Topic in, explainer video out—every frame drawn in React code

What it does

anything2explainer is a skill for Claude Code and OpenAI Codex that produces narrated explainer videos on any topic. You give it a subject and a target length; it researches, writes a script, generates TTS voiceover, storyboards every shot, then dispatches parallel build agents to write Remotion components. The result is a 1280×720 H.264 MP4 with synchronized subtitles, chapter cards, and a progress bar, plus a full paper trail of research docs, storyboards, and QC reports.

The interesting bit

Every frame is drawn in code via Remotion rather than synthesized by a generative video model, which means the output is deterministic, fully editable, and contains no stock footage. The repo is essentially a complete micro-studio: it ships style guides, motion vocabulary formulas, multi-agent build protocols, quantitative QC rubrics, and a reference film that serves as the quality bar.

Key highlights

• Deterministic React/TypeScript rendering with seeded randomness and computed text fitting
• End-to-end agent pipeline: research → narration → storyboard → parallel shot building → render → QC
• Chinese or English output with per-word TTS alignment and language-specific pacing models
• Four human checkpoints to approve length, narration, voice engine, and the first 30 seconds
• CPU-only rendering through headless Chromium; no GPU required

Caveats

• PolyForm Noncommercial license; commercial use requires prior author authorization
• Locked to a single visual style (black canvas, white line art, purple accents) with only two backdrop choices
• Landscape 1280×720 only; no vertical video support
• Developed and verified on macOS; Windows is untested

Verdict

Worth exploring if you want programmatic, revision-friendly explainer videos and already live inside Claude Code or Codex. Look elsewhere if you need commercial licensing by default, creative visual flexibility, or 9:16 vertical output.