Can Claude make videos? Here’s the honest answer

Can Claude make videos? The short answer

Not the way video generators do. Anthropic’s documentation describes Claude as an image understanding model: it can analyze images, but it can’t generate or edit them.¹ The help center puts it just as plainly: Claude doesn’t make photos or illustrations the way image generators do.² Even Anthropic’s own animation feature works without a video generation model.³

So where do the videos come from? Code. A video file is a stack of frames plus a soundtrack, and frames can be drawn by code. Claude writes that code, a renderer turns it into frames, and an encoder packs the frames into an MP4. The result is a real video file you can post anywhere.

There are two ways to get there. Claude Motion builds short animations from your prompt as code, animating text, charts, shapes and images, and exports them as MP4. It’s in beta on Team and Enterprise plans and doesn’t create realistic footage or people.³ Claude Code is the bigger toolbox: an agentic coding tool that reads your project, edits files and runs commands.⁴ Pair it with a video renderer, and Claude can build anything that renderer can draw.

How Claude video generation works with Claude Code

Claude video generation with Claude Code always has three layers. Claude writes each scene as code. A renderer draws that code frame by frame. An encoder turns the frames and the audio into a video file. The renderer is the part you choose.

HyperFrames

HyperFrames is HeyGen’s open-source framework for turning HTML, CSS, media and seekable animations into MP4 videos. A headless Chrome browser steps through the composition frame by frame, and FFmpeg encodes the result. It works with animation libraries such as GSAP, Lottie and Three.js, uses the Apache 2.0 license and needs Node.js 22 or newer plus FFmpeg. The project also ships skills that teach coding agents like Claude Code its workflow.⁵ Our puppy reels were rendered with it. The basic loop is four commands:⁵

npx hyperframes init my-video
cd my-video
npx hyperframes preview
npx hyperframes render

You won’t type these yourself very often. You describe the scene in plain words, and Claude Code runs the commands, reads the output and fixes what breaks.

Remotion

Remotion lets you create videos and motion graphics programmatically with React.⁶ If your project already lives in React, it’s a natural fit. The license works differently from HyperFrames: Remotion is free for individuals and small companies, while bigger companies pay a license fee.⁷ We compare the two side by side in HyperFrames vs. Remotion.

FFmpeg

HyperFrames hands the final encoding to FFmpeg,⁵ a multimedia framework that can decode, encode, transcode, filter and play media.⁸ You rarely see it at work. But it’s also the tool behind most of what people mean by Claude video editing. More on that below.

What Claude is good at, and what it isn’t

Where Claude shines

• Exact text. Every label, number and caption is typed text, so it’s spelled exactly the way you wrote it.
• Word-exact timing. When the voice track comes with timestamps, each image can appear on the frame where its word is spoken.
• Diagrams and 3D. Cutaways, charts, maps and mechanisms: anything you can describe precisely.
• A consistent series. Fonts, colors and camera rules live in files, so they carry over from one reel to the next.
• Plain-language fixes. “At 0:12 the phone is too small” is a valid instruction. Claude changes that one thing and renders again.

Where Claude struggles

• Realistic footage and people. No video model means no photoreal clips from a prompt.³ If a reel needs a real-looking scene, the image or clip has to come from somewhere else: a camera, a stock library or an image generator.
• Watching video. Claude’s vision works on still images. Even an animated GIF only counts with its first frame.¹ To check a render, Claude Code exports stills and looks at those.
• Taste. Claude can follow written rules, but it can’t know what you find beautiful until you say so. Plan for a few rounds of feedback on the first scene.
• Setup. HyperFrames needs Node.js and FFmpeg on your computer,⁵ and in our setup the rendering runs there too. It’s a one-time install, but it’s more than a sign-up form.

Claude video editing: what it realistically means

Claude video editing doesn’t mean Claude opens a timeline and drags clips around. It means Claude Code writes and runs FFmpeg commands on files you already have. FFmpeg reads your inputs, filters and transcodes them, and writes new files.⁹ Its options are famously fiddly, which is exactly why it helps to have Claude write them.

Typical jobs: trim a clip, cut out a section, join several clips, convert formats, crop horizontal footage to vertical, shrink a file for upload, pull out the audio. Here’s what a trim looks like:

ffmpeg -ss 00:01:05 -i interview.mp4 -t 20 -c copy clip.mp4

That starts at 1:05, keeps 20 seconds and copies the streams without re-encoding, which is fast and lossless.⁹ There’s a catch. In most formats FFmpeg can’t seek to an exact frame, so it jumps to the nearest seek point before 1:05. When copying, that extra bit stays in, so the clip may start a little early. When re-encoding, FFmpeg discards it by default.⁹ Trade-offs like this are where Claude saves you the most time: you say “cut at 1:05 exactly,” and it picks the right command.

The limit from above still applies: Claude doesn’t watch your footage. You give it timestamps, or it exports stills and looks at them. For a talking-head edit where every cut depends on what someone says, a regular editing app is still faster, in our experience.

The workflow behind our reels

Our reels come out of the same loop in Claude Code. Claude writes, times, builds and renders. We pick the question and judge the result.

1. Pick a question: an everyday thing with a surprising answer. “Why does my trackpad click?” beats “Trackpads explained.”
2. Claude writes the plan: facts with sources, three hook options, a script of roughly 150 to 170 words for 60 seconds and a scene list.
3. An AI voice reads the script. ElevenLabs can return the audio with a start and end time for every character,¹⁰ which Claude groups into words.
4. Claude builds the 3D scenes, and each image lands on its word.
5. The render: a vertical 1080 × 1920 MP4.
6. Checks before handover: loudness, text inside the safe zone, and no moment longer than 1.5 seconds without something new on screen. Music gets added later, in the app.

The trackpad reel shows why this approach suits explainers. On a Force Touch trackpad, sensors beneath the surface measure how hard you press, and the click you feel is haptic feedback, not a mechanical click.¹¹ That’s a mechanism with exact steps and labels, and each one has to land on its word. Code lets you place every one of them on the right frame.

Want the loop without building it yourself? Our course Explain It in 60 packages it: a guide plus a Claude kit with a CLAUDE.md that Claude Code reads at the start of every session,⁴ four rule files for hooks, voice and timing, sync and motion, and a starter project that renders a short sample reel. You need a paid Claude plan with Claude Code and an ElevenLabs account.

Claude or a video generator?

If you need a realistic shot of a beach at sunset, use a text-to-video generator. If you need a 60-second explainer where every word, number and label is right and on time, Claude Code gives you that control. You can also combine the two: our AirPods reel puts an AI-generated café photo inside a scene Claude built. We break down when to use which in AI video generator vs. Claude Code.

Ready to try it on a question of your own? See how to make 3D explainer reels with Claude Code, from the first hook to the finished MP4.