How can my Grok Bot make an instructional video?

The Grok Bot docs don't describe making videos. Here is what they do cover, plus the steps that worked for us when we made our own video, and what didn't.

All Grok Bot questions: https://www.soundminds.ai/faq.

Last updated Oct 18, 2026

Short answer

The Grok Bot docs don't describe making videos. They do say that Grok Bot "works from a persistent cloud computer," and that "It can use a browser, command line, files, and connected tools without depending on your laptop remaining open." xAI's developer docs also describe a text-to-speech service that can "Convert text into spoken audio with a single API call."

In our experience, a Bot can do most of the work on its own computer. Ours wrote the script, built the slides, made the narration with xAI's text-to-speech, and put the video together. We checked the script before any narration was made, and we approved the third version of the video as is before it was published.

xAI's quickstart for its API says to "Create an API key," and its text-to-speech page links to "Get API Key." Cursor's help says to "Keep API keys and other credentials out of chat and ordinary files."

Steps

The docs don't give these steps as one workflow. Each step starts with what the docs say. The sentences that start with "In our experience" are what worked for us, not documented features.

Step 1: Ask your Bot for a script first, with its sources. The docs suggest asking for "A document with headings and source links." In our experience, we made every fact in the script a word-for-word quote from a page we cited, and we checked the script's wording before any audio was made.

Step 2: Ask for the slides. The docs list "A slide deck with speaker notes" as one kind of result you can ask for. In our experience, the Bot built each slide as a simple web page and took a screenshot of it in a browser on its own computer, one slide for each part of the script.

Step 3: Try a few voices on a short sample. xAI's docs say, "Listen to samples and choose the best fit for your use case." In our experience, we tried our opening lines in six voices before we picked one. The "What didn't work for us" section below says which voices we turned down.

Step 4: Make the narration. xAI's docs describe timing data for each character of speech and say it is "Ideal for syncing captions, karaoke highlights, real-time lip-sync, or other time-aligned applications." In our experience, we made one narration file for each slide, and the timing data let the Bot line up the slides and the captions with the voice.

Step 5: Have your Bot put the video together. The Grok Bot docs don't describe making a video. In our experience, the Bot used a video tool that was already on its computer to join the slides and the narration, even out the volume, and make a captions file.

Step 6: Review the video before anyone else sees it. The docs say "Files, images, links, and tool results appear as cards in the conversation." They also list "Publishing content" among the actions to set a clear stopping point for. In our experience, we made three versions of the video, and the third one is the one we published.

Step 7: Save the method once it works. The docs say, "Start with a one-time task. Make it reliable, save the method as a skill, and only then automate it."

Your xAI API key

Don't paste your API key into a chat, including a chat with your Bot. Cursor's help says, "Use the secure secret card when a Bot asks for one." The Grok Bot docs also say that every Bot on your account shares one computer, and that "Command-line credentials are shared." So any of your Bots may be able to use a key that one Bot has.

What worked for us

This section comes from making our own video on October 18, 2026. It's what worked for us, not documented behavior.

  1. xAI's docs describe a [pause] tag and say to use it "to add dramatic timing or let a thought land." We marked each pause in the script and sent it as that tag.
  2. We evened out the loudness of each narration file and then of the finished video.
  3. The docs say to "Treat temporary directories, manually installed packages, and uncommitted application state as replaceable." Keeping the script, the slides and the build steps in project files let the Bot remake only the parts that changed. Our second version needed new narration for just four slides, and our third version needed no new narration at all.

What didn't work for us

This section also comes from our own notes on the video we made on October 18, 2026. It's our experience, not documented behavior. Where our notes don't give a reason, we say so.

  1. We made test clips of our opening lines with three voices from Piper, an offline voice tool, and we didn't use any of them. Our notes don't record why we turned them down. Before the test, our plan said the Piper voices would sound clear but obviously computer-made, and that they'd give us only limited control over expression.
  2. We made the same test clip with three of xAI's voices, Eve, Rex and Ara, and we used Eve for the whole video. xAI's docs say Eve "is the default" voice, but our notes don't record why we picked Eve.
  3. In our first version, we wrapped the last sentence of one slide in xAI's soft tag. The tag made the last words so quiet that they sounded cut off. We recorded that line again without the tag, and the new ending fades out naturally.
  4. In our first version, the Eve voice read the hyphen in a made-up web address out loud. We fixed the hyphen by spelling all three made-up addresses the way they should sound in the text we sent for narration, and the slides and captions kept the written addresses. xAI's docs also describe a "replace" option, which they call a "Map of phrases to spoken substitutions applied before synthesis." We didn't use it.
  5. For our second version, we set a stricter limit on the loudest peaks in the finished sound. Our notes don't say what problem that fixed.
  6. In our second version, we made the silence at the end of each slide a full second instead of 0.7 seconds, and our notes don't say why. For the third version, we went back to 0.7 seconds, which made the video 3 seconds shorter without changing any words.
  7. Our script estimated about 2 minutes and 48 seconds at a calm reading pace, but the narrated video ran about 3 minutes and 38 seconds. We wrote up cuts to shorten it, and in the end we kept the longer version.
  8. In our second version, one word came out quieter than the words around it. We measured it, found the whole word was there, and didn't change it.

What the docs don't cover

The Grok Bot docs don't describe building slides as pictures, making narration, or putting a video together. The steps on this page combine what the docs do say with what worked for us.

Copy this message to your Grok Bot

This message asks your Bot to draft the script and a slide plan first, and to wait for your yes before it makes any narration or video. Replace the part in square brackets with your topic.

I'd like a short instructional video about [your topic]. Before you make any audio or video, draft the script and a slide plan and show them to me.

1. Use only facts you can quote word for word from pages I approve, and list each source with its link.
2. Plan one slide for each part of the script.
3. Do all of the work on your own cloud computer, not on mine. Tell me before you install anything.
4. Wait for my yes before you make any narration. If you need an API key, show me a secure secret card. Never ask me for a key in chat.
5. Never publish, post, or send the video anywhere. Show it to me in this conversation first.
6. When the video works, offer to save the steps as a skill, and wait for my yes.

Sources

xAI docs, Text to Speech (developer docs): https://docs.x.ai/developers/model-capabilities/audio/text-to-speech

Convert text into spoken audio with a single API call.

Listen to samples and choose the best fit for your use case

Ideal for syncing captions, karaoke highlights, real-time lip-sync, or other time-aligned applications.

to add dramatic timing or let a thought land

Map of phrases to spoken substitutions applied before synthesis.

is the default

Get API Key

xAI docs, Quickstart (developer docs): https://docs.x.ai/developers/quickstart

Create an API key

xAI docs, Grok Bot: Use the computer and apps: https://docs.x.ai/grok-bot/computer-and-apps

Grok Bot works from a persistent cloud computer. It can use a browser, command line, files, and connected tools without depending on your laptop remaining open.

Command-line credentials are shared

Treat temporary directories, manually installed packages, and uncommitted application state as replaceable.

xAI docs, Grok Bot: Files and results: https://docs.x.ai/grok-bot/files-and-results

A document with headings and source links

A slide deck with speaker notes

Files, images, links, and tool results appear as cards in the conversation.

xAI docs, Grok Bot: Approvals, security, and privacy: https://docs.x.ai/grok-bot/approvals-security-and-privacy

Publishing content

xAI docs, Grok Bot: Skills and routines: https://docs.x.ai/grok-bot/skills-routines-and-automations

Start with a one-time task. Make it reliable, save the method as a skill, and only then automate it.

Cursor help, Store secrets securely: https://cursor.com/help/grok-bot/secrets

Keep API keys and other credentials out of chat and ordinary files.

Use the secure secret card when a Bot asks for one.

See how I set up Grok Bot: https://www.soundminds.ai/setup.