Programmatic video with Remotion
A series of experiments with pipelines that generate video from code: from auto-producing script, visuals, and voiceover from a single input, to motion graphics; scalable, repeatable video systems aimed at YouTube automation.
How it works
Sample output
What I did
I ran a series of experiments on generating video from code. The goal was to build content production not by hand but in a repeatable, scalable way, especially for short-form and YouTube automation. I tried different approaches separately and kept each in its own repo.
Short-form video template
A fully-loaded viral video template for TikTok and Instagram: it generates the script and visuals with OpenAI and the voiceover with ElevenLabs, and ships with a hook intro, B-roll layers, sentence-by-sentence subtitles, scene transitions, a CTA card, and sound effects. It even uses MediaPipe vision to auto-frame the shots.
End-to-end production pipeline
The most complete experiment: an end-to-end system that turns a single prompt into a markdown video plan, then generates AI visuals (Kie.ai) and voiceover with ElevenLabs and renders it with Remotion. It auto-adds continuation scenes when audio runs past 6 seconds, stamps dynamic elements like lower thirds and counters, and can batch several videos at once.
Motion graphics & kinetic typography
On the visual side I experimented with motion graphics using Remotion's shapes, paths, and noise: kinetic typography (animated quote scenes) and data-driven themed videos (like an 'AI Wars' narrative comparing AI companies). Clean, animated visuals produced from code, over and over.