Somewhere on your to-do list is a video you keep not making. A sermon series bumper. An announcement for the app launch. A 15 second clip for social. You know it would help, and you also know what it costs: hours in an editor you barely know, or a design contractor and a two week turnaround.
That cost just disappeared. You can now describe that video to an AI agent in plain English and get a finished, on-brand animation back in minutes. Not a template with your logo dropped in. A real motion piece, designed for your message, in your colors and your font, rendered as an MP4 you can post anywhere.
Here's the kind of thing we mean. One prompt, a couple branded assets, and a promo your church could post this week:
This is how our team makes video now. A first draft exists within minutes of describing it, and ours usually ship the same day.
When to reach for it
This approach fits a specific kind of video. It shines when your video is built from type, logos, icons and graphics:
- Announcements and feature reveals. A new ministry, a new app, a new sermon series.
- Social clips. Short, punchy, branded motion that stops the scroll.
- Captions and overlays. Layer animated titles, lower thirds and callouts onto footage you already have.
It's the wrong tool for filmed storytelling. If the heart of your piece is camera footage of real people, shoot that with a camera. Then come back here for the intro, the captions and the graphics package around it.
Set it up
These directions use Claude, but the skill works the same in other tools (OpenAI Codex, Grok Build, Hermes and friends). One heads-up: you'll need Claude Code.
Wait! Don't go anywhere. Claude Code sounds like a developer-only thing, and it isn't. You talk to it exactly the way you talk to any other AI agent, plain English, no special commands to memorize. The easiest way to get it is the Claude desktop app, where it's built right in. Nothing to configure.
1. Install it and make your first video
Paste one prompt. It installs the skill and makes your first video in a single step:
Install the HyperFrames skill from heygen-com/hyperframes, then use it to
make a 10 second video: dark background, one line of white text that says
"Hello from [your organization]", fade in, hold, fade out. Save it as an MP4.
Your agent pulls in the skill and sets up what it needs on its own. If it's missing a tool, it tells you and walks you through a quick one-time install, you don't have to know what any of it is. When it's done you'll have an MP4. That's the whole setup.
From here on, just say /hyperframes (or "use the hyperframes skill") in any prompt.
Prefer the command line? npx skills add heygen-com/hyperframes installs it directly, and it's the one method that keeps itself updated. The source lives at github.com/heygen-com/hyperframes.
2. Prep your assets
Before your agent can speak your brand, it needs to know your brand. A folder with your logo, fonts and hex colors gets you most of the way. Our team went further and published a machine-readable brand kit, a single brand-kit.json behind a public URL at branding.triumph.tech (open source if you want to copy the structure). Then one prompt does this:
Read the brand kit at https://branding.triumph.tech and treat it as the
source of truth for colors, fonts and logos.
Use /hyperframes to build a flashy 15 second vertical reel for social
media hyping our new whitepaper, "Lights, Camera, Algorithm". Fast cuts,
bold type, our colors everywhere. Open with "An AI agent made this video",
then the big idea, and end on triumph.tech.
3. Run a simple prompt of your own
Now make something real. The promo at the top of this page came from exactly this step. Point your agent at real assets and describe the video the way you'd brief a human editor:
Grab the free "How to Pray" series assets from open.life.church
(the key art and the intro bumper).
Use /hyperframes to build a 15 second vertical reel promoting the
series on Instagram. Hook people with "Does prayer actually work?",
use the real series art and footage, and end on "New Series ·
This Sunday" with our handle. Match the series look.
Here is that exact prompt, rendered:
A few guidelines will sharpen your results fast:
- Give the frame, not the frames. Length, aspect ratio, one core message. Let the agent choreograph.
- One idea per scene. If a scene has to say two things, it's two scenes.
- Name real assets. A URL to your brand kit or a path to your logo beats "use our branding."
- Fewer words on screen. Motion carries the energy. Copy carries the point. Cut every line in half.
- React like a director, not a developer. "The ending feels abrupt, hold the last card longer" is a perfect note. The agent handles the code.
4. Build your intuition
Your third video will be dramatically better than your first, not because the tool improved but because you did. You'll learn what motion language to ask for, and naming a move precisely ("stagger the icons in one by one", "a gentle spring, no bounce") gets you exactly that move. Spend ten minutes in a resource like the animation vocabulary at animations.dev and you'll direct with the confidence of someone who's done this for years.
The bottom line
Video used to be the most expensive way for an organization to communicate and the first thing cut when time ran short. With an HTML video skill in your AI agent and a brand kit it can read, video becomes what it should have been all along: something you ask for.
And if you ever want to nerd out: every video on this page is secretly a webpage your agent wrote. You don't need to know that to use any of this. It's a fun thing to know.
参考資料