🎬 Mythos Director: Where AI Becomes the Creative Director

🧐 The Pitch

Creating high-quality multimedia content usually requires switching between multiple tools and hours of manual labor. Writers struggle to visualize scenes, and filmmakers spend days on storyboarding. We needed a "One-Stop Production Studio" that builds the entire cinematic experience instantly. Mythos Director is that studio.

✨ The Solution

Mythos Director is an AI-powered production agent. Instead of a simple chatbot, it acts as a Creative Director:

• [cite_start]It breaks down a story prompt into a cinematic script[cite: 12].
• [cite_start]It determines exactly where to place illustrations and video clips to enhance the narrative[cite: 13].
• [cite_start]It produces human-like narration synchronized with the flow of the story[cite: 14].
• [cite_start]It streams everything live, creating a "Theater" experience in the browser[cite: 15].

🛠️ How We Built It

Core Models:

• [cite_start]Gemini 2.5 Flash (Vertex AI): Our "Creative Director" using Structured Output for media placement[cite: 18].
• [cite_start]Imagen 3.0 (Vertex AI): Generates 16:9 cinematic illustrations[cite: 19].
• [cite_start]Veo (Vertex AI): Generates cinematic video clips for dynamic action sequences[cite: 20].
• [cite_start]Cloud Text-to-Speech (Neural2-F): Provides warm, human-like narration via SSML[cite: 21].

Backend/Architecture:

• [cite_start]FastAPI (Python): Handles orchestration and Server-Sent Events (SSE) for live streaming[cite: 24].
• [cite_start]Google Cloud Platform: The backend is containerized via Docker for deployment on Cloud Run[cite: 25].

💡 Findings & Learnings

1. [cite_start]Prompt Design: Asking Gemini to act as a "Creative Director" led to much more artistic prompts for Imagen and Veo[cite: 51].
2. [cite_start]Streaming Efficiency: SSE improved UX by letting users read while videos "render" in the background[cite: 53].
3. [cite_start]Synchronization: We solved audio-visual timing using a specialized queue in the JavaScript frontend[cite: 54].

🚀 Proof of Cloud Deployment

The project exclusively uses Vertex AI endpoints. Implementation details:

• [cite_start]ai_gemini.py (Lines 47-51: Vertex AI Client initialization) [cite: 57]
• [cite_start]ai_gemini.py (Line 163: Gemini 2.5 Flash implementation) [cite: 58]
• [cite_start]ai_gemini.py (Line 61: Imagen 3.0 implementation) [cite: 59]
• [cite_start]ai_gemini.py (Line 97: Veo implementation) [cite: 60]
• [cite_start]ai_gemini.py (Line 237: Cloud Text-to-Speech client) [cite: 61]

Built With

• 2.5
• 3
• ai
• cloud
• fastapi
• flash
• gemini
• google
• imagen
• python
• text-to-speech
• veo
• vertex