agentleFS
adamperlisadamperlis/video-skills/skills/walkie-voiceover/SKILL.md

walkie-voiceover

Skill45 starsChanged 3 days agoMITView on GitHub

Give a video a voice with Walkie. Turns a film's script into a voiceover audio file by asking the user's own Walkie app to narrate it, then hands the file to the video step. Use with brand-film, narrated-product-film, or any video skill when the user asks for a voiceover, narration, VO, "give it a voice", or a different voice or read for an existing cut.

The file

We split this skill into the rules it gives an agent

8 rules. About 0 are common, 4 are distinctive, and 4 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈4

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 4

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Look for the Walkie tools list_narration_voices and narrate.

    line 14
  2. Distinctive:

    If they are missing, stop and tell the user plainly what to switch on.

    line 16
  3. Distinctive:

    If the tools are there but a call fails, read the message.

    line 23
  4. Very specific:

    Call list_narration_voices.

    line 41
  5. Distinctive:

    Keep one voice for the whole film unless the script has more than one speaker.

    line 45
  6. Very specific:

    Create the output folder first.

    line 49
  7. Very specific:

    Read the length of every take from the narrate result, or measure the file.

    line 64
  8. Distinctive:

    Leave about half a second of picture before the first word and enough after the last one for the ending to land.

    line 68

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Walkie voiceover

Walkie is a speech app that runs on the user's computer. When the user has connected it to their assistant, the assistant can ask it to read a script aloud and save the result as an audio file. This skill uses that to voice a video: script in, voiceover file out, picture timed to the voice.

It works alongside a video skill and does not replace one. Let brand-film or narrated-product-film decide the story and the picture; use this skill for the voice.

Check that Walkie is reachable

Look for the Walkie tools list_narration_voices and narrate. If they are there, carry on.

If they are missing, stop and tell the user plainly what to switch on. Do not swap in another voice service without asking.

  1. Install Walkie from https://trywalkie.com, open it, and sign in. Connecting an assistant needs the Walkie Work plan or above.
  2. Connect Walkie to this assistant on this computer. The steps for each assistant are at https://trywalkie.com/mcp/setup. Narration only works through the connection on the user's own computer, not the cloud one.
  3. In Walkie, open Settings, then MCP, and turn on "Let your assistant add recordings and narrate". It is off by default.
  4. Restart the assistant or start a new chat so the new tools appear.

If the tools are there but a call fails, read the message. The usual causes are that the Walkie app is closed or signed out.

Tell the user before the first narration that it counts toward their Read Aloud time in Walkie. Each new take counts again, so settle the script before recording it.

Settle the script first

Get the words right before making audio. Write short, speakable lines, one claim per line, and read them aloud. Spell out anything that could be misread: numbers, abbreviations, product names, and unusual names.

Mark the delivery:

  • Tone applies to a whole take: excited, happy, warm, calm, playful, serious, dramatic, deadpan, whispering, or sad.
  • Speed runs from 0.7 to 1.3, with 1.0 as normal. Change it to fit a duration only in small steps.
  • A single moment takes an English tag written into the text, such as [pause] before a turn, or [whispering] in front of one phrase. Tags are performed, not spoken. A [pause] is about one second.

Tone, speed, and tags apply to Walkie's cloud voices. If the voice list reports that Walkie is in On-Device Mode, one on-device voice speaks instead, the file is WAV, and tone and speed are ignored.

Choose a voice

Call list_narration_voices. It returns each voice's name, gender, and accent, the languages available, and the default voice. Any voice can speak any listed language; the accent is the voice's own.

Do not pick silently. Offer two or three voices that fit the brand and the audience, say why, and let the user choose. They can hear voices first at https://trywalkie.com/mcp/voices or in the Voice Lab inside the Walkie app. If the user has no preference, use the default and say which voice it was.

Keep one voice for the whole film unless the script has more than one speaker.

Narrate

Create the output folder first; narrate will not create it, and it will not overwrite an existing file. Put the audio where the video project keeps its sound, and number each take (voiceover-take-01.wav, voiceover-take-02.wav) so earlier takes survive.

Call narrate with the script as text, plus voice, tone, speed, language, format, and output_path. Ask for wav when the file goes into an edit or a render, and mp3 only for a quick listen. One call takes about as long as the audio it makes, and it returns the file's path and its length in seconds.

Choose how to split the script:

  • One take for the whole script gives the most natural read. Use it for short films where the picture can follow the voice.
  • One take per beat gives a separate file and an exact length for every line, so each beat can be retimed or re-read without touching the others. Use it when picture has to cut on the lines, or when the script will keep changing.

A single call takes up to 5,000 characters, about five minutes. Split longer scripts at paragraph breaks and join the files in order.

Hand the audio to the picture

The voice sets the timing. Before building or re-cutting the timeline:

  1. Read the length of every take from the narrate result, or measure the file.
  2. Find where each line starts. With one take per beat, the starts are the running total of the lengths plus the gaps you choose. With one long take, measure the pauses (for example with ffmpeg's silencedetect) or transcribe the file with word timings.
  3. Give the video skill a beat table: line, audio file, start, and length. Cut picture to those times rather than stretching the voice to fit the picture. If a line runs long, shorten the line and narrate it again.

Leave about half a second of picture before the first word and enough after the last one for the ending to land. Mix music well below the voice.

For Remotion, put the file in public/ and play it with an audio element at the frame the line starts. For an editor, import the WAV on its own track. To lay a finished take under an existing cut without re-rendering picture:

ffmpeg -i cut.mp4 -i voiceover-take-01.wav -filter_complex "[1:a]adelay=500|500,apad[a]" -map 0:v -map "[a]" -c:v copy -c:a aac -shortest with-voice.mp4

Check and deliver

Listen to the whole take before using it. Then transcribe the finished video's audio and compare it with the script: every word present, names said correctly, nothing clipped at the start or end. Watch once for sync, with each action landing on the words that name it.

The voice is AI-generated. Say so wherever the video is published, and tell the user that this is their responsibility when they post it. Report which voice, tone, and speed were used so the next take or the next film can match.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.