> ## Documentation Index
> Fetch the complete documentation index at: https://docs.stefanbrain.com/llms.txt
> Use this file to discover all available pages before exploring further.

> ## Agent Instructions
> Follow the content contract in AGENTS.md.
> Treat text as the canonical explanation; video and screenshots enhance it.
> Do not publish unverified product behavior or duplicate an existing canonical article.

# GrokBot + StefanBrain: Turn one brief into a reviewed video ad

> Coordinate prompting, generation, and editing so you can spend more time choosing the strongest creative and less time managing each handoff.

## Watch the walkthrough

<video controls playsInline preload="metadata" poster="https://stefanbrain.com/api/video/adaptive/tutorials/automate-video-ads-with-grokbot/poster-2026-10-27.jpg" alt="GrokBot and StefanBrain video workflow" aria-label="GrokBot and StefanBrain video workflow">
  <source src="https://stefanbrain.com/api/video/adaptive/tutorials/automate-video-ads-with-grokbot/full-frame-2026-10-27.mp4" type="video/mp4" />

  [Watch the walkthrough](https://stefanbrain.com/api/video/adaptive/tutorials/automate-video-ads-with-grokbot/full-frame-2026-10-27.mp4)
</video>

## Copy the group chat setup prompt

Paste this into GrokBot to define the team, learn from your existing videos, and agree on the process before generating anything.

```text theme={"system"}
Set up a Video Generation Groupchat. Goal here is autonomous video generation and editing. We need a QA Bot (to watch and judge videos based on specific skills and DR principles, pronunciation, and visual inconsistencies), a Prompting bot (to learn from QA, past prompts, and future prompts, upload all visual references, guidances, audio tracks, etc,), a Video Editor bot, to use SB’s native regeneration and editing platform to fix any inconsistencies. It will also use your own native video editor when needed to perform additional edits and cuts.

Additionally, use computer use, and analyze every video I ever generated with StefanBrain. Examine the prompts, and where the video went wrong. Use those findings to create a video prompting skill, so that we can ensure the highest quality results, every time. Use docs.stefanbrain.com to also look at what the team suggests for prompting video gen. Then, let’s align before you generate a video, because i want to QA your prompts and understand your reasoning. We will work towards this being fully autonomous. You will generate skill files as we go for each bot so that we follow a process that is repeatable and improves over time, resulting in this: as you generate more videos, learn from mistakes to continually improve and get better.

We need to setup a system here. Do you recommend we add any other bots into this groupchat before we proceed with full setup here?
```

Here, **DR** means direct-response advertising. Review the proposed team and prompting rules before approving the first generation. Learning from previous results helps refine the process; it does not guarantee that every generation will be usable.

## Connect StefanBrain to GrokBot

Set up **MCP** for StefanBrain tool access and **computer use** for working in the Video workspace and editor.

**1. Add the StefanBrain MCP connection**

In a GrokBot chat, paste:

```text theme={"system"}
Add StefanBrain as a custom MCP connection:
https://stefanbrain.com/api/developers/v1/mcp

Use OAuth and show me the sign-in step. If StefanBrain is already
connected, reuse that connection instead of adding it again.
```

Complete the StefanBrain sign-in and review **Allow access**. You need an active StefanBrain plan or trial. See the [StefanBrain MCP setup guide](https://docs.stefanbrain.com/developers/mcp.md#connect-an-oauth-client) for authentication options and troubleshooting.

Use the product endpoint above—not `https://docs.stefanbrain.com/mcp`, which only provides documentation.

**2. Sign in on GrokBot's computer**

Ask GrokBot to open the app:

```text theme={"system"}
Use your computer to open https://stefanbrain.com.
Pause so I can sign in, then confirm you can open my Video workspace.
Do not generate or edit anything yet.
```

Open the computer preview in GrokBot's right-hand panel, take control, and complete sign-in. Enter passwords and verification codes in the browser, not in chat.

MCP authorization and browser sign-in are separate. Connecting one does not sign you into the other. GrokBot's computer session is shared across your bots, so only sign in to accounts you want them to use. See [GrokBot's computer-use guide](https://docs.x.ai/grok-bot/computer-and-apps) for taking control and handling sign-in.

**3. Check both before setting up the group**

```text theme={"system"}
Confirm the StefanBrain MCP connection can list its available tools,
and that your computer can open my existing videos in StefanBrain.
Tell me which connection, if either, still needs attention.
Do not start a generation or change any videos.
```

If MCP cannot connect, follow the [MCP setup guide](https://docs.stefanbrain.com/developers/mcp.md). If the browser shows a login screen, take control and sign in there again.

Use MCP for interactive work with your assistant. For unattended pipelines or machine-scale generation, use the [Developer API](https://docs.stefanbrain.com/developers/quickstart.md).

## Set up a small production team

Create the group in GrokBot, not inside StefanBrain.

Create a group with clear responsibilities:

| Role | What it handles |
| - | - |
| Operations | Turns your brief into assignments and keeps track of the current version. |
| Prompting | Develops the concept, prepares the generation prompt, and carries forward useful references. |
| Editor | Makes targeted changes to the generated video. |
| QA | Checks pronunciation, visual consistency, and whether the result follows the brief. |
| Final review | Checks the proposed final version before sending it to you. |

You can begin with fewer roles. What matters is that someone owns each handoff and that the team knows when to stop and ask you.

## Give the team a clear brief

Include your product, audience, offer, creative direction, and references. Share examples of previous videos you liked and explain what worked: the opening, delivery, character, pacing, or product demonstration.

Ask for ideas before spending video allowance. For example:

```text theme={"system"}
Develop 10 distinct video-ad concepts for [product] aimed at [audience].
Use the attached product facts and references. For each concept, give me
the hook, visual direction, and reason it fits this audience.

Recommend one concept, but wait for my approval before generating.
Do not invent product claims. Ask me if a necessary detail is missing.
```

Once you choose a concept, give the team a concrete delivery target:

```text theme={"system"}
Produce one video using approved concept [number]. Keep the product,
character, and visual direction consistent with the references.

Show me the proposed prompt, style, format, duration, and usage cost
before generating. After generation, review the entire video. Identify
specific issues and propose focused edits rather than starting over.

Ask before any additional paid generation. Deliver the final candidate
and a short summary of changes. Do not publish or launch anything.
```

## Choose the style and format before generating

Use StefanBrain's Video workspace to choose the **Style**, **Format**, duration, and aspect ratio. Attach the correct Product and any visual references. Keep the prompt focused on the scene and message instead of packing it with unrelated instructions.

For a song ad, agree on the song direction and product pronunciation before generating. For a recurring character, reuse the same references so later clips have a clearer visual target.

See [how to prompt product videos](https://docs.stefanbrain.com/knowledge-base/video/prompt-product-videos.md) for the generation setup and [create AI song ads](https://docs.stefanbrain.com/knowledge-base/video/create-ai-song-ads.md) for the song workflow.

## Review the result before adding more edits

Have QA watch the entire video, then bring you a short list of concrete issues:

* Is the product name pronounced correctly?
* Does the opening match the concept you approved?
* Are the product and character consistent between scenes?
* Do the spoken claims match the product information?
* Can you follow the message without distracting overlays?

Keep the review practical. In the walkthrough, too much B-roll covers the original creative. An extra overlay is only useful when it makes the message easier to understand.

## Fix the weak spots without rebuilding everything

Open the generated video in StefanBrain's **Video Editor**. Use a focused dialogue edit for a mispronounced line, compare another hook when the opening needs work, and add captions or B-roll only where they improve the video.

Review the displayed usage cost before generating new takes. After an edit, replay the surrounding scene to check that the change still fits.

Follow [edit hooks, dialogue, and captions](https://docs.stefanbrain.com/knowledge-base/video/edit-video-dialogue-and-captions.md) for the individual controls.

For a multi-clip piece, the walkthrough also uses an external editing workflow to assemble clips and add audio. That assembly is separate from StefanBrain's native Video Editor. Check transitions, audio levels, and caption consistency across the finished piece.

## Keep what worked for the next brief

Save the approved prompt, references, chosen settings, and the edits that improved the result. Give the team specific feedback—such as “keep the character visible during the hook”—so future runs can reuse a useful rule instead of repeating the same mistake.

Review the finished video yourself before using it in an ad. Increase autonomy only after the workflow reliably produces work you would approve.

## Practical use cases

> [card] Choose a concept before spending video minutes
>
> Ask for 10 distinct concepts, compare the hooks and visual directions, then approve one for generation.

> [card] Fix a product-name pronunciation
>
> Have QA identify the exact line, let the Editor propose a replacement, and compare the new take before keeping it.

> [card] Assemble a consistent multi-clip ad
>
> Reuse character references across clips, then check transitions and caption consistency in the assembled video.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.