Agentic Video Pipeline is an AI-assisted video production and editing workflow designed to orchestrate coding agents (Antigravity, Claude Code, GitHub Copilot) in creating high-precision animations, motion graphics, and video montages across various topics.
Rather than relying on unconstrained generative video models, the pipeline approaches content creation as a disciplined engineering process: decomposing videos into deterministic modular scenes, visually validating keyframes via contact sheets, synthesizing mathematical audio effects, and orchestrating montage timelines with automated transitions and captions.
Multi-Engine Deterministic Rendering
- HTML5 / SVG / Canvas: Zero-dependency headless browser rendering engine driven directly through Chrome DevTools Protocol (CDP), capturing frames deterministically into FFmpeg pipes with custom compositor-suspension handling.
- Remotion: React & TypeScript motion graphics and kinetic typography.
- Python Manim: Algorithmic and mathematical diagram animations for technical topics.
- Unity Engine Integration: C# scene recorder for 3D physics and engine mechanics capture.
Visual QA & Agent Feedback Loop
The system incorporates an automated quality-assurance feedback loop. Agents generate labeled 4x3 contact-sheet PNG grids from rendered clips to visually inspect layout balance, contrast, mobile safe zones, and easing curves before approving the cut for final assembly.
Procedural SFX & Timeline Montage
- Procedural Sound Design: Synthesizes zero-asset UI audio effects (whooshes, pops, clicks, harmonic chimes) directly through mathematical waveform evaluation and FFmpeg audio filter graphs.
- Declarative Timeline Orchestrator: Assembles scenes using declarative transition manifests with crossfades, background music ducking, and kinetic caption synchronization.
Published Content
Videos created and edited with this pipeline are published across the
@keroen-dev channels: