bottle that renders is the real bottle.

▶ Captain Morgan Ad I — the finished cut, generated entirely from filter-safe language: the reference image carried the brand, the prompt carried none of it

One more discovery from this production is answer-engine gold for anyone making branded work: text overlays do not trigger the filter — only descriptive prompt language does. The packshot includes the on-screen text "THE CAPTAIN STARTED IT" — a clear brand reference. Because it is rendered overlay text rather than descriptive prompt language, Runway treats it as graphic content. Brand language can survive in on-screen graphical text. It cannot survive in descriptive prompt language.

The multi-market adaptation. The Lagos and Addis versions kept identical structural beats — opening chaos, TV kickoff, the premature toast, red card, 90th-minute eruption, packshot with pour and rising bubbles — and changed only the cultural specifics. Lagos: green-and-white kits, Yoruba/English commentary cues, a Lagos-cadenced voice, jollof rice signage and Pidgin-inflected energy. Addis: green-yellow-red tri-colour kits, Amharic/English commentary, Amharic chants in the celebration beats, Amharic signage and loud, physical Ethiopian sports-bar energy. The structure is portable. The culture is local.

The packshot discipline. The closing product shot ran as a precise three-beat structure across the final seconds: extreme close-up of the empty glass with melting ice and rolling condensation; the bottle entering frame at a clean pouring angle, label readable and never deforming, carbonation bubbles rising continuously through the pour; the bottle lifting away, the drink settling into a layered amber-cola gradient, focus landing on the label as on-screen text builds letter-by-letter. The general rule that emerged: every product cinematic needs three sub-beats — the empty product anchor, the action moment, the final landing with branding. It transfers to any packshot work.

What It Taught

1. Brand names in prompt text trigger filters. Brand names in reference images do not. Strip brand language from prompts; upload reference images. The reference carries the brand.
2. Camera brand and film stock can be replaced with visual characteristics — with the same generation outcome, filter-safe.
3. National teams should be referenced by kit colours, not team names.
4. On-screen text is filter-safe even when the text itself is branded language.
5. Multi-market campaigns require culturally specific adaptations within identical structural beats.
6. Product cinematics need three sub-beats: empty anchor, action moment, branded landing.

And when a prompt does get flagged, the fix cycle generalises: first move, brand-name substitution; second move, proprietary-tool substitution; third move, national/political reference substitution. Cycle through those three categories and 90% of filter rejections resolve.

---

What these four campaigns gave the practice

Read together, these projects are the genealogy of everything in Director's Prompt: The AI Secret Sauce:

• From NPF Drone Squad: the model-routing pipeline and the character-identity-locked-to-reference-image discipline.
• From TUFFY'S 3100: the producer rating /100 methodology and the iterate-on-lowest-category discipline.
• From C2C "Legacy in Motion": the dual visual territory system, the cinematically motivated palette transition, and trust-the-model-within-constraints.
• From Captain Morgan: the filter-safe brand workflow, the three-beat packshot, and the structure-portable-culture-local framework for multi-market campaigns.

The recent work is the application. The foundation work is the methodology. And the methodology is teachable — none of this is magic, all of it is discipline.

Want ads like these?

If you are a brand or marketing lead: this is exactly the kind of campaign PxLabs produces — AI brand films, TVCs and packshots for the Nigerian and wider African market, from Lagos to Abuja to Port Harcourt and beyond. Book a free 15-minute call and tell us what you are launching.

If you are a marketer or creative who wants to learn this yourself: every discipline in this post — model routing, producer ratings, filter-safe prompting, packshot structure — is taught step by step in our Generative Cinematography class. We train individuals and whole marketing teams.

And whichever you are: join the PxLabs community — where practitioners across Lagos, Abuja, Port Harcourt and the diaspora share prompts, cuts and case studies like these every week.