Frequently asked questions
Everything worth knowing to watch streams and videos with subtitles in your language: what it opens, what device it needs, how to use it and what to do when something is off.
How it works
What is subany.video?
A website that shows you live streams and videos with automatic subtitles in your language. Paste a link, pick the languages and press Watch. It is free, needs no account and there is nothing to install.
Where do speech recognition and translation happen?
In your browser. A speech recognition model (Whisper) turns what is said into text and a translator brings it into your language: Chrome’s built-in one when available, or compact translation models downloaded once. Your audio never leaves your device.
Why does the video run a few seconds behind?
So that each subtitle is ready when the line is spoken. On live streams, picture and subtitles run together about 15 seconds behind real time. You can change that: see how to change the delay.
Which languages can the subtitles be in?
English, Spanish, French, German, Portuguese, Italian, Japanese, Korean, Russian and Chinese. The original language is set in the “From … to …” selector on the home page; by default it is detected automatically (in beta).
What if I want to translate a book?
That is what subany.site is for: open a book or a document and read it translated, with the original alongside, also in your browser.
Supported platforms
This is what can be opened today. Each platform has its own page with the link to paste and its limits.
| Platform | Live | VOD | Clips | Page |
|---|---|---|---|---|
| Twitch | Yes | Yes | Yes | Twitch live subtitles · Twitch VODs and clips |
| Kick | Yes | Yes | Yes | Kick live subtitles · Kick clips |
| CHZZK | Yes | Yes | No | Subtitles for CHZZK |
| PeerTube | Yes | Yes | No | Subtitles for PeerTube |
| Picarto | Yes | No | No | Subtitles for Picarto |
| Owncast | Yes | No | No | Subtitles for Owncast |
| Rumble | No | Yes | No | Subtitles for Rumble |
| Odysee | No | Yes | No | Subtitles for Odysee |
| Wikimedia Commons | No | Yes | No | Subtitles for Wikimedia Commons |
| Podcasts (RSS or audio) | No | Yes | No | Subtitles for podcasts |
| .m3u8 links (HTTPS) | Yes | No | No | Translate a stream or video |
| File on your computer | No | Yes | No | Subtitle a video file |
What if my platform is not listed?
For now only the ones in the table work. If you paste another link, the home page tells you which ones do. If you have the .m3u8 address of a stream served over HTTPS, you can try that.
Can I open private or subscriber-only content?
No. subany.video never signs in to any platform, so it only opens what is public.
Can I subtitle a video stored on my computer?
Yes: MP4 or MOV videos and MP3, WAV, M4A, AAC, OGG, Opus or FLAC audio. The file is not uploaded. More in Subtitle a video file.
Requirements and performance
subany.video does all the work inside your computer. Here is, in plain words, what device you need, what to expect from yours and how to check it in a minute.
What gets downloaded and how much space it takes
Speech recognition and translation run in your browser. To do this, the browser downloads some models, which are ready-trained programs that understand speech and translate. They are downloaded once and kept in the browser, so later visits start much sooner. They take roughly 210 MB at the low level, 580 MB at the medium level and 1.7 GB at the high level. Each translation language pair adds 120 to 230 MB, and the sound labels add 16 MB.
WebGPU or WASM: the graphics card or just the processor
WebGPU is a browser feature that lets it use your graphics card for calculations. It is faster for these models and lets the site use the largest ones, which understand speech best. WASM is the fallback: it uses only the processor, is slower and needs smaller models, but works on almost any device.
Recent Chrome, Edge, Brave and Opera include WebGPU, and so do some recent versions of Safari. Firefox is always treated in basic mode (WASM), because it runs better that way.
You do not have to choose anything: when you open a video, the site runs a short test of your graphics and decides by itself. It repeats the test every 30 days or when the graphics change. With no WebGPU, with slow graphics or if the browser reports less than 8 GB of memory, it uses WASM.
Three levels depending on your device
The site picks the level by itself; if you want, you can set another one in ⚙ More settings › Advanced › Processing quality. The example devices and RAM are only a guide: we have not measured each one.
| Level | Example device | Speech model | Quality and delay | Voices and sounds | A 100,000-word book |
|---|---|---|---|---|---|
| Low | Laptop or desktop with Intel UHD or HD integrated graphics, or any device without WebGPU (such as Firefox). About 4 to 8 GB of RAM. | whisper-base (about 170 MB), on the processor (WASM). | Basic quality: good enough to follow the conversation, with more mistakes. Fixed delay of 15 s (you can set 10 or 20). With recorded videos, a 30 s wait before starting; the site waits up to 60 s for the first subtitle. | Does not separate voices. Sound labels: yes, if you turn them on in ⚙. | About 97 min or more with its own translator; about 10 min with Chrome’s built-in one. |
| Medium | Recent laptop with modern integrated graphics (for example AMD Radeon) or a modest dedicated card, with WebGPU. About 8 to 16 GB of RAM. | whisper-small (about 410 MB), on the graphics card (WebGPU). The first subtitle comes from whisper-tiny while it loads. | Good for following a live stream. Measured: 98% of subtitles on time with a 15 s delay. Video is requested at 720p so it does not compete with processing. With recorded videos, a 15 s wait before starting. | Separates voices and gives each one a colour (up to 3 at a time). Sound labels: yes. | About 97 min with its own translator (it uses the processor); about 10 min with Chrome’s built-in one. |
| High | Computer with a recent dedicated graphics card and WebGPU. 16 GB of RAM or more. | whisper-large-v3-turbo (about 1.6 GB), on the graphics card (WebGPU). | The best recognition quality: it is the largest model. We have not measured its timing yet. Fixed delay of 15 s by default and a 15 s wait with recorded videos. | Separates voices and gives each one a colour. Sound labels: yes. | About 97 min with its own translator (it uses the processor); about 10 min with Chrome’s built-in one. |
About books: translation does not use the graphics card, so the level barely changes the time. With the built-in translator of recent desktop Chrome you get about 9,900 words per minute; with the site’s own translator, about 1,000. We measured this on a PC with 16 GB and 16 threads, English to Spanish, with the tab in view; slower devices will take longer (not measured). It is machine translation: good for understanding, not for publishing.
Tips to keep it running well
The first time takes longer: the models are downloaded (from about 200 MB to 1.7 GB depending on the level, and more with translation). In a cold test at the medium level, the first subtitle came after 45 or 52 seconds; it depends mostly on your connection. After that they stay saved.
On a laptop, better plugged in and well ventilated. Processing speech makes the graphics or the processor work hard: on battery it drains fast and the device heats up.
Close heavy tabs and programs (games, video editing, other tabs playing video): they share graphics and memory. At the medium level, Chrome used between 1.6 and 2.5 GB of memory in the tab and about 2 GB on the graphics side.
If it struggles, lower the quality in ⚙ More settings › Advanced › Processing quality (Basic, Balanced, Maximum or Automatic): the device works less and heats up less. If subtitles arrive late, try a 20 s delay in the same section.
Do I need a powerful graphics card?
No. With no WebGPU or with modest graphics it works in basic mode (WASM): only the processor, a smaller model and no voice separation. With recent graphics it runs better.
How do I know which level my device is at?
Press “Check my device”: the diagnostics page tells you the level, whether it runs on WebGPU or WASM, and which model it uses.
How much space do the models take?
About 210 MB at the low level, 580 MB at the medium level and 1.7 GB at the high level, plus 120 to 230 MB for each translation language pair. They are kept in the browser.
Does it work on a phone?
It opens, but it is not designed for phones: the same rules apply as on a computer and it may stay at the low level. If you can, use a computer with Chrome, Edge, Brave or Opera.
How to
How to watch a video with subtitles in your language
- Paste the link. Copy the address of the stream or video and paste it into the box on the home page. You can also open a file from your computer.
- Pick the languages. In the “From … to …” selector, leave “Detect” or pick the original language, then the subtitle language.
- Press Watch. The first time, the models are downloaded and it takes longer; after that it starts much sooner.
- Adjust to taste. Aa changes the text size. ⚙ More settings has the background, the original text, voices, sounds and the delay.
How do I download the subtitles as .srt?
On VODs, clips, podcast episodes and files from your computer: open ⚙ More settings and, under Download subtitles, choose .srt or .vtt, or copy the full transcript. You get what has been generated so far, so let the video play on. Live streams have no download.
How does language detection work?
With “Detect (beta)”, the original language is recognised from the first sentences and the player shows “Detected: …”. If it is wrong, press that notice and pick the right language, or change it in ⚙ More settings › Original language. When the platform says what language the video is in, that is used as a hint.
How do I give each voice a colour?
In ⚙ More settings › Voices, turn on “Color-code voices”. Each person speaking gets a colour and keeps it, and you can give them a name. It needs WebGPU (medium and high tiers); on the medium tier it tells apart up to 3 voices at a time. Without WebGPU, subtitles appear in a single colour.
What are sound labels?
Notes such as [music] or [applause] when those sounds play without clear speech. Turn them on in ⚙ More settings › Sounds; they are off by default and the first time they download a model of about 16 MB. For now they do not recognise laughter or other sounds.
Can I change the playback speed?
Yes, on recorded videos, clips, episodes and files: the speed button goes from 0.5× to 2×. Not on live streams, and not on Twitch VODs, because their player does not allow it.
Can I change the live delay?
Yes, in ⚙ More settings › Advanced › Delay: 10, 15 or 20 seconds. With less delay, subtitles may arrive late on modest devices; with more, they have more room. On recorded videos, the first subtitles are prepared before playback starts; press “Start now” if you do not want to wait.
How do I change the quality?
There are two settings in ⚙ More settings › Advanced. “Processing quality” picks the speech model: Automatic (your device decides), Basic, Balanced or Maximum. “Video quality” picks the picture resolution. What to expect from each tier is in Requirements and performance.
Can I watch it on my TV?
Yes: in the browser menu, choose Cast and pick “Cast tab”. The quality depends on your device.
Common problems
No subtitles appear
Check, in order: that someone is actually speaking (music or silence gives nothing to transcribe); that the original language is right (look at the “Detected: …” notice); and that there is no notice saying the audio could not be read.
The first time, subtitles do not appear until the models finish downloading. If it still does not work, open ⚙ More settings › Diagnostics.
It keeps loading
The first time, the models are downloaded, from about 200 MB to 1.7 GB depending on your device, and you will see the progress. On a slow connection it can take several minutes; after that they stay saved. On recorded videos, the first 15 to 30 seconds are prepared before playback: press “Start now” if you do not want to wait. If nothing moves, reload the page.
I use a content blocker and something does not load
Some content-blocking extensions stop the Twitch player for VODs, or the model downloads, from loading. Turn it off for subany.video and reload the page.
Why is it worse in Firefox?
Firefox always runs in basic mode: it uses only the processor (WASM) and the small model (whisper-base), does not separate voices and has no built-in translator. It works, but Chrome, Edge, Brave or Opera give better results.
Does it work on a phone?
It opens, but it is not designed for phones: with little memory or no WebGPU it stays on the low tier, which is tight and drains the battery. On iPhone, the picture also cannot go full screen and the subtitles show on the page. Use a computer if you can.
It says the audio or video cannot be read
The platform or server does not let the browser read that content, or it is in a format that cannot be opened. It happens with some podcasts, part of the Odysee catalogue and, at times, Twitch clips. If the video keeps playing, it will do so without subtitles. Try later or with another link.
It says the channel is not live
The channel is not streaming right now. Channels only open while they are live; for a past broadcast, paste the VOD link if the platform has one.
Accessibility
subany.video wants more people to be able to follow what is said in a video. This is what exists today and what is still missing.
What does it offer today?
The Atkinson Hyperlegible typeface, designed to be easy to read, with a dark background behind each line; three sizes; a colour and a name per voice; [music] and [applause] labels; the original under the translation; and, on recorded videos, the full transcript to read separately. The player works with the keyboard (space or K to pause, M to mute, F for full screen and the arrows for volume) and the site respects the reduced-motion setting.
What does it not do yet?
Subtitles are automatic and make mistakes. They do not describe other sounds (laughter, doors, tone of voice). Screen readers do not announce the subtitles as they appear. Colours per voice need a graphics card with WebGPU. There is no browser extension and no sign language, and on live streams everything runs about 15 seconds behind.
Is it useful for deaf people?
For many, yes: it lets you follow streams that have no captions. There is more detail in Captions for deaf and hard-of-hearing viewers.
Privacy and price
Are my audio or video uploaded?
No. Speech recognition and translation happen in your browser. Files from your computer are read from your disk and are not uploaded.
Do I need an account?
No, neither on subany.video nor on the platforms.
What is stored in my browser?
The downloaded models, your settings (size, languages, delay…) and recent channels. You can clear all of it from the site data in your browser.
What does the diagnostics report contain?
Technical data about your device and the session. It contains no channel, links, IP addresses or keys, and it only leaves your device if you copy and share it.
How much does it cost?
Nothing. subany.video is free, with no subscription and no account.