DirectorGPT

• 

  Before generating

• 

  After generating

Inspiration

Generative AI video often lacks coherence—characters morph between scenes, pacing is erratic, and audio remains an afterthought. We built DirectorGPT to simulate an actual film production crew where specialized AI agents collaborate, critique, and align on visual style, narrative beats, and audio design before the cameras roll.

What it does

DirectorGPT takes a high-level story prompt and runs an autonomous end-to-end studio pipeline:

• Director Agent: Coordinates the crew, enforces pacing, and guides the production vision.
• Screenwriter Agent: Breaks narratives into three-act scenes with precise 5-second camera shots, angles, and movement cues.
• Casting Agent: Generates persistent visual anchors and character identity tokens to maintain facial and wardrobe consistency across cuts.
• Sound Designer Agent: Produces timestamped musical scores and ambient foley cues mapped to emotional arcs.
• Editor Agent: Compiles the complete Edit Decision List (EDL) with transitions, speed adjustments, and FFmpeg assembly commands.

How we built it

• Backend: Python, FastAPI, and Pydantic models orchestrated with autonomous multi-agent loops.
• Agent Intelligence: Google Gemini API for high-speed script breakdowns, critique passes, and prompt conditioning.
• Assembly & Video: Automated FFmpeg pipeline for shot stitching, aspect ratio normalization, and audio-video synchronization.
• Deployment: Render (Backend API) and GitHub Pages (Frontend studio UI).

Challenges we faced

• Character Continuity: Preventing visual drift across sequential shot prompts. We solved this by having the Casting Agent inject precise reference-sheet descriptor tokens into every prompt.
• Pacing & Audio Sync: Aligning emotional tone shifts with musical timestamps across variable scene lengths.

Accomplishments that we're proud of

• Delivering a functional multi-agent film studio that turns a one-sentence logline into a fully realized production package with shot lists, sound cues, and edit decisions in seconds.
• Building a decoupled architecture with a live interactive web UI backed by a production-grade FastAPI service.

What's next for DirectorGPT

• Direct integration with text-to-video APIs (e.g., Runway, Luma) for automated cloud rendering.
• Real-time director feedback chat allowing users to request shot revisions dynamically.

Built With

• agents
• css3
• fastapi
• ffmpeg
• github
• google-gemini
• html5
• javascript
• python
• render
• rest-api