Communitygithub.com

explainroo

Make an explainer video (MP4) with a voice-over using explainroo. The voice is made on this computer and no API key is needed. Use when the user asks for an explainer, how-to, walkthrough, launch or tutorial video, an animated diagram with a voice-over, or a short for YouTube, Reels or TikTok.

explainroo 是什么?

explainroo is a Claude Code agent skill that make an explainer video (MP4) with a voice-over using explainroo. The voice is made on this computer and no API key is needed. Use when the user asks for an explainer, how-to, walkthrough, launch or tutorial video, an animated diagram with a voice-over, or a short for YouTube, Reels or TikTok.

兼容平台~Claude Code~Codex CLI~Cursor
npx skills add https://github.com/maksiam07/siam-post-production/tree/main/skills/explainroo

在你喜欢的 AI 中提问

打开一个已预加载此 Agent Skill 的新对话。

文档

explainroo for coding agents

Run explainroo with npx --yes explainroo. --yes stops npx from waiting on its install question. In a source checkout, run node bin/explainroo.js instead. The command examples below shorten this to plain explainroo, so type the full form when you run them. npx does not install an explainroo command on PATH.

This page tells you, the coding agent, how to make a video with explainroo. The user gives you a topic, maybe with a length or a style. You write the words and the pictures, and explainroo makes the video. A long docs topic has sections, and explainroo docs <topic> --sections lists them. Print only the one you need. When the user wants their own recording edited instead, read explainroo docs cut.

What explainroo does

explainroo turns two files into an MP4 video with narration. script.md holds what the voice says. scenes.js draws what the viewer sees. Every drawing can appear on a word the voice says. A voice model (Kokoro) reads the script on this computer, or the narration comes from ElevenLabs, Cartesia or the user's own recording. Whisper notes when each word is spoken. Chrome draws the frames, and ffmpeg makes the MP4 with music and sound effects. All of this runs on this computer, so a video needs no account and no key. The documentation for people is at https://www.explainroo.com/docs/.

The hard limits

Check these first, and tell the user before you make anything that runs into one.

Languages. Set language in video.json to one of these codes. No other language works yet, with any voice.

LanguageCodeKokoro, the free voice, also needs
Englishennothing more
Spanish, French, Italian, Brazilian Portuguese, Hindies, fr, it, pt-br, hithe espeak-ng program
Japanese, Mandarin Chineseja, zhPython 3 with misaki
GermandeKokoro has no German voice

ElevenLabs, Cartesia and a recording of the user's voice work in all of these languages, and they need neither program. German works only with one of these three. explainroo docs languages has the details.

Keys, money and data. These services need a key or a sign-in. They run on the user's own account, and only when you run a command that uses them.

ServiceUsed forNeedsCosts moneyWhat it gets
OpenRouterAI images (image) and AI clips (clip)one key for both, from connect openrouteryes, per image and per second of clipthe prompt and any reference images, also for the company that runs the model
ElevenLabs, Cartesiaa paid voicea keyyes, per characterthe narration text, when narration is made
Blotatoposting with publisha keyyes, creating the key starts its paid planthe video, its title and its text
A presenter service, none yetAI presenters (presenter)a keyyesthe narration of a scene
YouTubeposting with publisha sign-in with the user's own Google clientnothe video, its title and its description
Bufferposting with publisha sign-inno, on Buffer's free planthe title, the text and a link to the video

connect, voices --provider and publish --dry-run send only the key or the sign-in to that service. When a post uses a link to the video, publish and its dry run also ask that link for the size of its file. clip --dry-run reads OpenRouter's public price list, without the prompt or a key. services reads the same list when it lists clips, at most once a day. share will upload the video and its title to explainroo.com, and sends nothing until that service runs. capture loads the page you name, and formats add clones the Git address you name. The models and fonts download the first time, and doctor --fetch may download espeak-ng's packages. Nothing of the user's goes along. Nothing else leaves the computer, and everything else is free.

Before you start

You only need to clone explainroo to change explainroo itself. It needs Node.js 20.11 or newer, ffmpeg, and Chrome or Chromium. Run this once:

npx --yes explainroo doctor --fetch

doctor checks Node, ffmpeg and Chrome, and the extra programs some languages need. With --fetch it also downloads the speech models that setup needs, the first time only: about 400 MB for English, and 252 MB for another language. Each part is ready, missing with how to get it, or not needed for this setup. doctor videos/<name> checks what that project's languages and voices need, and before the project exists, doctor --language es checks one language. On Debian and Ubuntu, doctor --fetch --language es also sets up espeak-ng without root when it is missing. Commands that work on a video take the project folder first, and every command accepts --json.

npx --yes explainroo docs prints this page, and npx --yes explainroo docs <topic> prints a topic listed at the end.

explainroo connect <service> sets up a key or a sign-in. It tests it and saves it in the settings folder, so every project can use it. Never print a key or put it in a script, a commit or a log. explainroo docs settings says where keys go and how explainroo finds them.

When to ask the user

Decide the normal things yourself, like the look, the length, the voice, the layout and the wording. Ask only when you cannot go on without an answer. When you ask, put everything you need in one message. Typical reasons to ask:

  • The topic is unclear, or you need a fact you cannot check.
  • The user chose AI images, clips, a paid voice or a post and has no key or login for it. Run explainroo connect openrouter, elevenlabs, cartesia, youtube, buffer or blotato, and show the user what it prints.
  • The user wants something explainroo cannot do, like video shot with a camera.

Explainroo makes the whole video for free. When the user asks for something that can use an outside service, run explainroo services <step> --json and tell the user the options, free ones first. Say which cost money and which are paid partners. Let the user choose. Never pick a paid service for them and never call a partner the best. The steps are voice, timing, images, clips, presenters, music, sharing and posting. publish and share work as names for the last two.

When the user has already picked a service or given a budget, use it and don't ask again. A budget for AI clips goes in video.json as "clipBudget": 0.40. clip then stops before a clip that would go over it, and it says how much is left. clip checks the price against the cap from --max-cost or the budget before it buys. The final cost is known only after the job, and it can come out a little higher.

share and publish are the only commands that upload a video, so run them only when the user asks. Run publish with --dry-run first and show the user the plan. Add --yes only after the user said yes. For a look at the post in Buffer before it goes out, add --draft: it saves a draft there and publishes nothing.

From a brief to a finished video

  1. Learn the subject. Read what the user points to, and write down the facts you will use. Never make up numbers, quotes or claims.
  2. Create the project with explainroo init videos/<name> --theme studio --size 16:9 --title "...". Pick the look for the audience, and the size for the platform (explainroo docs sizes). init --format <name> starts from a format, like vertical-short (explainroo docs formats). The video has a small "explainroo.com" watermark in a corner by default, and "watermark": false in video.json turns it off. With "captions": "auto", vertical, 4:5 and square videos get captions, and a 16:9 video gets none.
  3. Write script.md. The narration comes first, because it sets the timing for everything else (explainroo docs writing). When the user asks for a length, set it in video.json, like "targetSeconds": 40. After voice, a shorter video gets longer holds at scene ends to reach it, and voice and check warn when it is over by more than 0.5 s. explainroo set videos/<name> targetSeconds 40 writes one setting, and a dotted key like sizes.shorts.pace sets one of a single size.
  4. Make the voice with explainroo voice videos/<name>. Fix a word the speech check did not confirm with {shown|spoken}, and run it again (explainroo docs writing).
  5. Plan the pictures. Decide what appears on each cue, what leaves and what stays in place (explainroo docs pictures). explainroo words videos/<name> lists each scene's spoken words and times.
  6. Make images, if the video needs them and the user agrees (explainroo docs images).
  7. Write scenes.js, one function per scene, with each picture tied to a spoken word or a [#marker] (explainroo docs scene-api).
  8. Check with explainroo check videos/<name>, and fix every error and every warning (explainroo docs troubleshooting).
  9. Look at the frames with explainroo still videos/<name> and explainroo sheet videos/<name>. Open the images and judge them like a viewer would. For a platform's version, add --platform tiktok to check, still, sheet, render and verify.
  10. Make the video with explainroo render videos/<name>. Make a draft with --draft only when you will watch it, because still and sheet are the cheap way to see frames. Check the file with explainroo verify videos/<name>. It checks the loudness, black frames and how clear the voice is, and its last line says whether it passed.
  11. Tell the user where the video, its subtitles, chapters and cover are, how long it is, what verify said and what you could not check. To post it, read explainroo docs publishing.

Topics

Read the topic of the step you are on. A long topic has sections: npx --yes explainroo docs <topic> --sections lists them with their size, and npx --yes explainroo docs <topic> <section> prints one, like npx --yes explainroo docs publishing buffer. Print only the section you need.

TopicRead it when
making-a-videoyou want each step above in full
writingyou write script.md: the style, the length and the script syntax
picturesyou plan the scenes, with four layouts to start from and how the camera moves
scene-apiyou write scenes.js: everything a scene can draw, with its options
troubleshootinga scene breaks, or before you write the first one
voicesyou pick a voice, or the user wants a paid voice or their own recording
languagesthe video isn't in English, or comes in several languages
japanese-and-chinesethe video is in Japanese or Mandarin Chinese
imagesthe user wants AI images
clipsthe user wants AI video clips
musicthe user wants their own music, a sound file or original music from Soundroo
sizesthe video is for a platform: sizes, captions, pace and the watermark
formatsyou use, install or publish a format
short-videoyou make a Short, a TikTok or a Reel
product-demosthe user wants a demo of an app or a website, also one recorded with explainroo record
presentersthe user wants an AI presenter next to the product
settingsyou change video.json, make a look of your own or set up a key
commandsyou need a command or a flag that this page doesn't show
publishingthe user wants a package for a platform, a link or a post
web-playerthe video goes on a website, in an iframe
transparent-videothe video goes over camera footage in a video editor
cutthe user wants to cut their own recording by its words
connectorsyou add an outside service to explainroo
contributingyou change explainroo itself

相关技能