# Tamil captions that read correctly

> Tamil is the worst-affected script we measured: vowel signs like ெ ே ை are drawn before the consonant they follow, and most caption burners skip that step. Nova shapes Tamil properly.

*For Tamil Creators*

Canonical: https://novaclipper.com/for/tamil-video-clips

Paste a Tamil video and get vertical clips whose captions a Tamil reader can actually read. Of the twelve writing systems we tested, Tamil is the one most damaged by the shaping shortcut most video tools take — because ெ, ே and ை are all stored after their consonant and must be drawn before it.

## What you get

### The worst-affected script we measured

We rendered the same sentence with and without proper shaping across twelve writing systems and compared the pixels. Tamil diverged the most of any of them — 19 times more than a Latin control line, which shifts only by kerning. Every pre-base vowel sign in the sentence lands in the wrong place without HarfBuzz.

### Not a font problem

The usual advice is to install a font with Tamil coverage. Necessary, and completely insufficient: the glyphs are present and correctly drawn, they are simply in the wrong order. A font supplies shapes; deciding that ெ is painted before the letter it follows is a separate step.

### It clips, reframes and captions

Nova reads the transcript to find the strongest 25-50 second moments, crops them vertical, writes the hook, and burns word-timed captions — the whole clip, not just a subtitle track.

## FAQ

### Why do Tamil captions come out wrong in other tools?

FFmpeg's `subtitles` filter — the standard way to burn captions and the one most tooling is built on — exposes no text-shaping option, so the renderer falls back to simple shaping: bidirectional reordering and Arabic joining, but no HarfBuzz. Tamil needs HarfBuzz, because ெ ே ை are pre-base vowel signs stored after their consonant and drawn before it. It was reported to libass as issue #682 in February 2023, closed as 'not our bug' since the option belongs to FFmpeg, and has never been fixed. Check your own output: find a word with ெ and see which side of the letter it sits on.

### How bad is it really?

On our measurement Tamil showed the largest divergence of any script we tested — 0.740 against a Latin control of 0.038. Latin moves only by kerning; Tamil moves by whole glyphs. It is not a subtle rendering difference, it is words spelled wrong.

### Does Nova transcribe Tamil?

Yes, natively, and you can force the language if a code-switched video is being misdetected as English. Captions are word-timed so the highlight follows the speech.

---

Nova turns long-form video into vertical short-form clips with AI moment detection, face-aware 9:16 reframing, and word-by-word captions — and makes faceless videos from a script or a Reddit thread. Free tier: 30 minutes/month, no card. [Start free](https://novaclipper.com/signup) · [All use cases](https://novaclipper.com/for) · [Free tools](https://novaclipper.com/tools)