For Hindi Creators
Hindi captions that are
spelled right
Paste a Hindi video and get vertical clips with word-by-word captions that a Hindi reader can actually read. Most short-form tools burn subtitles through a default that cannot reorder Indic vowel signs, so दिन comes out दनि and क्योंकि falls apart — on roughly every other line. Nova shapes the text properly before it ever touches a frame.
Free tier included — no credit card required
Try it on your video — no signup
Paste a URL, see your viral moments
The matras land on the correct side
In Devanagari the short-i vowel sign is stored after its consonant and drawn before it. That reordering needs HarfBuzz, and the subtitle filter most video tools reach for runs without it — so the mark ends up after the letter and the word becomes a non-word. Nova renders with full complex shaping: दिन stays दिन.
Not just Hindi
The same reordering breaks Bengali, Tamil, Telugu, Kannada, Marathi, Thai and Khmer. Nova renders all of them through the same shaped path, and Arabic, Hebrew, Japanese, Chinese and Korean alongside them — checked on rendered pixels, not on whether a font happens to be installed.
Captions cut at phrases, not at a timer
Lines break where the sentence breaks and are sized to the width the frame actually has, so a caption is a readable phrase rather than three words chopped mid-thought. Hindi, English or anything else.
It finds the moments too
Nova reads the transcript to pick the strongest 25-50 second moments, reframes them vertical, writes the hook and the title, and burns the captions — the whole clip, not just the subtitle track.
From upload to posted in three steps
1.Upload or paste
Drop a file or paste a link. Nova handles download and conversion.
2.AI finds the moments
Transcription + moment detection rank and cut your most viral segments.
3.Download & post
Vertical, captioned clips ready for TikTok, Reels, and Shorts.
Ready for every platform
Questions? Answers.
Why do my Hindi captions look misspelled in other tools?
Almost certainly the subtitle burn, not the transcription. FFmpeg's `subtitles` filter — the standard way to burn captions and the one most tools are built on — exposes no text-shaping option, so the renderer falls back to simple shaping: bidirectional reordering and Arabic joining, but no HarfBuzz. Indic scripts need HarfBuzz, because a pre-base vowel sign like ि is stored after its consonant and has to be drawn before it. Without that step it is drawn where it was stored, and दिया becomes दयिा. You can check your own output: look for any word with a short-i and see which side the mark is on.
How much of a Hindi video is actually affected?
More than it sounds. Measured across the Hindi transcripts in Nova's own system, 9% of words carry that vowel sign — including क्योंकि, लिए, दिन and किया, which are among the most common words in the language. At around six words to a caption line, that works out to roughly every other line on screen carrying a visibly wrong word.
Why has nobody noticed this?
Because Arabic still looks fine. Arabic joining is handled by the simpler path, so the one non-Latin script a Western-built tool is most likely to spot-check comes out correct — and Latin scripts need no reordering at all, so nothing looks wrong to the people building or reviewing it. It is only visible to someone who reads the script.
Is this a known problem, or just your opinion?
It is known and it is public. It was reported to the libass project in February 2023 as issue #682 — 'ffmpeg always uses simple shaping with -vf subtitles' — and closed as 'not our bug', because the shaping option belongs to FFmpeg's filter rather than to the renderer underneath it. It has sat that way ever since, with no documented workaround. Any tool that burns captions through that filter inherits it, whether or not anyone there has looked at a Devanagari frame.
Isn't this just a missing font?
No, and that is the part that keeps it alive. The usual advice for Indic captions is to install a font with the right glyph coverage — Noto Sans is the common recommendation — and that advice is necessary and completely insufficient. Nova was already rendering with Noto and the glyphs were all present and correctly drawn; they were in the wrong ORDER. A font supplies shapes. Deciding that a vowel sign stored after its consonant must be painted before it is a separate step, and it is the step that was being skipped.
Does Nova transcribe Hindi, or only caption it?
Both. Speech recognition handles Hindi natively, and you can force the language if a code-switched video is being detected as English. The captions are word-timed, so the karaoke highlight follows the speech rather than sitting under it as a static block.
Can I get the clips in English too?
Yes — Nova translates captions, titles and descriptions into 90+ languages and can dub the voiceover, so one Hindi recording can go out to an English audience as well. See the translation page for how that works.
More ways to use Nova
Guides & resources