How to Create Marketing Videos With an AI Agent (Full Walkthrough)

Marketing teams spend hours in video editing tools, stitching together slides, adjusting animations, and exporting for different platforms. What if you could describe what you want in a prompt and have an AI agent produce the entire video for you?

That's exactly what we built at Gamut. In this post, I'll walk you through how I use a Gamut agent to create all of our marketing and product demo videos — from a simple text prompt to a fully rendered, on-brand video — without ever opening a video editor.

How It Works: Video as Code

The agent uses Remotion, a React-based framework that treats video as code. Every frame, animation, and transition is a React component. This means:

• Version-controlled: Your video is a codebase, not a binary file
• Programmatic: Text, colors, timing — everything is parameterized
• Customizable: The agent can iterate on specific scenes without re-rendering the whole video
• Reproducible: Same code, same output, every time

Because the video is just code, creating and editing it becomes a coding task — which is exactly what AI agents are built to do.

Step 1: Give the Agent a Prompt

Every video starts with a prompt. I explain what I want to show, the context, and any specific requirements. For example, here's what I provided for a recent LinkedIn video:

• The LinkedIn post text the video should accompany
• Ideas for the visual workflow and key points to highlight
• Product screenshots I wanted included
• A link to our website for brand reference

The agent reads everything, understands the goal, and starts planning the video structure.

Step 2: Automatic Brand Extraction

One of the things that impressed me most: the agent doesn't just guess your brand style — it goes to your website and figures it out.

The agent visited gamut.so, ran JavaScript in the browser to extract:

• Exact fonts used across the site
• Brand colors pulled directly from the CSS
• Visual style patterns and spacing

This gives it a complete brand guideline to work from, ensuring every video stays on-brand without you having to maintain a separate style document.

Step 3: The Self-Improving Render Loop

This is where things get interesting. The agent doesn't just write the code and ship it. It enters a continuous improvement loop:

1. Build the video script in Remotion
2. Render the video
3. Extract frames from key moments
4. Review each frame visually — checking composition, readability, and timing
5. Identify issues and fix them
6. Re-render and review again

The agent keeps cycling through this loop, catching things like text that's too small, animations that feel off, or colors that don't match the brand. It's like having a video editor who does their own QA.

Step 4: Interactive Feedback

Once the agent delivers a version, you can preview it right in the Gamut app. But here's the best part — you can leave comments directly on the video.

Each comment is:

• Timestamped to the exact moment in the video
• Positioned at a specific location on the frame
• Visible to the agent so it knows exactly what you're referring to

So instead of writing vague feedback like "the transition around the middle feels off," you can point at the exact frame and say "move this text up" or "this animation is too fast." The agent reads your comments and applies the fixes in the next iteration.

Multiple Video Styles, One Agent

The same agent handles completely different video styles. We've used it for:

• Animated explainers with branded motion graphics
• Product demos that mimic the actual product interface
• Social media clips optimized for LinkedIn, Twitter, and Instagram
• MCP integration demos showing how to connect Gamut to specific services

For the product demos, the agent actually pulls the source code to faithfully replicate the product environment inside the video. This means the demos look exactly like the real product — because they're built from the same codebase.