The best way to add animated captions, word by word

HyperFrames Studio is the best free way to add word-by-word animated captions to a talking-head video. Drop the clip into the app and name a style, such as Hormozi-style word pops or karaoke; HyperFrames transcribes the speech, times every word, highlights the word being spoken and renders a new MP4. Your original video stays untouched, and you can fix any word or switch styles by asking.

Short answer: most caption apps want an upload and a preset. In HyperFrames you pick from dozens of caption styles or describe your own, change any word or timing afterwards, and exports made on your computer have no watermark.

What do animated captions look like on one take?

We captioned one 20-second talking-head take five ways, one HyperFrames caption style per sentence. The take is a HeyGen AI presenter, and the footage under the captions is the same in every frame.

• caption-highlight: bold capitals, and a red box slides behind each word as it is said. The TikTok look.
• caption-pill-karaoke: the line sits in a light pill and each word turns dark when it is spoken.
• caption-neon-glow: words light up in cyan as they are said; key words glow pink and stay lit.
• caption-editorial-emphasis: small words build the line, then one big italic word lands on the beat.
• caption-parallax-layers: one tall red word sits behind the speaker's head, with the line in front.

Can you add animated captions in the Studio app?

Yes. The HyperFrames Studio app for Mac, Windows and Linux runs your agent beside a live preview and a timeline. Add the clip in Make something and ask for captions. The Captions panel shows caption styles, HeyGen's caption templates plus the HyperFrames catalog: click one, and it goes into the chat as the style to use.

In Studio, the words come from HeyGen's speech model on your account, and the language is detected for you. Studio renders exports on your computer for free, with no watermark, and its free plan includes captions, AI voiceovers, music, sound effects, 3 AI video clips and 5 cloud exports a month. Pro, at $25 a month, adds unlimited cloud exports and top AI models. Studio needs a free HeyGen account. Download Studio.

Above: verbatim captions under a presenter, with one word, FASTER, set behind her. It was made from the prompt in the HyperFrames captions guide.

How do you add animated captions with your own AI agent?

You need an AI coding agent such as Claude Code, Codex or Cursor, Node.js 22 or newer, and FFmpeg. Claude Code needs a paid Claude plan; HyperFrames itself is free.

1. Install the HyperFrames skills in your project folder: npx hyperframes skills update. Then open or restart your agent there.
2. Put the clip in the folder and say what you want. Name the style, where the words go and any names the transcript may miss. For example: "Add caption-highlight captions to talk.mp4, 9:16, keep the footage unchanged. The product is spelled HyperFrames."
3. Let the agent transcribe and build. It runs the commands below, installs the caption style as editable code and times it to every word.
4. Check it. The agent looks at still frames of its own work before it renders. You can open npx hyperframes preview to scrub the timeline, fix a misheard word or move a caption.
5. Render. npx hyperframes render --output captioned.mp4 writes the MP4 on your computer.

These are the commands behind the five-style take above, run with HyperFrames 0.9.19 on a MacBook Pro with an M4 Max chip, with the times they took.

Need a subtitle file for YouTube as well? npx hyperframes transcribe transcript.json --to srt writes an SRT from the same transcript, and --to vtt writes a VTT.

Which caption styles can you ask for by name?

HyperFrames has 16 caption styles in its catalog. Each one installs as source code you can restyle, so a font, a color or the animation can change without starting over. Asking by name locks the look; describing the mood ("hype", "clean", "neon") lets the agent pick.

The other eight are caption-camera-follow, caption-clip-wipe, caption-glitch-rgb, caption-gradient-fill, caption-matrix-decode, caption-neon-accent, caption-particle-burst and caption-texture. Each one has a moving preview in the captions catalog.

How do you put captions behind the speaker?

Text behind the speaker needs a cutout of the person. npx hyperframes remove-background makes one on your computer as a transparent video. HyperFrames then stacks the footage, the big word and the cutout, so the head covers part of the word. The captions guide has prompts for this and for designed cards.

• Use it once or twice. One big word at the strongest moment lands. Behind every line, it buries the words.
• Leave most of it visible. A word reads best when the head covers 30 to 55 percent of it.
• One person, a steady shot. Two speakers or hard cuts make the cutout flicker; split the clip first.
• The first cutout downloads a model of about 168 MB, then works offline.

Above: a captions film from the HyperFrames homepage, with words set in front of and behind the speaker.

Does the transcription run on your computer?

Yes, with your own agent. HyperFrames transcribes with whisper.cpp, an open-source speech model that runs on your machine, or with Parakeet when you install it. The video is not uploaded to a captions service.

• English: pass --model small.en. On our 20-second clip it took about 2 seconds. The first run downloads the model, about 490 MB.
• Other languages: pass --model small with --language and the code, such as es or ja. The default small.en model translates other languages into English.
• Parakeet covers English and 25 European languages: npx hyperframes models install parakeet.
• Speech over music: start with --model medium.en.
• HeyGen's speech model is an option with --engine heygen. It needs a HeyGen API key and uses API credits.

Read the transcript before you caption. Fix names and product terms, then caption the final cut, not the raw take, so the words stay on time.

How does HyperFrames compare with CapCut, Captions and Submagic?

Caption apps are quicker to start with a phone and no setup. HyperFrames wins when you want an agent to do the work, captions that are yours as code, no upload and no monthly limit. Prices checked on each site on 6 November 2026.

• Choose CapCut if you already edit on your phone and want a free start with many presets.
• Choose Captions if you want a phone app with polished templates and other AI tools in one place.
• Choose Submagic if you post many short clips and want a web app that does it with no setup.
• Choose HyperFrames if you work with an AI agent, want the captions as editable code, or need the video to stay on your computer.

What goes wrong with AI captions, and how do you fix it?

• Captions drift off the words. They follow the clip as it is. Cut the pauses first, then caption the cut.
• A name is spelled wrong. Our transcript wrote "hyperframes" in lower case. Tell the agent how names are spelled, and it fixes them in the captions.
• Captions cover the face. On 9:16 video, keep words below the chin and inside the area the app's buttons don't cover.
• The clip already has burned-in text. A second caption layer clashes with it. Pick a clip without it.
• Too many effects. Highlight key words, not every word. One style per video reads as designed; our five-style take is a side-by-side demo.

Related guides: kinetic typography, the best free AI video editor, AI motion graphics, Can ChatGPT make videos?.

Questions people ask

Can I correct transcription errors?

Yes. Ask the agent for the right word, for example “her name is spelled Siobhán”, and it changes that word without touching the timing or the footage. To bring in a corrected transcript file instead, run npx hyperframes transcribe transcript.json in the project.

Can I change fonts and highlight colors?

Yes. Ask for a font, a highlight color or a whole style by name, and keep several styles on one clip to switch between them. To use your own font, add the font file to the project or to a brand kit.

Are captions burned in, exportable, or both?

Both. The exported video has the captions burned in. For a separate caption file, run npx hyperframes transcribe video.mp4 --to srt (or --to vtt).

Is HyperFrames free for captions?

Yes. HyperFrames is free and open source under Apache 2.0. Transcription, caption styles, the cutout and the render run on your computer at no cost and add no watermark. The Studio app's free plan includes captions. Your AI agent may need its own paid plan.

Can HyperFrames do word-by-word or karaoke captions?

Yes. caption-highlight boxes each word as it is spoken, and caption-pill-karaoke fills each word with color in time with the voice. Both use word-level timings from the transcript, so every word lands on its sound.

How do I make TikTok captions with HyperFrames?

Ask for a 9:16 video and a social style such as caption-highlight or caption-kinetic-slam. Keep the words below the face and above the bottom of the frame, where the app's buttons and caption sit.

Does it work in languages other than English?

Yes. Transcribe with --model small and --language set to the language code, or install Parakeet for English and 25 European languages. The Studio app detects the language for you.

Does HyperFrames change my original video?

No. The captions are layers on top of the footage, and the render writes a new MP4. The original file stays as it was.