Build Multimodal Video Scripts with Make.com

Overview

This is a multimodal AI content creation automation solution.

Whether your material is text, audio, or video, it can be automatically transformed into stylized scripts:

1. Material Input - Input text or audio/video links in Notion
2. Smart Recognition - Auto-detect material type and select processing pipeline
3. Content Extraction - Gemini analyzes video / Replicate transcribes audio
4. Segment Processing - DataStore intelligently segments long texts
5. Style Generation - DeepSeek R1 generates social media-optimized scripts

Workflow supporting text, audio, and video multimodal inputs

---

Core Decision Factors

When choosing multimodal content generation solutions, consider:

• Multimodal Support - Can it handle text, audio, video, and other materials
• Content Quality - Logical coherence and stylization level of generated content
• Long Text Capability - Ability to process thousands of words
• Cost-effectiveness - Balance between API costs and output value
• Ease of Use - Complexity of workflow setup and daily operations

---

Technical Specifications

Specification	Value	Notes
Core Platform	Make.com	Workflow orchestration
Database	Notion	Material management & result storage
Video Analysis	Google Gemini Pro	Flash model deep analysis
Material Processing	OpenAI GPT-4o Mini	Initial extraction & processing
Data Storage	Make DataStore	Long text segmentation storage
Script Generation	Volcano Engine DeepSeek R1	Stylized writing
Audio Transcription	Replicate	Fast long audio transcription
Generation Cost	~$1/million tokens	Can generate 300-400k words
Video Analysis Time	150-200 seconds	Wait after video upload
Script Duration	280-300 words/minute	3000 words supports 10-minute video

---

Prerequisites

Before starting, ensure you have:

• Make.com account (free registration)
• Notion account and database
• Google Gemini API key
• Volcano Engine API key (ByteDance DeepSeek R1)
• Replicate account (audio transcription)
• Open-source direct-link cloud storage (for audio/video file storage)

Notion Database Structure

Create material management database with these fields:

• Material Type (Select) - Text/Audio/Video
• Material Content (Text) - Text content or audio/video link
• Status (Select) - Pending/Started/Completed
• Writing Style (Text) - Expected script style
• Additional Requirements (Text) - Other customization needs
• Generated Result (Text) - AI-generated script

---

Multimodal Processing Architecture

Text Material Processing

Directly pass text content to generation module:

Process:

1. Fetch text material from Notion
2. Use GPT-4o Mini for initial processing
3. Pass to DeepSeek R1 to generate script

---

Audio Material Processing

Use Replicate for audio transcription:

Configuration Points:

• Supports long audio (tens of minutes) processing
• Excellent Chinese and English recognition
• More stable than OpenAI official module

Process:

1. Get audio direct link URL
2. Replicate transcribes to text
3. Pass to generation module

---

Video Material Processing

Multimodal workflow module connections in Make platform

Use Google Gemini for deep video analysis:

Configuration Points:

• Upload video file to Gemini
• Wait 150-200 seconds for analysis
• Output precise transcript

Process:

1. Download video and get direct link
2. Upload to Google Gemini
3. Deep analysis to extract content
4. Pass to generation module

---

Long Text Segmentation Processing

Core mechanism to solve large model single output 1000-2000 word limits:

DataStore data storage and flow diagram

Implementation:

1. Smart Segmentation - Divide long materials into 500/1000-word segments
2. DataStore Storage - Save generated content as context
3. Repeater Loop - Generate and accumulate segment by segment
4. Differentiated Prompts - Use different strategies for first and subsequent segments

First Segment Prompt:

Based on the following material, generate the opening part of a script.
Requirements: Conversational social media style, capture audience attention...

Subsequent Segment Prompt:

Continue generating script content, maintaining coherence with previous text.
Previously generated content: {{previous_content}}
Current material segment: {{current_segment}}

---

Stylized Writing

AI-generated script content and segmented layout

Volcano Engine DeepSeek R1’s stylization capabilities:

Features:

• Supports separating thinking process from content
• Transforms serious content into conversational expression
• Adapts to finance, film, parenting, and multiple domains

Style Transformation Examples:

• Economic theory → “Cycle of seasons” metaphors
• Technical jargon → Vivid analogies and storytelling
• Formal language → Social media conversational hooks

News text transformed into social media style script

---

Gotchas

Common issues during setup:

1. Manual Preprocessing - Video downloading and direct-link generation require manual work

2. Learning Curve - Make workflow setup and logic understanding require time investment

3. Over-stylization - DeepSeek R1 may add elements not in original text; requires human review

4. Notion Permission Config - New databases need separate Make authorization access

5. File Size Limits - Make free tier has small file download limits; large videos need manual upload

6. Content Expansion Risk - With limited material, AI expansion may introduce non-original elements

---

Use Cases

Recommended Users

• Content Creators - Short video, live stream professionals needing efficient scripts
• Content Repurposers - Users transforming audio/video materials into text content
• Style Differentiation Seekers - Creators wanting to transform serious content into conversational style
• Efficiency Pursuers - Willing to invest time learning for scale production

May Not Suit

• Users completely unwilling to learn new tools
• Users with extremely high accuracy requirements unwilling to review
• Users resistant to API configuration and third-party tool integration

---

FAQ

What material types are supported?

Supports three types: text, audio (MP3), and video (MP4). Can source materials from YouTube, social platforms, or any video sharing sites.

How to handle long text output limits?

The workflow uses Make DataStore and Repeater modules for intelligent segmentation, with different prompts for first and subsequent segments to ensure context coherence.

Is generation cost high?

Volcano Engine DeepSeek R1 costs ~$1/million tokens, can generate 300,000-400,000 words of scripts. Extremely cost-effective for multiple iterations.

How long does video analysis take?

Google Gemini video analysis takes approximately 150-200 seconds, depending on video length and complexity.

---

Next Steps

After mastering basics, you can try:

• Adding more writing style templates
• Integrating auto-download tools to reduce manual steps
• Adding multi-platform one-click distribution
• Building script quality scoring and filtering mechanisms

Questions? Feel free to leave comments!