AI motion graphics workflow

How to Make Motion Graphics Videos With Remotion and AI

By Shreyas Raj. Written from my YouTube video, about 10 minutes long.

Video: Google Antigravity + Remotion is INSANE (Motion Graphics in 10 Minutes) on the AI with Shreyas Raj YouTube channel.

Short answer: Install Remotion with its agent skills inside Google Antigravity, drop your voiceover in the public assets folder, generate a detailed video prompt with the master prompt, and paste it to a Claude model in the agent. Remotion builds the video, which you can preview and render.

What you will learn

What Remotion is and what it can produce
How to install Remotion into Google Antigravity with agent skills
How to organize an assets folder for the agent
How to write a video prompt using the master prompt
The prompting rules that keep the output from being poor

What Remotion does

Remotion lets you make videos programmatically, and it now offers an agent skill so an AI agent can write the video for you. In my test videos I made a voiceover driven promo for a coffee brand and a motion graphic for a to do list app with features like smart scheduling, focus mode and cross platform sync.

It can also use 3D models through Three.js, add captions, and edit full length videos, for example removing cuts. The simple demo in this guide just shows the basic flow.

If you make videos for clients or for your own channel, this is useful in two ways. You can generate short promo clips and explainers from a script or voiceover, and you can use the same setup for captions and edits. I describe it in the video as a way to edit your own videos on autopilot, though you should test that on your own footage first.

The reason Remotion pairs well with AI agents is that videos are written as code. An agent that understands the rules can write, adjust and re-render a scene without you touching a timeline.

Why I install it in Antigravity

The Remotion site offers several ways to install it. I use Google Antigravity, because Claude Code requires a paid Claude subscription. You can also install into VS Code or Claude Code if you have them. If you have not installed Antigravity yet, do that first.

Make a new folder on your desktop, for example remotion test, and open it in Antigravity using open folder. Everything for the project will live here.

Step 2: Install Remotion and the agent skills

Open a terminal in Antigravity, copy the install command from the Remotion page and run it. Answer the prompts as follows:

Once the prompts are answered, the installer sets everything up. The scope choice is about convenience. Global means you can call Remotion from any folder at any time. Project scope keeps the files inside this one project, which is cleaner for a demo.

Choose the blank template
Say yes to adding Tailwind
Say yes to adding agent skills
Select Antigravity as the place to install the skills
Choose project scope (global would make Remotion available from any folder, project scope only installs into this folder)
Choose the symlink install method, which is the recommended option

Step 3: Install dependencies and add assets

Run npm i to install the remaining packages. While it installs, go into the public folder and create a new folder called assets. Put your files in here: images, videos or audio. In my demo I added a voiceover about coffee.

What matters is that you tell the agent there is something in this folder and that it should use it for the video.

Run npm run dev in the terminal. This starts a local development server and opens the Remotion dashboard. You cannot do much in the dashboard yet. The real work happens by prompting the agent.

If you are working with a voiceover, keep the file name simple and descriptive, such as the brand name followed by voiceover. Later you will paste its path into the prompt, and a clean name avoids typing errors.

Step 5: Build the prompt and run it

In the agent panel, select a Claude model. I used Claude Sonnet 4.5 with thinking, because Claude models work best with Remotion.

Open the Remotion master prompt, which I link in the description, and copy it. Paste it into an LLM of your choice (I used Perplexity) and add a request such as: based on the above prompt, write a motion graphic video prompt promoting a coffee shop and talking about Arabica coffee, and attach the voiceover so it understands the content.

Copy the long prompt it produces, paste it into the Antigravity agent, and add the path of your audio file inside the public assets folder. Press enter and the agent writes the video code.

Why go through a second AI first? Because the master prompt is long, and a general LLM can combine it with your voiceover and your goal to produce a detailed scene by scene prompt. The Antigravity agent then has enough information to build the whole video, instead of guessing.

Prompting rules you should always follow

Remotion ships with rules and skills the agent must read, and without them the output is poor. Whenever you write a Remotion prompt, tell the agent to first read the skills folder and the best practices for prompting, and to go through the rules folder and the skill file, so it understands how to structure the output.

The skills live in a GitHub repository with rules for when to use what. You do not need to intervene, because the AI follows them on its own once you point it there.

In practice this means one extra sentence at the top of every Remotion prompt. It costs nothing, and it is the difference between a messy video and one that follows the framework properly. I would not skip it, even for a quick test.

Preview and render

Once the agent finishes, the video shows up in the Remotion dashboard. In my test it played the voiceover with motion graphics. I call the result decent, not perfect, and it could have been better, which is normal for a first attempt.

If you like it, render it and use it however you want. You can keep prompting the agent to refine the animation.

For the coffee demo the result was a clean motion graphic that followed the voiceover. My earlier test videos for a to do app and some geography videos looked even better, which tells me that the quality depends heavily on the prompt and on how well the agent follows the Remotion rules.

I also tested the whole flow by assuming the video was already generated, so I could show the download options. The dashboard lets you preview each result before you render, which saves time when you are iterating.

Watch the full 10 minute video to see the install prompts, the prompt flow and the finished coffee promo preview.

Watch the full video See EducateXAI

Results vary and depend on your effort. Read the results disclaimer.

Questions people ask

Do I need Claude Code to use Remotion with AI?

No. In the video I install Remotion inside Google Antigravity and select a Claude model in its agent panel. You can also install the skills into VS Code or Claude Code if you use those.

What goes in the assets folder?

Put images, videos or audio you want in the video inside public, then assets. The folder type does not matter much. You just need to tell the agent the files are there and how to use them.

Why is my Remotion output poor?

Usually the agent has not read the Remotion rules. Add an instruction to read the skills folder, the prompting best practices and the rules folder with the skill file before it writes anything.

Can Remotion edit existing videos?

In the video I say it can add captions and edit full length videos, for example removing cuts. My demo only shows a simple motion graphics example, so test that on your own footage.

More from the channel

Prefer short reads? Start with the EducateXAI guides.