Vincentwei1021/anything2explainer

What it is

anything2explainer is a Claude Code / Codex skill that takes a topic (or an article you supply) plus a length and language, and produces a 1280×720 H.264 explainer video with TTS voiceover, word-aligned subtitles, chapter cards and a bottom progress bar. All frames are React/TypeScript components rendered with Remotion, so there is no stock footage or generative video model; visuals are white line art plus purple accents on a black canvas with either a star-field or dot-field backdrop source:2.

The repository is not a CLI. It ships the method an agent follows: a compilable Remotion template, a primitives and lighting library, voiceover/storyboard/render/QC scripts, written style and motion specs, a multi-agent build protocol, and a reference film under examples/rag/ with its research, narration, storyboards, shot source and QC reports source:31.

Expect roughly 1–3 hours of wall-clock time depending on video length, most of it spent by parallel build agents writing one Remotion component per shot source:16.

Quick start

Prerequisites: Node ≥18, ffmpeg, Python 3, and either Claude Code or Codex. Clone the repo and symlink it into the agent's skills directory:

git clone https://github.com/Vincentwei1021/anything2explainer.git
ln -s "$PWD/anything2explainer" ~/.claude/skills/anything2explainer   # Claude Code
ln -s "$PWD/anything2explainer" ~/.codex/skills/anything2explainer    # Codex

Then install the Python tooling in a virtualenv (the pinned edge-tts version matters because it targets a Microsoft endpoint that changes across upgrades) source:18:

python3 -m venv ~/.venvs/a2e && source ~/.venvs/a2e/bin/activate
pip install 'edge-tts==7.2.8' numpy pillow scipy
pip install kokoro soundfile && brew install espeak-ng   # English narration only

In Claude Code or Codex, ask for the video in natural language, e.g. "Make me an explainer video about vector databases." The skill runs nine stages: scaffold, research, narration and TTS timeline, storyboard, overlays and icons, a pilot shot group with a 30-second cut for review, parallel shot building, full render with frame metrics, then QC and fixes source:25.

If you prefer to drive the template yourself, create a project and work through the numbered steps:

template/scripts/new_project.sh ~/work/my-video myslug
cd ~/work/my-video
python3 scripts/tts_build.py
python3 scripts/render_storyboard.py
scripts/preview.sh 30
VER=v1 scripts/render.sh
python3 scripts/frame_metrics.py

The README notes that the working directory contents (research doc, script/narration.txt, storyboard source, src/config.ts) must be filled in at the corresponding steps source:26.

Checkpoints and control

The run pauses at four points: choosing length and language before the script, narration sign-off before voiceover, TTS engine choice before synthesis, and reviewing the first 30 seconds before the remaining shots are built source:27. Choose length carefully: it fixes line/shot count and how many agents are used, and changing a word after voicing re-times the whole film because shot code hard-codes frame numbers.

Defaults are edge-tts zh-CN-YunxiNeural for Chinese and kokoro-82m am_liam for English. You can bring your own TTS or finished audio by placing it at public/assets/<slug>/audio.wav and filling in the timeline and subtitle files source:29.

Constraints

• Platform: shell scripts are zsh + Python 3 and were developed and verified on macOS; Linux is expected to work, Windows is untested source:21. Raspberry Pi 5 (ARM64) is verified but needs system Chromium via REMOTION_BROWSER_EXECUTABLE source:22.
• No GPU is required; rendering runs through headless Chromium on CPU.
• Output is landscape 1280×720 only; there is no vertical 9:16 mode, and only Chinese and English are supported, each with its own pacing and subtitle budget source:39.
• One visual style with two backdrops (bg: 'stars' | 'dots'); changing more means editing the style guide and src/ui.tsx.
• Parallel builds need at least 5 GB free disk, and tmux panes are capped around 12, so larger films dispatch in waves.
• The toolkit is PolyForm Noncommercial 1.0.0: noncommercial use is free, commercial use needs prior authorization, and videos you make are yours source:35. Bundled fonts are SIL OFL 1.1 source:37, and Remotion has its own company license terms source:38.

HotRepo's take

There are no HotRepo ranking records, star-curve samples or Hacker News threads for this repository, so there is no momentum or discussion history to assess. The notable technical fact from the materials is the split between a fixed style system and generated content: every animation is a pure function of the frame number with seeded randomness, so re-renders are identical, but the fixed frame numbers mean any late narration edit cascades through every shot source:29. That makes the four checkpoints the real control surface, and the first 30-second review the point where style changes are cheapest.