Nasl3yn — Autonomous Cinematic AI Agent

A sovereign agent that turns a text brief into a film: Gemini plans scenes, ffmpeg renders. The machine calculates, the human decides.

Comment

Inspiration

What it does

How we built it

Challenges we ran into

Accomplishments that we're proud of

What we learned

What's next for Nasl3yn — Autonomous Cinematic AI Agent

All Things Agentic — Project Details (About)

Inspiration

Climate and civilization knowledge is trapped in static reports. Communities engage with video, not PDFs. We wanted an agent that takes a plain-text brief — "make a film about sovereign AI and the environment" — and autonomously produces a finished documentary, with a human approving each creative decision. The axiom: "The machine calculates, the human decides."

What it does

Nasl3yn is an autonomous cinematic AI agent. Given a source-text brief it:

  1. Researches & plans — breaks the brief into scenes using a Gemini (Vertex AI) orchestrator.
  2. Generates script & VO — writes a cinematic script and narration.
  3. Renders film — assembles scenes with ffmpeg into a deliverable mp4.
  4. Local-first / sovereign — runs on Google Cloud Run; no external creative dependency, full asset lineage.

It is exactly the hackathon's "Taskmaster" track: a goal in, a multi-step workflow executed autonomously, while you do something else.

How we built it

  • Gemini (Vertex AI, us-central1) via the Google GenAI SDK for planning/script generation.
  • Google Cloud Run — nasl3yn-api deployed and serving live traffic.
  • Firestore for project/state persistence.
  • ffmpeg for deterministic, lineage-tracked rendering.
  • Local-first QSR runtime for sovereign execution.

Challenges we ran into

  • Keeping the agent's creative output coherent across a long film (scene-to-scene continuity).
  • Rendering reliably without cloud video APIs (blocked) — solved with a local ffmpeg pipeline.
  • Arabic/English bilingual overlays with correct RTL/LTR layout.

Accomplishments that we're proud of

  • A working autonomous agent that produces a 5+ minute documentary from a one-line brief.
  • Verified deployment on Google Cloud Run (Gemini 3.5 + Vertex AI + Firestore).
  • A reusable sovereign pipeline reusable for any civic/environmental storytelling.

What we learned

Agentic pipelines shine at planning and orchestration; the human's job is curation, not production. "The machine calculates, the human decides."

What's next for Nasl3yn

  • Multi-language narration, interactive/web delivery, and a public creative-director API on Cloud Run.

Built With

Share this project:

Updates

Submission history