How to make motion graphics with AI without writing code

Every prompt below has four parts: the goal, the inputs, the constraints and the definition of done. After that you ask for one change at a time and look at the result before you send the next prompt. That habit is most of what "vibe coding done well" means. It matters even more for video, because a vague prompt gives you a video that's wrong in ten small ways at once, and you can't tell which words caused which.

You don't need design training for this. You need to watch your own preview the way a stranger would and say exactly what bothers you.

What to do: write one page that fixes the message, the length, the format, the brand and the scenes. One idea per scene.

Tool: Claude (or ChatGPT).

Choose the format first. 16:9 is for YouTube and websites, 9:16 for Reels, Shorts and Stories, 1:1 for LinkedIn and Instagram feeds. The QuickBilling video starts in 16:9 and gets adapted later. The LinkedIn number starts in 1:1. Keep to one or two fonts, a small palette and one transition style, and make the text big enough to read on a phone.

For QuickBilling that comes to five scenes of four seconds: the problem, three features, a call to action. The SolutionPlus post needs only three beats: the hook, the number, the call to action.

Goal: write a reusable motion graphics brief for a 20-second explainer for QuickBilling, an invoicing app for freelancers.
Inputs: audience: freelancers who chase late invoices. Message: invoicing takes minutes, not evenings. Structure: problem, then three features (send an invoice fast, automatic payment reminders, export for the accountant), then the call to action "Try QuickBilling".
Constraints: five scenes of four seconds. 16:9 master, later 9:16 and 1:1. Dark navy background, white text, one sans-serif font, one transition style, text readable on a phone. Do not invent features.
Definition of done: a table with scene, seconds, what we see, on-screen text and voice-over line. Ask me for brand colours and fonts if I have not given them.

Goal: write a second brief in the same style for a 10-second animated number for a SolutionPlus LinkedIn post.
Inputs: one result from client work, stated as a single number. I will fill in the number; use [NUMBER].
Constraints: three beats of about three seconds, 1:1, same fonts and colours as the first brief, a very large number and a small caption.
Definition of done: a table with beat, seconds, what we see and on-screen text.

What to do: turn the brief into a storyboard with one row per scene. Fix the words and the timing here. Changing a table costs nothing, and changing a rendered video costs a render.

Tool: Claude, in the same chat as the brief.

Read every voice-over line out loud with a timer running. If you stumble, or a line runs past its four seconds, it's too long. When the table is right, tell Claude it's final. Every later prompt points back to it.

Goal: turn the QuickBilling brief into a final storyboard.
Constraints: columns: scene, time range, what we see, on-screen text (six words at most), voice-over line (ten words at most). One idea per scene. Total exactly 20 seconds. Do not make any video yet.
Definition of done: only the table, followed by up to three questions about anything unclear.

Keep the same table. Shorten all on-screen text to four words at most and keep the total at 20 seconds. Return the full table.

Three tools cover most of what a founder or marketer needs, and you can mix them: titles from one, footage from another. I describe each from its own documentation, which is listed at the end.

Claude Motion lives inside Claude. It makes editable animations from text, charts, shapes and images and exports an MP4. Anthropic says no video model is involved, so Claude animates your content and never invents footage. To start one, select Output in the message box and choose Motion, or just ask Claude for an animation. The catch is the plan. When I last checked Anthropic's help page, Motion was a beta on Team and Enterprise plans only. I keep the dated plan details in my post on Claude Motion and Higgsfield Katana, so check there and on Anthropic's pricing page before you count on it.

HyperFrames, from HeyGen, is a free, open-source (Apache 2.0) framework that turns web pages, media and animations into MP4 video, rendered on your own computer. There are two ways to use it without writing code. The first is the HyperFrames Studio app for Mac, Windows and Linux, which is free with a free HeyGen account. You describe the video, an agent on your computer builds it, and you change it by typing, pointing at the picture, drawing or commenting. That agent is a separate program. The docs list Claude Code, Codex, Grok, Antigravity, GitHub Copilot and OpenCode, and Studio can install one for you from its Connect an agent screen. Most are billed by their own provider, and OpenCode is listed as free. The second way is for people who already use Claude Code, Codex, Cursor or Gemini CLI: install the HyperFrames skills once, then ask your agent to make a video. Studio's paid Pro plan adds HeyGen's video, image, voice and music models.

Higgsfield is for scenes that need real-looking footage: a person at a desk, a product on a table, a camera move. It says it offers more than 35 image and video models. Its Elements let you save a character, a location or a prop and call it up again with an @ tag across images and video.

In both of my videos the hardest scene is text and a number. So I'd start in HyperFrames Studio or Claude Motion, and add a Higgsfield shot only if a scene needs a person.

What to do: paste the final storyboard, but ask for scene one only. Approve it, then ask for the next. Going scene by scene keeps every problem small.

Tools: HyperFrames Studio, Claude Motion, or your coding agent with the HyperFrames skills.

In Studio, you type into the Make something box on the home screen and choose Let's go. The agent may ask you a few questions first, each as a card with options. By default it then shows its own storyboard in the Storyboard tab before it builds anything. Compare that with your table, ask for changes to any frame that differs, and only then choose Build the video.

How big should the first request be? The HyperFrames README gives its own example: a 10-second product intro with a fade-in title, a background video and subtle background music. Mine below is about that size, with the storyboard attached.

Goal: build scene 1 of the storyboard below, and only scene 1.
Inputs: the final storyboard table.
Constraints: on-screen text exactly as in the table. Dark navy background, white text, one sans-serif font. The title fades in with a gentle ease-out. Exactly four seconds. Do not start other scenes.
Definition of done: a preview I can watch, and a short note on what you built and what I should check.

A four-second shot for an invoicing app explainer: a freelancer at a desk in soft morning light, laptop open, slow push-in. Keep the upper third of the frame empty for a title. Calm, natural colours that sit well on dark navy. One shot only.

What to do: watch the whole preview. Then name one problem, with the scene, the second and the object. The four words I reach for most are timing, easing, overlap and contrast.

Tool: whichever tool built the scene. In HyperFrames Studio you can also point at the frame or draw on it, which is quicker than describing where something sits. In Claude Motion you can ask in the chat or adjust the animation yourself in the editor.

Don't batch fixes. If a change makes the preview worse, ask the tool to undo that one change. For the LinkedIn number, check the frame where the number lands. It has to be huge, sharp and readable on a phone with the sound off.

Keep everything else the same. In scene 1 only, start the title half a second later and make the ease-out slower. Keep the scene at four seconds. Show me the new preview.

Keep everything else the same. Make the scene 2 text larger so it is readable on a phone, keep white on navy, and keep the text clear of the laptop image. Show me the new preview and wait for my check.

What to do: when every scene is approved, add captions, then sound, then the other sizes. In that order.

Tool: HyperFrames can do all three in the same project. Anthropic's help page for Claude Motion doesn't mention audio, captions or aspect ratios, so I'd export the MP4 from Motion and add sound and captions in a video editor.

Assume some people will watch with the sound off, and treat captions as part of the video. Keep them large, with strong contrast, and check every number twice. The voice-over is your own recording unless you pay for a voice model. Keep one calm music track low under the voice, and make sure you have the right to use it.

Render the 16:9 master first. Then adapt it to 9:16 and 1:1 with the titles re-centred inside safe margins. A plain crop cuts words off. In Studio, Export is in the project's top bar. MP4 at 1080p is the default, rendering on your own computer is free, and the file lands in your Downloads folder. Watch every file on your phone, once with sound and once without.

Goal: finish the approved QuickBilling video.
Inputs: the final voice-over lines from the storyboard.
Constraints: captions from those lines, large white text with strong contrast. One subtle music track, low under the voice. Render the 16:9 master first, then 9:16 and 1:1 with titles re-centred inside safe margins.
Definition of done: three MP4 files, and a list of anything that is cut off or overlaps on a phone screen.

What to do: save the brief, the storyboard format, the style rules and the prompts that worked. Next time you change the message and the numbers and leave the rest alone.

Tool: Claude, at the end of the chat, then wherever you keep notes. Each Studio project is a normal HyperFrames project on your disk, so your agent can start the next video from the last one. In Higgsfield, keep your saved Elements and reuse their tags.

This is where the time comes back. The first video in a series is a design project. The second should be closer to filling in a form.

Goal: turn this project into a reusable template.
Inputs: the brief, the final storyboard and the prompts that worked in this chat.
Constraints: keep the style rules (navy, white text, one sans-serif font, one idea per scene, one transition, phone-readable text) and the four-part prompt shape. Replace the specifics with placeholders: message, length, format, scenes.
Definition of done: plain text I can copy, plus the three prompts for the next video: brief, storyboard, build scene 1.

I wrote about why motion graphics became something you describe instead of keyframe in Will AI replace motion designers?. This post is the practical half. Designers still win on taste, and on the details a viewer feels but can't name. The tools take away the hours of keyframing. They don't take away the judgement.

If you want a video pipeline like this inside your own product, for example videos generated from your customers' data, that's the kind of thing I build in my agentic platforms work.