AI Video Ad Generator - Automatic video ad generation system with Sora 2

Process description

Automation for creating authentic UGC (User-Generated Content) video ads using OpenAI's latest Sora 2 model. The system analyzes the product image, creates a detailed profile of the ideal ambassador, generates a 12-second “raw” video script in the iPhone-style shooting style, and automatically creates the finished video. The whole process from uploading a product photo to getting a professional video takes only 5-10 minutes, replacing days of work by an entire production team.

API keys and services:

1. OpenAI API - GPT-4.1 for analysis and Sora 2 for video generation
2. Google Gemini 2.5 API - image processing and prompt generation
3. Google Drive OAuth2 - storing ready-made videos
4. N8n Form Trigger - download interface

System architecture by blocks

SECTION 1: DATA INITIATION AND PREPARATION

1.1 Form Trigger — Entry Point

Purpose: Web form to download the product and enter the name

Form settings:

• Form Title: “Video Generator”
• Field 1: Product (file, required)
• Field 2: Product Name (text, required)

Interface: Simple drag-and-drop form for product images

1.2 Extract from File - Image conversion

Purpose: Extracting binary image data for processing

Settings:

• Operation: BinaryToProperty
• Binary Property Name: Product

1.3 Convert to File — AI Preparation

Purpose: Conversion to a format for transfer to AI models

Settings:

• Operation: Tobinary
• Source Property: data

SECTION 2: AI ANALYSIS AND PERSONA CREATION

2.1 analyze_product - In-depth product analysis

Purpose: OpenAI Vision analyzes the product and creates a detailed profile of the target user

Settings:

• Model: chatgpt-4o-latest
• Operation: analyze
• Input Type: base64

Mega-prompt for analysis (abbreviated):

//ROLE AND PURPOSE//

You are an expert in casting and consumer psychology...

Your only task is to analyze the product and create

A detailed profile of the ideal person for UGC advertising.

‍

//PROFILE STRUCTURE//

I. Basic identity (name, age, location, profession)

II. Appearance and style (detailed description of the image)

III. Personality and communication (demeanor, speech style)

IV. Lifestyle (hobbies, values, pain points)

V. The rationale for trust (why this particular person)

```

Result: A full character profile of 500+ words

2.2 set_model_details - Save profile

Purpose: Structuring personal data for the following stages

Settings:

- Assignment: prompt = $json.content

SECTION 3: VIDEO SCRIPT GENERATION

3.1 set_build_video_prompts - Preparing a master prompt

Purpose: Creating a detailed 12-second UGC video script

Key elements of the promotion:

```

Master Prompt: Raw 12-second UGC video script

AESTHETICS:

✓ Shaking when shooting handheld

✓ Natural camera movement

✓ Real lighting and locations

✓ Authentic imperfections

WE AVOID:

✗ Tripods or stabilization

✗ Text overlays

✗ Professional installation

✗ Clean backgrounds

FRAME-BY-FRAME STRUCTURE:

[0-2 sec] A clue is the middle of a conversation

[2-9 sec] Product demonstration in action

[9-12 sec] Natural completion

```

3.2 generate_ad_prompts - Creating the final script

Purpose: Gemini 2.5 Pro generates a detailed frame-by-frame scenario

Settings:

- Model: gemini-2.5-pro

- Endpoint: GenerateContent

- Input: Prompt+ product image

Output: Per second breakdown with a description of each frame, camera movements, and dialogues

3.3 Message a model - Clearing a prompt

Purpose: GPT-4.1 clears the generated text from unnecessary comments

Prompt:

```

Filter comments, leave only a clean prompt for the video

SECTION 4: PREPARING THE FIRST FRAME

4.1 generate_frame - Format adaptation

Purpose: Gemini adapts the product image to the 9:16 vertical format

Settings:

• Model: gemini-2.5-flash-image-preview
• Special prompt: Adapting to the aspect ratio while preserving the composition

Technique: Intelligent background extension without distortion

4.2 set_frame_result - Extracting the result

Purpose: Parsing Gemini's response to get a base64 image

Expression:

jascript

$json.candidates [0] .content.parts.filter (item => Item.inlineData) .first () .inlineData.data

4.3 get_frame_image+resize_image

Purpose: Final preparation of the first frame

Resize parameters:

• Width: 720px
• Height: 1280px
• Option: IgnoreasSpectratio

SECTION 5: GENERATING VIDEOS VIA SORA

5.1 generate_video — Launching Sora 2

Purpose: Submitting a video generation request to OpenAI Sora

API parameters:

• Endpoint: https://api.openai.com/v1/videos
• Model: sora-2
• Duration: 12 seconds
• Size: 720x1280 (vertical format)
• Input: First frame+detailed prompt

Request format:

json

{

“prompt”: “[Detailed video script]”,

“model”: “sora-2",

“seconds”: 12,

“size”: “720x1280",

“input_reference”: “[base64 first frame]”

}

```

5.2 Status monitoring cycle

Components:

1. delay (Wait) - Wait 15 seconds

2. get_video_status - Checking the generation status

3. check_status (If) - Completed/processing check

4. Cycle - Return to delay if not ready

Logic:

```

generate_video → delay → get_video_status → check_status

↑___________________________|

```

5.3 get_video - Download the finished video

Purpose: Getting the final video after the generation is complete

Endpoint: https://api.openai.com/v1/videos/ {id} /content

SECTION 6: SAVING THE RESULT

6.1 upload_video — Uploading to Google Drive

Purpose: Automatically save the finished video

Settings:

- Name: Video # {runIndex + 1}

- Drive: your google drive

- Folder: your folder

- Credentials: Google Drive OAuth2

Required services and their configuration

OpenAI setup:

1. API key with access to:
   
   • GPT-4.1 or GPT-4o
   • ChatGPT Vision API
   • Sora 2 API (requires special access)
2. Limits:
   
   • Sora 2: Check video generation quotas
   • Recommended balance: $50+ for testing

Google Gemini setup:

1. Obtaining an API key:
   
   • Google AI Studio
   • Enable Gemini 2.5 Flash and Pro
2. Setup in N8n:
   
   • HTTP Header Authentication
   • Header: x-goog-api-key

Google Drive setup:

1. OAuth2 connection:
   
   • Create a project in Google Console
   • Enable Drive API
   • Configure OAuth2 credentials

System capabilities

Types of generated content:

1. UGC video style:
   
   • Product reviews
   • Unboxing
   • First impressions
   • Usage tutorials
   • Before/after comparisons
2. Formats and styles:
   
   • iPhone selfie video
   • Mirror photography
   • POV demonstrations
   • Lifestyle content
   • Testimonial video

Unique features:

• Hyperrealism - simulating a real shoot on a phone
• Personalization - a unique character for each product
• Authenticity - natural defects and imperfections
• Speed - 5-10 minutes vs production days

Use cases

1: Cosmetic product

Login: Photo face serums

AI analysis creates a persona:

• 28 year old woman, graphic designer
• Minimalist style, natural beauty
• Speaks fast, enthusiastically

The result: A 12-second video of a morning routine with natural light from the bathroom window

2: Technical gadget

Login: Wireless headphone photos

AI analysis creates a persona:

• 32-year-old male developer
• Casual style, glasses, beard
• A calm, technical way of speaking

The result: Unboxing video on a cluttered desktop with code on the monitor at the back

3: Food product

Login: Photo of a protein bar

AI analysis creates a persona:

• 25 year old woman, fitness trainer
• Sportswear, high ponytail
• Energetic, motivating delivery

The result: Video of a snack after a workout in the gym locker room

Advanced settings and optimization

Fine-tuning prompts:

1. Regional adaptation:
   
   • Accents and dialects
   • Local references
   • Cultural characteristics
2. Platform optimization:
   
   • TikTok: More dynamic editing
   • Instagram Reels: aesthetics
   • YouTube Shorts: informative
3. Audience targeting:
   
   • Gen Z: fast paced, memes
   • Millennials: Authenticity, Stories
   • Gen X: practicality, details

Advanced techniques:

1. A/B testing:
   
   • Different people for one product
   • Scenario variations
   • Different locations and lighting
2. Mass production:
   
   • Batch product processing
   • Automatic series generation
   • Cross-product campaigns

Integrations and extensions

Possible additions:

1. Social Media Publishing:
   
   • Auto posting on TikTok/Instagram
   • Publication planner
   • Hashtag generator
2. Performance Tracking:
   
   • Connecting to Meta Ads
   • Google Analytics integration
   • ROI calculator
3. Brand Safety:
   
   • Content moderation
   • Brand guidelines check
   • Compliance filters
4. Multimodal generation:
   
   • Add music/sounds
   • Subtitles and captions
   • Multilingual versions

Practical value

For small businesses:

• Affordable video marketing no production budget
• Launch campaigns quickly for new products
• Testing creatives at minimal cost

For agencies:

• Scaling up production without increasing the team
• Customization for customers commercially
• Innovative offer for customers

For e-commerce:

• Videos for every SKU in the catalogue
• Dynamic generation for seasons/promotions
• Content localization for different markets

The result of the system

This automation is a full-fledged video studio in the cloud that creates professional-grade content in minutes. The combination of product analysis through AI, the generation of authentic scenarios and the use of advanced Sora 2 technology makes possible what seemed fantastic a year ago - the instant creation of realistic videos without actors, studios and equipment.

An example of automation: