Skip to main content

​
Overview

The Content Generation feature uses Google’s Gemini AI to transform your topic into a structured presentation. It analyzes your input and creates a complete slide-by-slide breakdown, deciding what content appears on each slide and whether visuals (animations or images) are needed.

​
How It Works

When you provide a topic, the Content Generator:
  1. Structures the Presentation: Creates a logical flow of slides (typically 5-10 slides) that introduces, explains, and summarizes your topic
  2. Determines Visual Needs: For each slide, intelligently decides whether it needs:
    • Text only (most common - 70-80% of slides)
    • An image from Unsplash
    • A Manim animation video
  3. Estimates Timing: Calculates appropriate duration for each slide based on content complexity (typically 4-10 seconds)
  4. Saves the Structure: Outputs a JSON file containing all slide metadata for downstream processing
The system uses mutual exclusivity - each slide can have either an animation OR an image, never both. Most slides are text-only to maintain focus on the narrative.

​
Slide Structure

Each generated slide contains: 
{
  "slide_number": 1,
  "title": "Introduction to Quantum Physics",
  "content_text": "Quantum physics describes the behavior of matter at atomic scales.",
  "needs_image": false,
  "image_keyword": "",
  "needs_animation": false,
  "animation_description": "",
  "duration": 6.0
}

​
Key Fields

  • slide_number: Sequential slide identifier (starts at 1)
  • title: Concise slide heading
  • content_text: Main narration text (2-4 sentences, optimized for TTS)
  • needs_image: Boolean flag for Unsplash image requirement
  • image_keyword: Search term for image fetching (only if needs_image=true)
  • needs_animation: Boolean flag for Manim animation requirement
  • animation_description: Detailed instructions for animation generation (only if needs_animation=true)
  • duration: Slide display time in seconds

​
Animation vs Image Logic

​
When Images Are Used

Images are selected for:
  • Historical figures or famous people
  • Real-world objects, places, or phenomena
  • Static diagrams that support understanding
  • Background context or examples

​
When Animations Are Used

Animations are used very sparingly (maximum 1-2 per presentation) as they’re resource-intensive to generate and render.
Animations are only created when:
  • The concept is impossible to understand without motion
  • Mathematical proofs require visual demonstration (e.g., Pythagorean theorem)
  • Vector operations need directional representation
  • Physical motion must be illustrated (e.g., circular motion with velocity vectors)

​
Text-Only Slides (Most Common)

The majority of slides (70-80%) use text-only format for:
  • Definitions and explanations
  • Lists of concepts or steps
  • Summary slides
  • Theoretical concepts
  • General information

​
Configuration

​
API Settings

Content generation is powered by the Gemini API, configured in config.py: 
GEMINI_API_KEY = os.getenv("GEMINI_API_KEY")
GEMINI_MODEL = "gemini-1.5-flash"  # Fast, cost-effective model

​
Response Format

The generator uses structured JSON output to ensure consistent, parseable responses: 
self.model = genai.GenerativeModel(
    model_name=Config.GEMINI_MODEL,
    generation_config={"response_mime_type": "application/json"}
)

​
Customizing Slide Count

You can adjust the number of slides when generating content: 
content_generator = ContentGenerator()
content = content_generator.generate_content(
    topic="Introduction to Machine Learning",
    num_slides=7  # Default is 5
)

​
Validation & Error Handling

The system includes robust validation: 
@model_validator(mode='after')
def validate_mutually_exclusive(self):
    if self.needs_animation and self.needs_image:
        raise ValueError(
            f"Slide {self.slide_number}: Cannot have both animation and image."
        )
    return self
If Gemini generates invalid content:
  • Missing fields are automatically added with defaults
  • Conflicting flags (both needs_animation and needs_image set to true) are resolved by prioritizing animation
  • Malformed JSON is cleaned and parsed (removes markdown code blocks)
The system prints a breakdown after generation:
Slide breakdown: Text=5 Image=2 Animation=1
This helps you verify the visual distribution across your presentation.

​
Output Structure

Generated content is saved to: 
workspace/source/data/slides/<topic_name>_content.json
Example output: 
{
  "topic": "Introduction to Quantum Physics",
  "total_slides": 5,
  "slides": [
    {
      "slide_number": 1,
      "title": "What is Quantum Physics?",
      "content_text": "Quantum physics describes how matter behaves at atomic scales...",
      "needs_image": false,
      "image_keyword": "",
      "needs_animation": false,
      "animation_description": "",
      "duration": 6.0
    }
  ]
}

​
Best Practices

  1. Specific Topics Work Best: “The Pythagorean Theorem Proof” generates better content than “Math”
  2. Educational Focus: The prompt is optimized for learning content - tutorials, concepts, explanations
  3. Trust the AI: Gemini’s decisions on visual needs are based on educational best practices
  4. Review the JSON: Check the generated _content.json file to verify structure before video generation

Build docs developers (and LLMs) love

Get started for freeTalk to us