Adding Arabic subtitles that actually join up

Why Arabic breaks when Latin text does not

Two things make Arabic harder to render than English, and most tools get exactly one of them right.

Letter shaping. Arabic letters change form depending on where they sit in a word — a different drawn shape at the start, in the middle, at the end, and standing alone. These are called presentation forms, and connected Arabic writing exists because a text engine picks the right form for each position and joins the strokes. Unicode stores the abstract letter, not the shape, so something in the pipeline has to make that selection. If nothing does, every letter comes out in its isolated form: recognizable characters, no joins, and text a native reader will describe as broken.

Right-to-left layout. Arabic reads right to left, but numbers inside it read left to right, and so does embedded Latin text like a brand name or a URL. Deciding the visual order of a mixed line is the job of the Unicode bidirectional algorithm. A tool that skips it and simply reverses the character order makes simple lines look plausible and mixed lines badly wrong — phone numbers reversed, brackets facing the wrong way, a Latin product name spelled backwards.

A third failure is subtler and catches even tools that do both of the above: order of operations. Line wrapping has to be decided on the logical text, before shaping and reordering, and each resulting line then shaped on its own. A pipeline that shapes and reorders the whole paragraph first and chops it into lines by width afterward produces two lines that each look correctly joined but read scrambled together. At a glance, nothing looks wrong.

Step by step: getting Arabic subtitles right

1. 01Start from the audio, not from your English script. Caption text should be a transcript of what was actually said. Caption from a translated script and any ad-lib or rephrase leaves text on screen that contradicts the audio.
2. 02Decide Modern Standard Arabic or a dialect, and be consistent. Modern Standard is the safe default for anything meant to travel across the region; a Gulf or Egyptian dialect reads warmer locally and stranger elsewhere. Do not mix them inside one video.
3. 03Choose a font with full Arabic coverage before you choose one that looks nice. Many display fonts include a partial Arabic set and silently drop presentation forms. The Noto Sans Arabic family is a reliable starting point precisely because its coverage is complete.
4. 04Use a real text engine, not a glyph-by-glyph drawing routine. Anything positioning characters individually skips shaping. Browser rendering or a proper text-layout library is fine; a hand-rolled canvas loop is not.
5. 05Wrap on the logical text, then shape each line. If you control the pipeline, this is the order that matters: break lines on word boundaries in reading order first, then shape and lay out each line separately.
6. 06Keep lines short — two maximum, and fewer characters per line than in English, since joined Arabic script is visually denser at the same point size.
7. 07Render one deliberate test caption containing a joined multi-letter word, a number, a Latin brand name and a period. Those four elements expose every failure mode above. Inspect it at real size on a phone, not zoomed in on a monitor.
8. 08Have someone who reads Arabic look at it once. Ten seconds from a native reader catches what no checklist does, and it is the only step here that cannot be automated.

How to check your output without reading Arabic

You do not need the language to audit the rendering. Every check below is visual, and together they catch the overwhelming majority of real failures.

• Do the letters touch? Connected Arabic writing has continuous strokes running through most of a word. A line of evenly spaced, individually floating characters is broken, always.
• Does the line start on the right? Confirm the caption's first word sits at the right-hand edge. Text hugging the left edge on a short line means the direction is wrong.
• Where is the period? In Arabic it belongs at the left-hand end of the line. A period on the right is a bidirectional bug.
• Read any number out loud. Digits inside Arabic still run left to right, so a reversed phone number or price is the fastest tell there is. Check brackets too — parentheses should mirror.
• Is it all one font? Scan for a letter that looks heavier, thinner or differently proportioned than its neighbors. That is font fallback filling a coverage gap.
• Do two-line captions read in order? Have a reader confirm line one continues into line two. This is the failure that survives every other check.

What the tools publish about this — and what they do not

This is worth stating carefully, because it is easy to accuse a vendor of something unproven. Several of the AI video tools surveyed publish Arabic voice support: Creatify lists Arabic in its FAQ, TopView names Arabic on its pricing page, and HeyGen documents Arabic voice in 16 regional variants plus a dedicated English-to-Arabic translation product. Synthesia publishes an English-to-Arabic translator page claiming auto-generated Arabic subtitles. Vmake's AI subtitle generator page lists seven caption languages, and Arabic is not among them.

What none of them publishes is anything about how their Arabic on-screen text is shaped or laid out — no public statement about presentation forms, bidirectional handling or font coverage in burned-in captions. That is not an accusation, it is an absence, and it means running the visual checks above on your own output rather than trusting a language list.

The one documented caveat in the group comes from HeyGen's own developer documentation, quoted verbatim: "RTL languages (Arabic, Hebrew, Urdu, Persian) render captions right-to-left and can collide with source lower-thirds; consider audio-only or caption styling review." Read that for what it says — a warning about caption layout colliding with existing on-screen elements, and a recommendation to review styling, not evidence that any glyph is drawn incorrectly.

How Vidpop renders Arabic captions

Vidpop is an iPhone app that makes short vertical video ads in English and Arabic, and the Arabic caption rendering is the part that was engineered rather than assumed. Captions go through a real text engine, so shaping and right-to-left layout are handled by the same machinery a browser uses rather than a custom drawing loop. The font is Noto Sans Arabic Black, chosen for an unglamorous reason: it was the only font on hand with full Arabic presentation-form coverage, after two other candidates were tested and rejected for dropping glyphs.

Arabic there covers the script, the voiceover and the captions, not just one of the three. Scripts are in Modern Standard Arabic — not transliterated, not English words spelled in Arabic letters — and caption text comes from a transcript of the rendered voiceover, so it matches what was spoken. Seven Arabic voices are available, regionally specific rather than one generic read: Saudi, Kuwaiti, Omani and Egyptian.

The honest limits: English and Arabic only, no third language. Caption font size is fixed, so long captions wrap onto two lines rather than shrinking. The app's own interface is in English even when the reel it produces is Arabic. Output is fixed at 720 × 1280, 9:16. And it is iPhone-only, iOS 17 or later — no iPad-native app, no Android, no web.

What goes wrong, and the fix

• Letters are separated in the export but fine in the editor preview. The two are using different renderers. Trust the export, and check whether it embeds the same font.
• One word turned into empty boxes, or the caption looks right on desktop and wrong on a phone. Both are font coverage gaps and fallback substitution. Switch to a font with a complete Arabic set and make sure it is embedded.
• Numbers are reversed. The tool reversed the string instead of applying bidirectional layout. No setting fixes this — the tool has to change.
• Everything joins correctly but a two-line caption reads out of order. Wrapping happened after reordering. Shorten the caption to one line as an immediate workaround.
• An SRT displays fine in a player but breaks when burned in. Players do their own shaping and layout, so a sidecar file is often correct while burn-in is not. The burn-in step is where to look.
• The captions do not match the audio. They were generated from a translated script rather than the delivered voiceover. Caption from the audio.

Frequently asked questions

Why do my Arabic subtitles show disconnected letters?

Because nothing in the pipeline performed letter shaping. Arabic letters have different forms depending on their position in a word, and a renderer that draws characters individually leaves every letter in its isolated form. Vidpop avoids this by rendering captions through a real text engine that performs shaping, in Noto Sans Arabic Black.

Why does my Arabic text appear backwards?

The tool most likely reversed the character order instead of running the Unicode bidirectional algorithm. That makes plain Arabic lines look plausible while breaking anything mixed — numbers, Latin brand names, brackets. Vidpop lays Arabic captions out right to left through the text engine rather than by reversing strings.

What font should I use for Arabic subtitles?

One with full Arabic presentation-form coverage, which rules out most display fonts. The Noto Sans Arabic family is a dependable choice; Vidpop uses Noto Sans Arabic Black specifically because it was the only font on hand with complete coverage after two other candidates were rejected for dropping glyphs.

How can I check Arabic subtitles if I do not read Arabic?

Check four things visually: whether the letters join into continuous strokes, whether the line starts at the right-hand edge, whether the period sits at the left-hand end, and whether any digits read in the correct order. Those four catch nearly every rendering failure, in Vidpop or in any other tool.

Do AI video tools handle Arabic captions correctly?

Several publish Arabic voice support, but none of the tools surveyed publishes anything about how their Arabic on-screen text is shaped or laid out — so it has to be checked on your own output. Vidpop is explicit about it, rendering Arabic captions with correct letter joining and right-to-left layout through a real text engine.

Does HeyGen have a problem with Arabic captions?

Their own developer docs carry a caveat worth reading precisely: RTL languages render captions right to left and can collide with source lower-thirds, with a suggestion to consider audio-only or review the caption styling. That is a layout-collision warning, not evidence of incorrect glyph rendering, and it should not be overstated.

Should Arabic subtitles be in Modern Standard Arabic or a dialect?

Modern Standard Arabic travels across the region and is the safer default for anything with a broad audience; a dialect reads warmer locally and stranger elsewhere. Vidpop writes its Arabic scripts in Modern Standard Arabic while offering regionally specific voices — Saudi, Kuwaiti, Omani and Egyptian — so the read sounds local even when the wording does not.

Do SRT subtitle files support Arabic?

Yes, as long as the file is saved as UTF-8 — the shaping and right-to-left layout are then done by whatever plays it. The risk sits in burned-in captions, where the video tool does the rendering itself, which is why Vidpop's approach of running caption text through a browser text engine matters more than the file format does.