MoFA Effect

AI-native video generation using DaVinci Resolve Fusion templates.

Enter a text prompt, and MoFA Effect selects matching motion graphics templates from a curated library, generates background images, writes text content, synthesizes a voiceover, patches everything together, and hands off to DaVinci Resolve for cinematic rendering.

---

How It Works

MoFA Effect is an 8-step AI pipeline that turns a text prompt into a rendered video with professional Fusion templates.

The Pipeline

Step	What Happens	Technology
1. Prompt	User describes the video they want	Interactive CLI
2. Template Selection	AI picks 2-6 matching templates based on mood, energy, and visual style	GPT-4o / LLM
3. Content Generation	AI generates text content for each template's text elements	GPT-4o / LLM
4. Background Images	AI generates topic-specific background images	DALL-E 3 / HuggingFace / Pollinations
5. Voiceover	Text-to-speech narration synthesized from AI-written script	Edge TTS
6. Template Patching	Two-phase patching: LLM edits text/fonts, programmatic injects backgrounds	GPT-4o + Regex
7. Resolve Render	DaVinci Resolve renders each Fusion template as a video segment	DaVinci Resolve
8. Final Assembly	FFmpeg concatenates segments and merges voiceover audio	FFmpeg

---

Architecture

mofa-effect/
|-- main.py                    # Entry point - orchestrates the full pipeline
|-- config.py                  # Central configuration (loads .env)
|-- interactive_review.py      # Pre-render content review UI
|
|-- intelligence/              # AI-powered template intelligence
|   |-- _llm_client.py        #   Centralized LLM API client (OpenAI/Groq)
|   |-- selector.py           #   AI template selection + content generation
|   |-- llm_patcher.py        #   Two-phase template patching (LLM + programmatic)
|   |-- universal_patcher.py  #   Programmatic .comp file transforms
|   |-- comp_curator.py       #   One-time library curation (613 -> 25 templates)
|   |-- profiler.py           #   Structural analysis of .comp files
|
|-- providers/                 # Swappable service providers
|   |-- image/                #   OpenAI DALL-E 3, HuggingFace, Pollinations
|   |-- tts/                  #   Edge TTS
|   |-- llm/                  #   Groq (for provider registry)
|   |-- renderer/             #   DaVinci Resolve
|   |-- registry.py           #   Provider registration and factory
|
|-- generator/                 # Asset generation services
|   |-- image_generator.py    #   Image generation with retry/fallback
|   |-- placeholder.py        #   Placeholder image generator
|
|-- renderer/                  # DaVinci Resolve integration
|   |-- resolve_bridge.py     #   Handoff JSON + script installation
|   |-- mofa_render.py        #   Runs inside Resolve (Workspace > Scripts)
|
|-- comp_usable/               # Curated template library (25 verified .comp files)
|   |-- manifest.json         #   Template metadata index
|
|-- utils/                     # Shared utilities
    |-- helpers.py            #   File reading, encoding fallbacks
    |-- logger.py             #   Logging configuration

---

Template Patching System

The core innovation of MoFA Effect is its two-phase template patching system that modifies DaVinci Resolve Fusion .comp files.

Phase 1: LLM-Native (Creative)

• Sends the .comp Tools section to GPT-4o
• LLM edits text content, adjusts fonts, modifies styles
• Validation: brace balance, tool count preservation, size sanity
• Falls back to regex if validation fails

Phase 2: Programmatic (Structural)

• Background image injection behind template output
• Ken Burns zoom animation on background
• Text overlay injection for visual-only templates
• Render range clamping to target duration
• Anti-aliasing enforcement on 3D renderers

---

Template Library

MoFA Effect ships with 25 curated Fusion .comp templates across 13 unique visual families.

Family	Templates	Type	Best For
CH16 3D Text	6 variants	3D extruded text	Titles, headings
CH17 3D Titling	8 variants	Animated 3D text paths	Dynamic intros, reveals
afg poster	1	3D poster layout	Announcements
neural networks	1	Node visualization	Tech, science topics
Scoreboard	1	Sports overlay	Sports, competitions
Particles (stx)	2	3D particle effects	Energy, action scenes
ObjectsFromParticles	1	Particle generation	Abstract, creative
se LEDWall	1	LED wall simulation	Events, concerts
UT Water	1	Water simulation	Nature, calm scenes
flag	1	Flag animation	National, patriotic
BigNoodle	1	Stylized text	Clean text overlays
playerjoinedtext	1	Text animation	Gaming, esports

Adding Custom Templates

1. Place your .comp file in the Comp_Library/ directory
2. Run the curator: python -m intelligence.comp_curator
3. Verified templates are copied to comp_usable/ with metadata in manifest.json

---

Output Organization

Each run produces an isolated timestamped directory:

output/
  mofa_20260402_183033/
    comps/      # Patched .comp files
    images/     # AI-generated background images  
    audio/      # Voiceover WAV
    final/      # Rendered MP4 video

---

Setup

Prerequisites

• DaVinci Resolve (Free or Studio) — for template rendering
• FFmpeg — for video concatenation and audio merge
• Python 3.10+

Installation

git clone https://github.com/YOUR_USERNAME/mofa-effect.git
cd mofa-effect
pip install -r requirements.txt
cp .env.example .env
# Edit .env with your API keys

Configuration

Edit .env to set your providers:

# LLM (for template selection + content generation)
LLM_PROVIDER=groq
LLM_API_KEY=your-groq-key
LLM_MODEL=llama-3.3-70b-versatile

# Image generation (pick one)
IMAGE_PROVIDER=openai          # Best quality (paid)
OPENAI_API_KEY=your-openai-key

# IMAGE_PROVIDER=pollinations   # Free, no key needed
# IMAGE_PROVIDER=huggingface    # Free tier
# HF_API_KEY=your-hf-key

# DaVinci Resolve path
RESOLVE_EXE=C:\Program Files\Blackmagic Design\DaVinci Resolve\Resolve.exe

---

Usage

python main.py

The interactive CLI will guide you through:

1. Video topic — describe what you want
2. Duration — 15, 30, 45, or 60 seconds
3. Orientation — landscape (16:9) or portrait (9:16)
4. Asset reuse — optionally reuse images/audio from a previous run
5. Content review — edit AI-generated text before rendering

After the pipeline completes, open DaVinci Resolve and run:

Workspace > Scripts > MoFA_Render

The render script automatically:

1. Creates a project with separate timelines per template
2. Renders each template segment
3. Concatenates all segments
4. Merges the voiceover audio
5. Saves the final video and cleans up intermediate files

---

Provider System

MoFA Effect uses a pluggable provider architecture. Swap any service by changing one line in .env.

Provider Registry
|
|-- LLM Providers
|   |-- groq (default) -- Llama 3.3 70B, free tier
|
|-- Image Providers
|   |-- pollinations (default) -- Free, no API key
|   |-- huggingface -- FLUX.1-schnell, free tier
|   |-- openai -- DALL-E 3, highest quality
|
|-- TTS Providers
|   |-- edge_tts (default) -- Microsoft Edge TTS, free
|
|-- Renderer Providers
    |-- resolve (default) -- DaVinci Resolve Fusion

Adding a Custom Provider

1. Create a class implementing the interface in providers/base.py
2. Register it in providers/registry.py
3. Set the provider name in .env

---

License

MIT