Local AI video · Public case study

OmniReel AI

A local-first AI video pipeline that turns a short topic into a narrated educational video through local planning, media generation, offline speech and a Rust compositor.

Orchestration
Python
Media compositor
Rust and FFmpeg
Generation backend
Local ComfyUI
Privacy posture
Local-first and loopback-only

Why this system exists.

Many AI video workflows depend on paid cloud APIs, send prompts or generated media to external services and hide the engineering between a topic and a finished video. OmniReel demonstrates that complete orchestration can be made inspectable and local-first.

Topic → local lesson and scene plan → base image → offline narration → procedural or ComfyUI video backend → subtitle timing → Rust subtitle burn and audio mux → final MP4

What the public implementation demonstrates.

  • Supports deterministic procedural QA, local ComfyUI video generation and an external local animation hook.
  • Uses offline text-to-speech with a pyttsx3 implementation and a Piper integration point.
  • Restricts the ComfyUI service to loopback URLs rather than exposing a remote generation endpoint.
  • Combines Python orchestration with Rust-based decoding, encoding, subtitle rendering and audio muxing.
  • Provides reproducible command-line workflows for fast QA and real local generation.

What I designed and built.

  • Pipeline architecture and backend-switching strategy.
  • Python orchestration and local model integration.
  • ComfyUI workflow automation and safety-aware prompt wrapping.
  • Python-to-Rust interface design and FFmpeg media processing.
  • Offline narration, subtitle timing, QA workflows and technical documentation.

Reliability, safety and responsible use.

Engineering decisions

  • No hosted LLM API or cloud video API is required for the local workflow.
  • Generated media and model weights remain outside version control.
  • A procedural backend enables fast deterministic validation before expensive generation is attempted.

Limitations

  • CPU-only local video generation is slow and requires reduced resolution and duration for practical tests.
  • The repository is a portfolio-grade AI systems project rather than a polished consumer application.
  • The educational prompt wrapper is a technical safeguard, not a certified child-safety system.

The implementation stack.

Public and inspectable. The repository contains the implementation, documentation and setup guidance available for this project.
PythonRustComfyUIFFmpegOffline TTSLocal AICLI Orchestration