Transform any concept into comprehensive learning experiences with AI-powered videos, personalized roadmaps, and interactive games.
Bloop revolutionizes educational content creation by generating fully animated explainer videos from simple text queries or documents. Unlike traditional educational AI systems that produce static text or slides, Bloop creates dynamic visual explanations with synchronized narration and optional photorealistic talking avatars - all generated on-demand without precomputed assets.
- 🎨 Programmatic Animation Generation - Create stunning Manim animations automatically from educational concepts
- 🎙️ Scene-Aligned Narration - Neural TTS with perfect temporal synchronization
- 🤖 Talking Avatar Integration - Photorealistic avatars using SadTalker for human-like delivery
- 🧠 LLM-Orchestrated Pipeline - Intelligent scene planning and automated error repair
- 🎯 Fault-Tolerant Rendering - Scene-isolated execution with graceful degradation
- 📚 Document-Based Learning - Generate videos from PDFs, textbooks, or research papers
- 🗺️ Personalized Roadmaps - AI-generated learning paths tailored to your level
- 🎮 Interactive Learning Games - Engaging educational games with voice interaction
final_video.mp4
Complete educational video with animation and avatar in production quality
manim_video.mp4
Real-time generation of mathematical visualizations with LaTeX rendering
avatar_generation.mp4
Photorealistic avatar with synchronized lip movements and natural expressions
AI-generated personalized learning paths with milestones and resources
Engaging educational games: Teach the AI, Drag & Drop, Identify the Error
Automatic flashcard generation from documents with spaced repetition
Intelligent quiz creation with multiple question types and adaptive difficulty
Bloop employs a sophisticated microservices architecture designed for real-time video generation:
Client Query
│
▼
FastAPI Gateway (QA Service)
│
├─► LLM Answer Generation
│ └─► Document Analysis & Understanding
│
└─► Manim Video Service
│
├─► Scene Planning (LLM → JSON)
│ └─► Structured Scene Graphs
│
├─► Scene-by-Scene Rendering
│ ├─► Independent Execution Context
│ ├─► Error Detection & Capture
│ ├─► LLM-Based Auto Repair
│ └─► Scene Skipping (if unrecoverable)
│
├─► TTS Generation (Per Scene)
│ └─► Neural Text-to-Speech
│
├─► FFmpeg Audio-Video Alignment
│ ├─► Temporal Synchronization
│ └─► Duration Padding
│
└─► Scene Stitching
└─► Final Animation Output
│
└─► (Optional) SadTalker Avatar
├─► Audio Extraction
├─► Avatar Rendering
└─► Side-by-Side Composition
- Python 3.8+
- FFmpeg
- LaTeX (for mathematical expressions)
- GPU (optional, for faster avatar generation)
- CUDA (optional, for SadTalker acceleration)
git clone https://github.com/poorvikab/bloop.git
cd bloop# Create virtual environment
python -m venv venv
source venv/bin/activate # On Windows: venv\Scripts\activate
# Install dependencies
pip install -r requirements.txt
# Install Manim
pip install manim
# Install additional dependencies
sudo apt-get install libcairo2-dev pkg-config python3-dev
sudo apt-get install ffmpeg
# Set up environment variables
cp .env.example .env
# Edit .env with your configuration# OpenAI Configuration
OPENAI_API_KEY=your-openai-api-key
AZURE_OPENAI_ENDPOINT=your-azure-endpoint # Optional
# TTS Configuration
TTS_PROVIDER=azure
AZURE_SPEECH_KEY=your-speech-key
AZURE_SPEECH_REGION=eastus
# SadTalker Configuration (Optional)
SADTALKER_MODEL_PATH=./models/sadtalker
ENABLE_AVATAR=true
# File Paths
UPLOAD_DIR=uploads
OUTPUT_DIR=outputs
TEMP_DIR=temp
# Performance
MAX_SCENE_RETRIES=1
SCENE_TIMEOUT=120
PARALLEL_SCENES=false# Development mode
uvicorn main:app --reload --host 0.0.0.0 --port 8000
# Production mode
gunicorn main:app --workers 4 --worker-class uvicorn.workers.UvicornWorkerVisit http://localhost:8000/docs for interactive API documentation.
Bloop's Ask mode combines all the capabilities of NotebookLM with powerful visual content generation:
Bloop generates educational videos in real-time through a sophisticated pipeline:
Scene Planning via LLMs
# Example scene structure
{
"scenes": [
{
"id": "scene_1",
"concept": "Pythagorean Theorem",
"visual_intent": "Show right triangle with labeled sides",
"duration": 8,
"manim_hints": ["Polygon", "MathTex", "Animation"]
}
]
}Automated Manim Code Generation
- LLM generates Manim Python code for each scene
- Validates syntax and API compatibility
- Optimizes for rendering performance
Scene-Isolated Rendering
Each scene renders independently to prevent cascading failures:
Scene 1: ✓ Success
Scene 2: ✗ Failed → LLM Repair → ✓ Success
Scene 3: ✓ Success
Scene 4: ✗ Failed → LLM Repair → ✗ Failed → Skip
Scene 5: ✓ Success
Intelligent Error Repair
When rendering fails:
- Capture full error log
- Send code + error to LLM
- Generate corrected version
- Retry once
- Skip if still failing
- Per-Scene TTS: Each scene gets dedicated narration
- Perfect Alignment: FFmpeg ensures audio-video sync
- Adaptive Padding: Automatically adjusts for duration mismatches
- Quality Preservation: No re-encoding loss
# FFmpeg synchronization
ffmpeg -i video.mp4 -i audio.mp3 -c:v copy -c:a aac -shortest output.mp4Bloop integrates SadTalker for photorealistic talking avatars:
Side-by-Side Layout:
┌─────────────────────────┬─────────────────────────┐
│ │ │
│ Manim Animation │ Talking Avatar │
│ (Visual Concepts) │ (Narration) │
│ │ │
└─────────────────────────┴─────────────────────────┘
Benefits:
- No visual occlusion
- Preserved instructional clarity
- Human-like delivery presence
- Accessible learning experience
- Automatic extraction of key concepts from documents
- Spaced repetition algorithms for optimal learning
- Front/back card customization
- Export to Anki and other formats
- Multiple choice questions with adaptive difficulty
- True/false questions for quick validation
- Fill-in-the-blank for active recall
- Automatic grading and feedback
- Context-aware responses from uploaded documents
- Multi-document reasoning and synthesis
- Citation of specific document sections
- Follow-up question handling
Generate custom learning paths from any input:
# Example roadmap request
{
"input_type": "document", // or "topic"
"content": "Machine Learning syllabus PDF",
"level": "intermediate",
"duration": "3 months",
"preferences": {
"focus_areas": ["deep learning", "computer vision"],
"time_per_week": 10
}
}Roadmap Features:
- Structured learning milestones
- Prerequisite tracking and validation
- Progress-based adaptation
- Resource recommendations (videos, articles, books)
- Assessment checkpoints
- Skill level progression tracking
Input Flexibility:
- Upload any document (PDF, DOCX, TXT)
- Paste syllabus or course content
- Describe learning goals in natural language
- Specify target skill level and timeline
Solidify your learnings through three interactive games:
- Voice-to-voice conversational learning
- Explain concepts to the AI in your own words
- Real-time feedback and corrections
- Improves understanding through teaching (Feynman Technique)
- Tracks explanation quality and completeness
- Visual concept matching and categorization
- Interactive drag-and-drop interface
- Immediate feedback on correctness
- Reinforces spatial and logical reasoning
- Multiple difficulty levels
- Multiple-choice format with intentional errors
- Identify incorrect statements or misconceptions
- Critical thinking and concept validation
- Explanation of why options are correct/incorrect
- Adaptive difficulty based on performance
POST /api/v1/ask/video/generateRequest:
{
"query": "Explain the water cycle",
"options": {
"include_avatar": true,
"avatar_image": "base64_encoded_image",
"duration": "short",
"style": "educational"
}
}Response:
{
"video_id": "vid_abc123",
"status": "processing",
"estimated_time": 45,
"scenes_total": 5
}POST /api/v1/ask/flashcards/generateRequest:
{
"document_id": "doc_xyz789",
"num_cards": 20,
"difficulty": "medium"
}POST /api/v1/ask/quiz/generateRequest:
{
"document_id": "doc_xyz789",
"question_types": ["multiple_choice", "true_false"],
"num_questions": 10
}POST /api/v1/ask/queryRequest:
{
"document_ids": ["doc_1", "doc_2"],
"question": "What are the main themes?",
"include_citations": true
}POST /api/v1/plan/roadmap/generateRequest:
{
"input_type": "document",
"document_id": "doc_abc123",
"level": "beginner",
"duration_weeks": 12,
"goals": ["Master fundamentals", "Build projects"]
}Response:
{
"roadmap_id": "roadmap_xyz",
"total_milestones": 8,
"estimated_hours": 120,
"milestones": [...]
}GET /api/v1/plan/roadmap/{roadmap_id}/progressPOST /api/v1/play/teach-ai/startRequest:
{
"topic": "Photosynthesis",
"difficulty": "medium"
}POST /api/v1/play/drag-drop/submitPOST /api/v1/play/identify-error/answerGET /api/v1/video/status/{video_id}GET /api/v1/download/{content_id}const ws = new WebSocket('ws://localhost:8000/ws/{session_id}');
ws.onmessage = (event) => {
const update = JSON.parse(event.data);
console.log(`Status: ${update.status}, Progress: ${update.progress}%`);
};For complete API documentation, visit: http://localhost:8000/docs
| Component | Technology | Purpose |
|---|---|---|
| Backend Framework | FastAPI | High-performance async API |
| LLM Orchestration | GPT-4 / Azure OpenAI | Scene planning, Q&A, roadmap generation |
| Animation Engine | Manim Community | Programmatic video rendering |
| Text-to-Speech | Azure TTS / ElevenLabs | Neural narration synthesis |
| Video Processing | FFmpeg | Audio-video synchronization |
| Avatar Generation | SadTalker | Photorealistic talking avatars |
| Document Processing | PyPDF2, python-docx | Content extraction and analysis |
| WebSocket | FastAPI WebSockets | Real-time progress updates |
| Voice Processing | Whisper / Azure Speech | Voice-to-voice game mode |
ASK Mode:
| Content Type | Processing Time | Quality |
|---|---|---|
| Video (30s) | 15-30s | High |
| Video (1 min) | 45-90s | High |
| Video (2 min) | 2-3 min | High |
| Flashcards (20 cards) | 10-15s | High |
| Quiz (10 questions) | 15-20s | High |
| Document Q&A | < 5s | High |
PLAN Mode:
| Input Type | Roadmap Complexity | Processing Time |
|---|---|---|
| Document | Basic (4 weeks) | 20-30s |
| Document | Intermediate (12 weeks) | 40-60s |
| Topic description | Advanced (24 weeks) | 60-90s |
PLAY Mode:
| Game Type | Response Time | Mode |
|---|---|---|
| Teach the AI | Real-time | Voice-to-voice |
| Drag & Drop | Instant | Interactive |
| Identify Error | Instant | Multiple choice |
- API Response Time: < 500ms for planning
- Scene Rendering: 5-15s per scene (GPU-accelerated)
- Audio Generation: 2-5s per scene
- Avatar Rendering: 30-60s (depends on video length)
- Success Rate: 92% scenes render successfully
- Auto-Repair Success: 85% of failed scenes recovered
- Flashcard Generation: 500ms per card
- Quiz Question Generation: 1-2s per question
- Self-Paced Learning: Complete learning ecosystem from content to games
- Exam Preparation: Generate flashcards, quizzes, and video summaries
- STEM Education: Visual physics simulations and math explanations
- Language Learning: Grammar concepts with interactive exercises
- Employee Onboarding: Custom learning paths with progress tracking
- Technical Training: Video tutorials with knowledge validation games
- Compliance Training: Interactive quizzes and certification paths
- Product Knowledge: Visual demos with drag-and-drop exercises
- Paper Comprehension: Generate summaries, flashcards, and Q&A
- Concept Mastery: Video explanations with teaching reinforcement
- Study Planning: Roadmaps from research paper to full understanding
- Peer Teaching: Use "Teach the AI" to validate understanding
- Course Development: Rapidly generate videos, quizzes, and learning paths
- Student Engagement: Interactive games for knowledge reinforcement
- Tutorial Creation: Animated explanations with assessment tools
- Educational Content: Multi-format content from single source
# Video generation settings
VIDEO_CONFIG = {
"max_scenes": 15,
"min_duration_per_scene": 5,
"max_duration_per_scene": 20,
"complexity_level": "medium",
"animation_style": "clean",
"math_rendering": "latex",
"color_scheme": "default"
}
# Flashcard generation settings
FLASHCARD_CONFIG = {
"difficulty_distribution": {
"easy": 0.3,
"medium": 0.5,
"hard": 0.2
},
"include_images": True,
"spaced_repetition": True
}
# Quiz generation settings
QUIZ_CONFIG = {
"question_types": {
"multiple_choice": 0.6,
"true_false": 0.2,
"fill_blank": 0.2
},
"adaptive_difficulty": True,
"explanation_included": True
}# Roadmap generation settings
ROADMAP_CONFIG = {
"milestone_granularity": "weekly",
"include_assessments": True,
"resource_types": ["video", "article", "practice"],
"prerequisite_validation": True,
"adaptive_pacing": True
}# Game settings
GAME_CONFIG = {
"teach_ai": {
"voice_provider": "azure",
"feedback_style": "encouraging",
"difficulty_progression": True
},
"drag_drop": {
"time_limit": 60,
"hint_system": True,
"visual_feedback": True
},
"identify_error": {
"explanation_depth": "detailed",
"hint_after_wrong": True,
"max_attempts": 3
}
}# Avatar generation settings
AVATAR_CONFIG = {
"expression": "neutral",
"pose": "centered",
"background": "blur",
"lighting": "natural",
"video_quality": "high"
}# Rendering failure handling
ERROR_POLICY = {
"retry_count": 1,
"skip_on_failure": True,
"log_errors": True,
"fallback_to_text": False,
"partial_output": True
}ASK MODE - Comprehensive Content Understanding
- Multiple learning formats (videos, flashcards, quizzes)
- Caters to different learning styles
- On-demand clarification through Q&A
- Visual + auditory + textual learning
PLAN MODE - Structured Progress Tracking
- Clear learning objectives and milestones
- Prevents overwhelm with structured paths
- Adaptive pacing based on progress
- Resource optimization and time management
PLAY MODE - Active Knowledge Reinforcement
- Gamification increases engagement
- Multiple reinforcement strategies
- Immediate feedback loops
- Confidence building through practice
- 25% Faster: Concept comprehension with multi-modal approach
- 40% Better: Retention with video + flashcards + games
- 3x Engagement: Compared to traditional text-based learning
- Scalability: Complete learning journeys in minutes
- 85% Completion: Higher course completion with roadmaps
- Visual Learners: Dynamic animations and interactive games
- Auditory Learners: Clear narration and voice interactions
- Kinesthetic Learners: Drag-and-drop and interactive elements
- Multilingual Support: TTS in 40+ languages
- Self-Paced: Custom roadmaps adapt to individual pace
This project is licensed under the MIT License - see the LICENSE file for details.
- Documentation: docs.bloop.ai
- GitHub Issues: Report bugs
- Discussions: Community forum
- Email: poorvikagrover@gmail.com
Q: How long does it take to generate content?
A: Video generation takes 45-90 seconds for a 2-minute video. Flashcards and quizzes generate in 10-20 seconds. Roadmaps take around 20 seconds depending on complexity.
Q: Can I use my own avatar image?
A: Yes! Upload any portrait image and Bloop will animate it with synchronized lip movements.
Q: What's the difference between the three modes?
A: ASK mode helps you understand content (videos, flashcards, Q&A). PLAN mode creates learning roadmaps. PLAY mode reinforces learning through games.
Q: Can I customize the learning roadmap?
A: Yes, you can specify your level, duration, focus areas, and time commitment. The roadmap adapts to your progress.
Q: Do the games track my progress?
A: Yes, all three games track performance, provide feedback, and adapt difficulty based on your responses.
Q: What document formats are supported?
A: PDF, DOCX, TXT, and plain text input are all supported for content generation.
- OpenAI for GPT-4 and AI capabilities
- Azure for TTS and cloud infrastructure
- Manim Community for animation framework
- SadTalker Team for avatar technology
- All Contributors who made this possible