Last week I made a 39-second animated video.
Glowing title, colored bullets that pop in one by one, an underline that draws itself across the screen. The kind of motion graphics that usually means After Effects and a weekend.
I made it with five prompts, all typed in plain English.
I never opened a timeline or touched a keyframe. I described the video I wanted, an AI agent built it, and I asked for changes until it looked right.
The tool that makes this possible is open, free, and runs through the same AI assistant a lot of you already use for writing and research.
Today, I'm walking you through how to make your first video 👇
Your best prompts are the ones you'd never bother typing.
The detailed ones. The ones with examples and edge cases. Wispr Flow lets you speak them instead — clean, structured, ready to paste into any AI tool. Free on Mac, Windows, and iPhone.
Your AI agent can learn to be a video editor
First, two quick definitions, because the rest of this falls into place once you have them.
An agent is an AI that takes actions instead of just chatting. It writes files, runs commands, and carries a task to a finished result on its own. Claude Code is one of these. So are Codex, Cursor, and Gemini CLI. You talk to them in normal language and they do the work.
The second piece is HyperFrames, a free tool from the video company HeyGen.
HyperFrames builds video out of HTML, the same basic code behind every web page.
That’s good for us, because AI agents are already excellent at HTML. They write it all day.
So the moment your video is "just a web page that moves," your AI agent becomes a video editor without learning anything new.
You describe a scene, it writes the HTML, HyperFrames turns it into a real MP4 file.
What you'll need (It's lighter than it looks)
You need three things, and you set them up once.
An AI coding agent. I'll use Claude Code here since it's what I reach for, but the exact same steps work in Codex, Cursor, or Gemini CLI. These run in a terminal, the plain text window on your computer where you type commands. Don't let that scare you off. You will be typing sentences, not code.
The HyperFrames skills. A skill is a recipe a tool follows for a specific job. HyperFrames ships a set of them that teach your agent exactly how to build a correct video. You install them with one line:
npx skills add heygen-com/hyperframes
Paste that into your agent, pick the core set when it asks, and you're done. Now your agent knows the rules of building video, so it gets things right the first time instead of guessing.Two helpers running quietly underneath. HyperFrames needs Node.js (which runs the tool) and FFmpeg (which saves the final video file). Your agent can install both for you if you ask. One-time setup, then you forget they exist.
That's the entire kit. The work itself is just talking.
Make your first video in five prompts
Here's the flow I used, start to finish. Each step is one prompt for your agent.
1. Point it at HyperFrames and describe what you want. Be specific about length, look, and mood:
“Using /hyperframes, make a 10-second product intro. Dark background, a title that fades in, three colored bullet points that appear one at a time, and subtle background music.”
2. Preview it in your browser. Ask the agent to open a live preview:
“Preview it so I can watch it play.”
A window opens and your video plays. Every time you change something, it updates instantly.
3. Watch it, then ask for fixes in plain words. This is where it feels like magic. You talk to it the way you'd talk to a human editor (more on that below).
4. Keep going until it's right. Two or three rounds of small changes is normal. Mine took five total.
5. Save it as a video file. When you're happy:
“Render it to an MP4.”
You get a finished output.mp4 you can post anywhere, the same as any other video.
Talk to it like a video editor
Real examples of little edits that you can make:
Make the title twice as big and switch to a dark background.
Add a fade-out at the very end.
Add a lower third at the 3-second mark with my name and title. (A "lower third" is that name banner that slides in across the bottom of the screen on TV.)
Slow the bullet points down so each one holds for a full second.
It goes well beyond title cards, too:
Point it at a spreadsheet and ask for an animated bar chart race.
Hand it a PDF and ask for a 45-second summary video.
Give it a blog post and ask for a vertical clip for TikTok with captions that bounce in time with a voiceover.
Describe what you want. Hyperframe executes. Repeat until it's done 🤝
Open your AI agent, make sure the HyperFrames skills are installed, and paste this in.
Fill in the bracket with whatever you want to make 👇
Using /hyperframes, make me a short video.
What I want:
[Describe it in plain English. For example: "A 12-second intro for my newsletter. Black background, my logo fades in, then the words 'Build with AI' type out one letter at a time, then a soft glow."]
Do this:
1. Build it.
2. Open a live preview so I can watch it play.
3. Wait for my feedback, then make the changes I ask for.
4. When I say it's ready, render it to an MP4.
If you need Node.js or FFmpeg installed first, set them up for me before you start.Start with one short, simple video.
Once you've watched the preview update from a single sentence of feedback, the whole thing clicks.
Want to get the most out of ChatGPT?
ChatGPT is a superpower if you know how to use it correctly.
Discover how HubSpot's guide to AI can elevate both your productivity and creativity to get more things done.
Learn to automate tasks, enhance decision-making, and foster innovation with the power of AI.
Your roundup of the latest model releases and updates from the biggest AI labs.
The AI coding agents are in a price-and-power war. A wave of new agentic coding models shipped this month, with Microsoft and Google going head to head against Anthropic and OpenAI. These are the same agents you'd use to drive HyperFrames, and they're getting faster, cheaper, and more capable by the week. For anyone building with these tools, the work they can do for you keeps expanding while the cost keeps dropping. (CNBC)
xAI's Grok Imagine Video 1.5 goes live. On June 16, xAI released its newest image-to-video model (it turns a still picture into a moving clip) and claimed the top spot on the public quality leaderboard, at a fraction of a competitor's price. It points to a useful split: generated video like this is great for lifelike footage, while the code-driven approach in today's edition wins when you need precise titles, charts, and on-brand graphics. Knowing which tool fits which job is becoming its own skill. (xAI)
OpenAI and Broadcom unveil a custom AI chip. The two announced an inference chip tuned to run large models faster and cheaper. Inference is just "the AI doing its job after it's trained," so cheaper inference means the assistants you use every day cost less to run. Why it matters: the price of running agents keeps dropping, which is why doing real work with them is getting accessible. (CNBC)
Advertise with Build with AI
Get in front of an audience of professionals using AI day-to-day: founders, engineers, operators, and product builders.
Interested in advertising? Respond to this email for rates and details.
Until next time,
William Ryan
Editor-in-Chief @ Build with AI
PS: Follow me on X for daily updates and AI workflows.