Inky Clips

Caption languages: 99 recognised, 68 drawn

Inky Clips listens to your video with Whisper, an open speech-recognition model, running inside your browser. Pick the language that is spoken, and the captions write themselves onto the video word by word, in time with the voice. Nothing is uploaded.

Start for freeFree, with your Google account.
  • Sveiki no orbītas
  • Cześć z orbity
  • Hola desde la órbita
  • Päikest orbiidilt
  • Grüße aus dem Orbit
One greeting in five languages, each written with its own letters.

How auto captions work

  1. Pick the spoken language

    In the editor, click the speech bubble and choose the language of the video. The one you used last waits at the top.

  2. Download the model once

    The first time, your browser fetches Whisper large-v3-turbo: about 1.6 GB on computers with WebGPU, about 1.1 GB otherwise. After that it is reused.

  3. Let it listen

    Speech is recognised on your computer, with a timestamp for every word, then split into short phrases at pauses and sentence ends.

  4. Watch the words write

    Each word draws itself onto the video as it is spoken, stroke by stroke, like a pen keeping up with the voice.

  5. Fix and restyle

    Click a caption to correct its words. Long-press one on the timeline to change the font, colour, size and position of all of them at once. Then export an MP4.

Read about it in your language

Pages written in Latvian, Lithuanian, Polish and Estonian, with the details that matter for each.

  • Automātiski subtitri video latviešu valodā

    With ā, č, ē, ģ, ī, ķ, ļ, ņ, š, ū and ž, and the ģ comma drawn above the letter.

    Lasīt latviski

  • Automatiniai lietuviški subtitrai jūsų vaizdo įrašams

    With ą, č, ę, ė, į, š, ų, ū and ž in every font.

    Skaityti lietuviškai

  • Automatyczne napisy do filmu po polsku

    With ą, ć, ę, ń, ó, ś, ź and ż everywhere, and ł in the caption fonts.

    Czytaj po polsku

  • Automaatsed subtiitrid videole eesti keeles

    With õ, ä, ö, ü, š and ž in every font.

    Loe eesti keeles

Every language it can caption

Whisper recognises 99 spoken languages, listed here by their English name as in the editor, with each language’s own name beside it. Whether the captions can be drawn depends on the letters, so they are grouped by which fonts can write them.

Latin script (55)

All 23 fonts draw letters with accents, carons, ogoneks and umlauts: where a handwriting font lacks one, Inky Clips writes the letter and then adds its mark. A few letters of their own are still missing from some handwriting fonts: German ß, Danish and Norwegian æ and ø, Icelandic þ and ð, Croatian đ, Turkish ı, Polish ł and Maltese ħ. The default caption font, Relief, has all of these, and Montserrat also covers Azerbaijani ə and the rarer letters of languages like Yoruba and Hausa.

  • AfrikaansAfrikaans
  • AlbanianShqip
  • AzerbaijaniAzərbaycan
  • BasqueEuskara
  • BosnianBosanski
  • BretonBrezhoneg
  • CatalanCatalà
  • CroatianHrvatski
  • CzechČeština
  • DanishDansk
  • DutchNederlands
  • EnglishEnglish
  • EstonianEesti
  • FaroeseFøroyskt
  • FilipinoTagalog
  • FinnishSuomi
  • FrenchFrançais
  • GalicianGalego
  • GermanDeutsch
  • Haitian CreoleKreyòl ayisyen
  • HausaHausa
  • HawaiianʻŌlelo Hawaiʻi
  • HungarianMagyar
  • IcelandicÍslenska
  • IndonesianBahasa Indonesia
  • ItalianItaliano
  • JavaneseBasa Jawa
  • LatinLatina
  • LatvianLatviešu
  • LingalaLingála
  • LithuanianLietuvių
  • LuxembourgishLëtzebuergesch
  • MalagasyMalagasy
  • MalayBahasa Melayu
  • MalteseMalti
  • MāoriMāori
  • NorwegianNorsk
  • Norwegian NynorskNorsk nynorsk
  • OccitanOccitan
  • PolishPolski
  • PortuguesePortuguês
  • RomanianRomână
  • ShonaChiShona
  • SlovakSlovenčina
  • SlovenianSlovenščina
  • SomaliSoomaali
  • SpanishEspañol
  • SundaneseBasa Sunda
  • SwahiliKiswahili
  • SwedishSvenska
  • TurkishTürkçe
  • TurkmenTürkmen dili
  • UzbekO‘zbek
  • WelshCymraeg
  • YorubaÈdè Yorùbá

Cyrillic, Greek and Vietnamese (13)

These draw in the bold caption fonts: Cyrillic and Vietnamese in TikTok Sans and Montserrat, Greek in TikTok Sans.

  • BashkirБашҡорт
  • BelarusianБеларуская
  • BulgarianБългарски
  • GreekΕλληνικά
  • KazakhҚазақ тілі
  • MacedonianМакедонски
  • MongolianМонгол
  • RussianРусский
  • SerbianСрпски
  • TajikТоҷикӣ
  • TatarТатар
  • UkrainianУкраїнська
  • VietnameseTiếng Việt

Recognised, not drawn yet (31)

Whisper understands these, but none of the fonts can write their scripts yet, so the captions would come out blank.

  • Amharicአማርኛ
  • Arabicالعربية
  • ArmenianՀայերեն
  • Assameseঅসমীয়া
  • Banglaবাংলা
  • Burmeseမြန်မာ
  • Chinese中文
  • GeorgianᲥართული
  • Gujaratiગુજરાતી
  • Hebrewעברית
  • Hindiहिन्दी
  • Japanese日本語
  • Kannadaಕನ್ನಡ
  • Khmerខ្មែរ
  • Korean한국어
  • Laoລາວ
  • Malayalamമലയാളം
  • Marathiमराठी
  • Nepaliनेपाली
  • Pashtoپښتو
  • Persianفارسی
  • Punjabiਪੰਜਾਬੀ
  • Sanskritसंस्कृतम्
  • Sindhiسنڌي
  • Sinhalaසිංහල
  • Tamilதமிழ்
  • Teluguతెలుగు
  • Thaiไทย
  • Tibetanབོད་སྐད་
  • Urduاردو
  • Yiddishייִדיש

How accurate it is, honestly

The model is Whisper large-v3-turbo from OpenAI, a faster version of its largest model, running in full on your computer. It is good, and it still makes mistakes. Plan to read your captions through.

Strongest where it heard the most

English, Spanish, German, French, Portuguese, Italian, Polish and other widely spoken languages come out with most words right when the speech is clear.

Smaller languages need more edits

Latvian, Lithuanian, Estonian and many others had far less training audio. Expect more wrong words, especially names, places, slang and dialect.

Clean audio helps most

Record close to the speaker, turn music down under speech, and pick the right language. Then click any caption to fix it before you export.

Your videos stay on your computer.

Most caption tools upload your footage before they can listen to it. Inky Clips runs the speech model inside your browser instead, so interviews, meetings and anything personal stay with you. We keep your sign-in and nothing else.

Start for free

Questions

Which languages can Inky Clips caption?

The Whisper model recognises 99 spoken languages. Captions draw onto the video in 68 of them today: Latin-script languages in all 23 fonts, and Russian, Ukrainian and other Cyrillic languages, Greek and Vietnamese in the TikTok Sans or Montserrat fonts. Scripts such as Chinese, Japanese, Korean, Arabic and Hindi are recognised, but no font can draw them yet.

Is my video uploaded to a server?

No. The speech model runs inside your browser, on your computer. Your video, its audio and the captions never leave it. Our server only handles Google sign-in.

Why is there a download the first time?

That is the speech model itself, Whisper large-v3-turbo. It is about 1.6 GB on computers with WebGPU and about 1.1 GB otherwise. Your browser keeps it, so captions after the first are made straight away.

How accurate are the captions?

Very good for clear speech in widely spoken languages such as English, Spanish, German or Polish. In languages with less training data, such as Latvian, Lithuanian or Estonian, expect to correct more words, especially names. Check the captions before you publish.

Does it detect the language automatically?

No, you pick the spoken language from the list. Telling the model what to listen for gives better results than letting it guess, especially for smaller languages.

Can it translate the captions?

No. Captions are written in the language that is spoken in the video.

Are accented letters drawn correctly?

Yes. The default caption font has them built in, and where a handwriting font lacks one, Inky Clips draws the letter and then its accent, the way a hand would. A few letters with no accent to add, such as ß, æ, ø, đ, ı and the Polish ł, are still missing from some handwriting fonts; the caption fonts have them.

Can I download an SRT file?

Yes. Once the captions are made, choose Export subtitles in the editor to save an SRT, a VTT or a plain text transcript. You can also burn them into the picture and export an MP4.

Is the editor available in other languages?

The editor itself is in English for now. The captions are in whichever language your video is in.