Automated Car Commercial Generation Workflow

Portfolio project

CASE STUDY: Automated Car Commercial Generation Workflow

Client Brief & Challenge

The Client Request A automotive marketing agency approached me with a specific need: they wanted to generate professional car commercials quickly and cost-effectively. Their requirements were:

Client Needs:

Simple Input System - Only 2 inputs: car model + city location

Multiple Camera Angles - 7 different cinematic shots per commercial

Driving Footage - Car must be shown in motion, not static

Beautiful Environments - Cinematic city settings with proper lighting

Professional Quality - Broadcast-ready output

Easy Variation - Ability to change car/city instantly for new videos

Mini-Movie Feel - Complete narrative arc, not just random clips

The Problem:

Traditional car commercial production costs $50,000-$150,000 and takes weeks. They needed:

Cost: Under $10 per commercial

Speed: 10-15 minutes per video

Scale: Ability to produce hundreds of variations

Consistency: Professional quality every time

My Analysis & Strategic Approach

Why Traditional Solutions Failed

Option 1: Stock Footage

❌ Limited variety

❌ Generic, not brand-specific

❌ Can't customize car models or cities

❌ Licensing costs add up

Option 2: Manual AI Generation

❌ Requires 7 separate prompts

❌ Inconsistent style between shots

❌ Time-consuming prompt engineering

❌ No automation

Option 3: Traditional Video Editing

❌ Still requires source footage

❌ Labor-intensive

❌ Not scalable

My Solution Strategy

I identified that the client needed automated orchestration of AI video generation with:

Intelligent prompt generation - AI creates cinematography descriptions

Systematic shot variety - Predefined camera angle archetypes

Workflow automation - No manual intervention between inputs and output

Style consistency - Unified aesthetic across all 7 shots

Modular architecture - Easy to modify parameters

The Weavy Workflow Solution

Architecture Overview

I designed a node-based pipeline that transforms 2 text inputs into a complete commercial:

INPUT LAYER → PROCESSING LAYER → GENERATION LAYER → OUTPUT LAYER

[Car Model Input] ──┐
                    ├──→ [Prompt Concatenator] ──→ [GPT-4] ──→ [Array Splitter] ──→ [Text Iterator] ─┬──→ [Veo #1: Wide Shot]
[City Input] ───────┘         ↓                       ↓              ↓                                  ├──→ [Veo #2: Tracking]
                         Master Prompt          7 Descriptions   Distribution                          ├──→ [Veo #3: Aerial]
                         Template               Split by ***     Mechanism                             ├──→ [Veo #4: Close-up]
                                                                                                       ├──→ [Veo #5: POV]
                                                                                                       ├──→ [Veo #6: Action]
                                                                                                       └──→ [Veo #7: Hero]
                                                                                                            ↓
                                                                                                    [Video Concatenator]
                                                                                                            ↓
                                                                                                    [Final 45-sec Commercial]

Detailed Workflow Breakdown

Layer 1: Input Collection (Nodes 1-2)

Node 1: Car Model Text Input

Purpose: Capture specific vehicle

Example: "2024 Porsche 911 Turbo S"

Why separate node: Enables instant car swapping

Node 2: City Location Text Input

Purpose: Define environment and lighting

Example: "Tokyo at night"

Why separate node: Enables instant location changes

Design Rationale: By separating these as independent nodes rather than a single combined input, I enabled maximum reusability. The client can generate 100 different combinations from just 10 cars × 10 cities without rebuilding the workflow.

Layer 2: Intelligent Prompt Generation (Nodes 3-4)

Node 3: Prompt Concatenator

Function: Combines user inputs with master cinematography template

Contents:

User inputs (car + city)

Cinematography instruction set

Shot type definitions

Technical requirements

Master Prompt Template Structure:

[Car Model] driving through [City]

You are an award-winning automotive cinematographer. Generate 7 cinematic shot descriptions.

SHOT TYPES REQUIRED:
1. Wide Establishing - Full scene context
2. Low Angle Tracking - Speed and power
3. Aerial Drone - Urban landscape overview
4. Close-Up Detail - Car craftsmanship
5. Driver POV - Immersive perspective
6. Dynamic Action - High-energy maneuver
7. Hero Beauty - Signature money shot

REQUIREMENTS:
- Car DRIVING in every shot (wheels turning, motion)
- Consistent time of day (night with city lights)
- 30-40 words per description
- Cinematic camera movements
- Professional lighting terminology

FORMATTING:
- Separate shots with ***
- NO numbering or labels
- Pure descriptions only

Generate now:

Design Rationale: This template was refined through 15+ iterations to achieve:

Specificity: Detailed enough for quality output

Flexibility: Works with any car/city combination

Consistency: Enforces unified aesthetic

Technical precision: Uses professional cinematography language

Node 4: GPT-4 LLM

Model Choice: GPT-4 (not 3.5) for superior creative writing

Temperature: 0.7 (balanced creativity + consistency)

Function: Generates 7 unique, professional shot descriptions

Why GPT-4:

Better understanding of cinematography terminology

More creative variation within constraints

Consistent quality across generations

Follows complex multi-part instructions reliably

Wide establishing shot of 2024 Porsche 911 Turbo S cruising Tokyo's neon-lit expressway at night, elevated camera capturing sweeping cityscape with illuminated skyscrapers, LED headlights cutting through mist ***

Low angle tracking shot racing alongside the Porsche as it accelerates through Shibuya, camera hugging asphalt, wheels spinning, city lights streaking, aggressive stance emphasized ***

Aerial drone view circling above the Turbo S navigating Tokyo's elevated highways, bird's eye perspective revealing intricate road networks, car's sleek form cutting through urban geometry ***

[... continues for all 7 shots]

Layer 3: Distribution System (Nodes 5-6)

Node 5: Array Splitter

Function: Converts single text block into 7 separate items

Split Character: *** (three asterisks)

Output: Array of 7 independent descriptions

Why This Is Critical: Without the Array node, all 7 descriptions would go to EVERY video generator, creating duplicate content. The Array creates clean separation.

Technical Detail:The *** separator was chosen because:

✅ LLMs reliably output it ✅ Unlikely to appear in natural text ✅ Easy to parse programmatically ✅ No whitespace sensitivity issues

Node 6: Text Iterator

Function: Distributes array items sequentially to connected nodes

Mechanism: Round-robin distribution

Result: Shot 1 → Veo #1, Shot 2 → Veo #2, etc.

Design Rationale: The Text Iterator is the "traffic controller" of the workflow. It ensures each video generator receives EXACTLY ONE unique shot description. This creates:

✅ Shot diversity (7 different angles)

✅ No duplicates

✅ Predictable ordering

✅ Scalability (could expand to 10, 15, 20 shots)

0 likes · 47 views · Feb 15, 2026

A automotive marketing agency approached me with a specific need: they wanted to generate professional car commercials quickly and cost-effectively.