Readable captions: a guide for Shorts and Reels

The captions on a Short or a Reel are not decoration: many people watch short videos without sound, in the street or on public transport, and for anyone who is hard of hearing they are essential. Karaoke captions, which highlight each word as it is spoken, are now a very widespread style in vertical video. This guide covers what decides whether captions get read or get in the way: size, short chunks, contrast, each platform’s safe zones, the highlighted word, language and accessibility. It also tells you what to check when you use automatic captions.

Why captions matter in vertical video

Captions do three jobs at once. They make the video understandable without sound. They make it accessible to people who are deaf or have hearing loss: on the web, the WCAG 2.2 accessibility guidelines call for captions for all the audio in prerecorded videos. And they help viewers follow names, figures and technical terms that are easy to miss by ear.

But badly made captions take something away. If they cover the image, move faster than anyone can read, change style every second or contain errors, viewers pay attention to them instead of to what you are saying.

Size and typeface

Vertical video is watched on a small screen, at arm’s length. So:

• Use a sans-serif typeface with a heavy stroke. Thin or decorative typefaces get lost against the image.
• Go for a large size and two lines at most. A short, large chunk is better than a long, small sentence.
• Be careful with capitals. In chunks of a few words, capitals read well; with longer sentences, mixed upper and lower case usually reads better.
• Keep the same style across the whole channel. Consistency helps people read faster and gives the channel an identity.
• Test on a real phone. What looks huge on a computer monitor may be only just big enough on a phone.

Chunks of two or three words

A very widespread style in short video shows captions in chunks of two or three words that stay on screen long enough to be read. It works because the eye takes in the chunk at a glance and does not have to scan a long line.

For a chunk to make sense, split by units of meaning. Do not separate the article from the noun, a figure from its unit or a verb from its negation: “not / rising” reads worse than “not rising”. Also avoid single-word chunks that flicker: they are tiring and leave no time to read.

If your priority is accessibility, for example in a long or educational video, subtitles for deaf and hard-of-hearing people follow other conventions, with longer lines and reading-speed rules. In Spain, the reference is the UNE 153010 standard on subtitling for deaf and hard-of-hearing people.

Contrast and background

A video’s background changes with every shot, so the text has to be readable over the lightest shot and over the darkest. You have three tools, which you can combine:

• Outline: a dark border around light letters.
• Shadow: a soft shadow that separates the text from the background.
• Box: a semi-opaque rectangle behind the text, the safest option over very busy backgrounds.

As a reference, on the web WCAG asks for a minimum contrast of 4.5:1 between text and its background, and 3:1 for large text. In video it cannot be measured precisely shot by shot, but if your caption clears that bar over the most difficult scenes, it will normally read well in all of them.

Safe zones by platform

Each platform covers parts of the video with its interface: buttons on the right, the account name, text and audio at the bottom, controls at the top. Whatever sits underneath cannot be seen. The exact figures change with every version of the apps and not every platform publishes them, so treat them as approximations:

The rule of thumb is the same for all of them: keep important text in the central band of the screen, above the bottom third and away from the column of buttons on the right. And check the result in each app’s preview before publishing.

The highlighted word: karaoke captions

In karaoke captions, each word changes colour or size at the exact moment it is spoken. Doing this well means knowing when each word starts and ends. Those timings come from the speech synthesiser itself, if the narration is synthetic, or from aligning the audio with the text using speech recognition.

Some guidelines:

• A single highlight colour, consistent with the channel’s identity and with good contrast.
• Few effects. If every word comes in with a different animation, the caption competes with the video.
• Exact sync. A highlight that runs ahead of or behind the voice is noticeable, and it usually bothers people more than having no highlight at all.
• Highlight what matters. Some styles also emphasise figures or keywords; use this sparingly.

Automatic captions save a lot of work, but speech recognition fails precisely on the most delicate parts: proper names, figures, acronyms and technical terms. Always check those words before publishing.

Language and translations

Captions burned into the image go in the language being spoken. To reach other markets, do not put two languages on screen: upload separate caption tracks. YouTube lets you add subtitles in several languages, and its automatic captions are generated from the language spoken in the video. You can also translate the title and the description.

Accessibility: captions for everyone

Burned-in captions have a limitation: viewers cannot enlarge them, change their colour or hide them. That is why it is worth:

• Also uploading a caption file where the platform allows it, as on YouTube.
• Indicating relevant sounds in square brackets, for example [music] or [laughter], when they add information.
• Identifying who is speaking when there are several voices, with a colour or a name.
• Reviewing captions as carefully as the script: an error written on screen is more noticeable than one that is spoken.

Checklist

• Two lines at most, in chunks of two or three words split by units of meaning.
• A heavy typeface that is legible on a real phone.
• Outline, shadow or box: legible over the lightest scene and over the darkest.
• Text inside the safe zone of every platform you publish on.
• Highlighted word in sync with the voice.
• Names, figures and technical terms checked.
• Captions in the language of the audio; translations as separate tracks.
• Caption file uploaded wherever possible.

How CLAQON handles it

CLAQON adds captions to the videos it produces and warns you when something falls outside the safe zone.

• Dynamic captions in the factory. Chunks of two or three words that stay on screen for between 0.6 and 1.2 seconds, a hook card that shows the first sentence from the first frame with each word highlighted as it is spoken, and on-screen labels for the data. You will see it in the Shorts factory.
• Word-by-word timing. It comes from the local voice or from ElevenLabs. If you upload your own narration, it is aligned on your machine with Faster Whisper.
• Safe zones. The summary of each edit warns you if the caption style or an insert falls outside each platform’s safe zone, and the Studio draws that zone over the preview. It uses approximate margins, not official figures.
• Editing Studio (beta). Captions by hand, from an SRT file or by local transcription, with an editable karaoke style, uncertain words underlined and the option to realign them. More in the Editing Studio.
• Subtitle Room (beta). It puts karaoke captions on your own video, with transcription on your machine and no paid AI. You will find it in captions and shorts.
• Caption tracks on YouTube. After publishing, it uploads the original-language track and tracks translated into up to six languages (Spanish, Brazilian Portuguese, French, German, Hindi and Indonesian), and translates the title and the description. Whatever does not fit in the day’s quota is postponed to the next day.