Choosing a font for captions in Indian scripts

A typeface only contains the characters its designers drew. Anton and Bebas Neue, two Latin display faces used in AutoCap's own caption styles, have no Devanagari, Tamil, Bengali or Urdu in them. That is scope, not quality: Devanagari alone needs hundreds of extra shapes for its conjuncts and vowel signs, a separate design job that many Latin families never had done.

When a caption asks for a letter its font lacks, the text engine tries the next font in the style's list, then the fonts on the device, and borrows the first that has it. Only if nothing can draw the letter does it become an empty box, often called tofu.

That rescue keeps words legible but loses the style: the stand-in may be lighter, narrower or sit at a different height, and each operating system can pick a different one. Burned-in captions avoid that last problem, because the export draws them once for every viewer.

Fonts are chosen character by character, not word by word. In ये font बहुत clean लगता है, the letters of font and clean are found in the Latin face, and the engine keeps going down the list for the Devanagari ones, so a single line ends up in two typefaces.

Every built-in AutoCap style keeps a fallback face further down its font list for each script AutoCap transcribes: members of the Baloo 2 family for Devanagari, Gujarati, Gurmukhi, Tamil, Telugu, Kannada, Malayalam and Odia, Noto Sans Bengali for Bengali, and Noto Nastaliq Urdu for Urdu.

Under Fonts, Native script font swaps that fallback for a face you pick, used for Indian-script and Urdu letters while English words keep the Family face. A font you upload yourself brings no fallbacks with it, so when you use one on a native-script project, set Native script font too.

Judge a candidate on real lines from your own transcript, at the size a phone will show them:

AutoCap's font list has 60 families for Devanagari, which Hindi and Marathi share, and 26 for Telugu. Gujarati has 21, Tamil 16, Bengali, Kannada and Urdu 10 each, Gurmukhi 9, Malayalam 8 and Odia 5.

With sixty families the work is narrowing down; with five, it is finding one heavy enough. Typing a script's name, such as Tamil, into the font search brings up its faces, and each language's templates page opens the gallery on that language: Hindi, Bengali, Gujarati, Punjabi, Tamil, Telugu, Kannada and Odia.

Urdu is a separate case. It runs right to left, its letters join and change shape with their neighbours, and it is traditionally set in Nastaliq, where each joined group slopes down towards the baseline. Nastaliq needs generous line height, and its dots matter: they alone tell ب, ت and پ apart.

AutoCap styles fall back to Noto Nastaliq Urdu, whose heaviest weight is 700, so a heavier style's Urdu is drawn at 700. Several Urdu faces in the list, Noto Naskh Arabic and Noto Kufi Arabic among them, are Naskh or Kufi designs, which readers used to Nastaliq may find look like Arabic rather than Urdu.

Any caption containing Urdu is laid out right to left, English words included, and letter-by-letter entrances bring each Urdu word in whole, since joined letters can't be split. Skip extra letter spacing, and keep outlines thin enough for the dots to stay separate.

SRT and VTT files hold only words and timings, so the app or player showing them picks the font, which for native-script subtitles can mean a system fallback again. When the typeface matters, burn the captions in; SRT, VTT or burned-in captions compares the three.

The browser, not AutoCap, joins letters and places marks. AutoCap loads every face a caption needs before an export draws its first frame, but it can't judge whether a face is well made, and a character missing from your chosen family comes from whichever font further down the list has it. Watch the export on a phone, and if you don't read the script, ask someone who does.

No typeface makes a script readable for viewers who don't read it. If your audience reads Hindi in Latin letters, see native script, romanised or English.