Offline Subtitle Generators in 2026: 7 That Run on Your Machine
Offline subtitle generators compared for 2026: seven tools that transcribe on your own computer, what each one does after the transcript, and what they cost.
An offline subtitle generator runs speech recognition on your own computer, so the video never leaves it. Nearly every good one in 2026 is built on OpenAI's open Whisper models, which means the transcripts start out similar; what separates the tools is everything after the transcript — editing, reading speed, translation, styling and burn-in.
The short version: Subtitle Edit if you want free and thorough, MacWhisper for a quick transcript on a Mac, DaVinci Resolve Studio if you already edit there, and Sablate when the subtitles have to be styled, burned in or delivered in several languages. We make Sablate. The other six were not benchmarked by us, and their details come from their own documentation, checked on 28 September 2026; Subtitle Edit, Buzz and MacWhisper were checked again on 12 November 2026.
| If you need… | On Windows | On a Mac |
|---|---|---|
| Free, thorough subtitle editing | Subtitle Edit | Subtitle Edit |
| A quick transcript or SRT, free | Buzz | MacWhisper (free version) or Buzz |
| Subtitles inside your video edit | DaVinci Resolve Studio or Kdenlive | DaVinci Resolve Studio or Kdenlive |
| Styled, burned-in or multilingual subtitles | Sablate | Sablate |
| A folder of files processed by script | Whisper command line | Whisper command line |
What makes a good offline subtitle generator
Six things decide whether a local tool saves you time or only moves the work somewhere else:
- It makes subtitles, not only a transcript. Most offline transcription software stops at text with timestamps. Subtitles are short lines at a readable pace, broken at phrase boundaries, and reading speed is where raw Whisper output most often fails.
- The engine and the models it runs. Most tools wrap faster-whisper or whisper.cpp. Which Whisper model you can choose matters more than which wrapper runs it, and the model comparison has the numbers; the Whisper GUI comparison shows which app runs which engine.
- GPU support on your hardware. An NVIDIA GPU is the big speed-up on Windows. On Apple Silicon, only tools built on whisper.cpp or Apple's own frameworks use the GPU; faster-whisper runs on the CPU there.
- What happens after transcription. An editor with timing tools, translation, styling and burn-in are the difference between a draft and a deliverable.
- Whether the cloud is really optional. Several local apps also offer cloud engines. That is useful, as long as you always know which one is active.
- How you pay. Free, one-time or subscription, and whether the free version covers the video lengths you actually work with.
Is offline transcription accurate enough now?
For most speech, yes, and the reason is that local and cloud Whisper are the same model.
Whisper's weights are public. Most of the local tools on this list run the same large-v3 and large-v3-turbo that anyone can download, and OpenAI's own Whisper API serves large-v2, an older version than the one you can run at home. For Whisper-quality transcription, a cloud Whisper service has no model advantage — only a hardware one, and that gap has mostly closed. In faster-whisper's published benchmark, 13 minutes of audio took about a minute with large-v2 on an NVIDIA RTX 3070 Ti, and 1 minute 42 seconds with the small model on an eight-thread Intel Core i7-12700K, using int8.
The concession is real, though. Proprietary cloud models have moved past Whisper: OpenAI reports that its gpt-4o-transcribe model has a lower word error rate than Whisper large-v2 and large-v3 across the languages of the FLEURS benchmark. On heavy accents, noisy rooms and dense jargon, a current cloud model can be measurably better than any local Whisper. That is why several apps here, Sablate included, offer an optional cloud engine next to the local one. In Sablate it is off by default, runs on your own API key, and the app warns you whenever audio is about to leave the machine.
The practical rule: for clear speech in a major language, local Whisper is close enough that your correction pass decides the final quality. For difficult audio, run a two-minute clip through both before you commit a whole recording.
The 7 best offline subtitle generators in 2026
1. Sablate — for subtitles you will style, burn in or translate
Sablate is a desktop app for Windows and Mac that transcribes locally with faster-whisper in four tiers — Fast (tiny), Balanced (small), Accurate (large-v3-turbo) and Best (large-v3) — across 100 source languages. Where it differs from everything else here is what happens next.
Each additional language is a layer on the same transcription rather than a new job, and each layer carries its own style and its own script font, so Arabic, Korean or Devanagari render instead of showing empty boxes. Translation runs offline through Argos, where pairs without a direct package, such as Turkish to German, go through English, or through DeepL, OpenAI or Claude with your own key. The editor has timeline retiming, split and merge, find and replace, and a custom vocabulary for names Whisper keeps misspelling. The 13 caption presets include word-by-word karaoke styles and the Hormozi look, and the burn-in is rendered by libass at 16:9, 9:16 or 1:1. Export covers .srt, .vtt, .txt, .ass, a burned-in .mp4, and one .mkv with a subtitle track per language.
On Windows, an NVIDIA GPU transcribes roughly five to ten times faster than the CPU, and the CUDA runtime downloads the first time you use it. On an Apple Silicon Mac, Sablate switches to whisper.cpp and transcribes on the GPU through Metal, because faster-whisper has no Apple GPU path; in our test on an M4, eight minutes of English took 27 seconds with large-v3-turbo instead of 103 on the CPU. Intel Macs use the CPU.
$29 one-time. The free plan covers videos up to ten minutes, three jobs a month, the Fast and Balanced models and one offline translation layer; .ass, the multi-track .mkv and an unmarked burn-in are Pro. It is more than you need for a one-language transcript, and the right fit when one video has to ship styled or in several languages, for the reasons the multilingual workflow sets out.
2. Subtitle Edit — the free all-rounder
Subtitle Edit is free and open source, built for Windows first, with macOS and Linux builds. Its Video → Speech to text menu lists 12 engines in the documentation: local ones such as whisper.cpp, Purfview's Faster-Whisper-XXL, Whisper CTranslate2 and WhisperX, alongside optional cloud ones. It downloads the engine and model you pick on first use.
After transcription it is the deepest tool on this list: a waveform for timing, reading-speed and line-length checks, automatic translation through several engines, more than 380 subtitle formats, and burn-in through FFmpeg from the Video menu. For format conversion, timing repair and subtitle QC, nothing here, paid or free, does more.
What it asks of you is decisions. Choosing an engine, a model and a GPU backend is up to you, and it is not built for designed social captions such as word-by-word highlights.
Free. Start here if you are unsure, and move on only when you hit something it will not do.
3. Buzz — the simplest free transcriber
Buzz is free and open source under the MIT licence, and runs on Windows, Linux and Apple Silicon Macs; the last version for Intel Macs is 1.4.5. It wraps Whisper, whisper.cpp and faster-whisper, uses an NVIDIA GPU through CUDA and other GPUs through Vulkan, and exports .txt, .srt and .vtt. A watch folder and a command-line mode let you point it at a directory and walk away, and it can also transcribe live from the microphone and label speakers.
It is a transcriber with a subtitle export rather than a subtitle editor. There is no styling and no burn-in, and translation means Whisper's translate-to-English task or an OpenAI-compatible model you connect. For "give me an SRT of this interview", that is exactly the right amount of software.
Free.
4. MacWhisper — the most polished on a Mac
MacWhisper is a native Mac app that runs Whisper locally and uses the Mac's GPU, which the faster-whisper-based tools on this list cannot. The free version exports .srt, .vtt and .txt; Pro adds more export formats and automatic subtitle translation, and optional cloud models sit alongside the local ones.
It is a transcription app first. Styling and burning subtitles into the picture are left to your video editor.
Free, or €64 one-time for Pro. Mac only. On an Apple Silicon Mac, the GPU support is the reason to pick it for long recordings. On Windows, the MacWhisper alternatives cover the closest equivalents.
5. DaVinci Resolve Studio — if you already edit in Resolve
DaVinci Resolve Studio generates subtitles with its own AI model, on your machine, from Timeline → AI Tools → Create Subtitles from Audio. The result lands on a subtitle track you edit like any other part of the timeline, and since Resolve 20 you can animate it, including a Word Highlight style. Delivery is .srt, WebVTT or burned in, from the Deliver page.
Three limits. Automatic subtitles are a Studio feature; the free version of Resolve has none. Transcription covers 14 languages out of the box, with more through an extra download, against roughly a hundred for Whisper. And there is no built-in translation of subtitles.
$295 one-time, for Windows, Mac and Linux. Too much for subtitles alone, and nothing extra if you already own it. If you use the free version, auto subtitles without Studio shows how to bring an SRT in from any tool on this list.
6. Kdenlive — the free editor with Whisper inside
Kdenlive is a free, open-source video editor for Windows, Mac and Linux with speech recognition built into the timeline, using Whisper or the lighter VOSK engine. On first use it sets up a Python environment and downloads the model, after which recognition runs locally. The subtitles land on a dedicated track where you edit and style them, then export them as .srt or .ass or burn them in when you render. Translation into languages other than English uses the SeamlessM4T model, a download of about 9 GB.
The setup is the weak spot: a Python environment and a large first download. Once it works, it is the only free option here that puts the transcript, the edit and the burn-in in one timeline.
Free.
7. Whisper on the command line — for scripts and batches
The openai-whisper command-line tool is free and open source, runs anywhere Python and FFmpeg do, and writes subtitle files directly:
whisper interview.mp4 --model turbo --language en --output_format srt
Wrap it in a shell loop and a folder of recordings becomes a folder of .srt files overnight. whisper.cpp ships its own command-line tool and faster-whisper has third-party ones, and both run faster than the reference implementation on the same hardware. There is no editor, no styling and no preview: you get the file and do everything else somewhere else.
Free. The right choice when subtitles are a pipeline step rather than a creative one.
Offline subtitle generators compared
| Feature | Sablate | Subtitle Edit | Buzz | MacWhisper | DaVinci Resolve Studio | Kdenlive | Whisper CLI |
|---|---|---|---|---|---|---|---|
| Platform | Windows and Mac | Windows, Mac and Linux | Windows, Mac and Linux | Mac | Windows, Mac and Linux | Windows, Mac and Linux | Windows, Mac and Linux |
| Price | $29 one-time, free plan | Free | Free | Free, €64 one-time for Pro | $295 one-time | Free | Free |
| Local speech engine | faster-whisper | Several Whisper builds | Whisper, whisper.cpp, faster-whisper | Whisper | Blackmagic's own | Whisper or VOSK | Whisper |
| Subtitle editor with timing tools | Yes | Yes | Basic | Basic | Yes | Yes | No |
| Translation | Offline layers, or DeepL, OpenAI, Claude with your key | Several engines | Into English, or an OpenAI-compatible model | Pro | No | SeamlessM4T, or into English | Into English only |
| Styled burn-in | Yes | Yes, via FFmpeg | No | No | Yes | Yes | No |
| Subtitle file formats | .srt, .vtt, .txt, .ass, .mkv | 380+ | .srt, .vtt, .txt | .srt, .vtt, .txt, more on Pro | .srt, WebVTT | .srt, .ass | .srt, .vtt, .txt, .tsv, .json |
Read the rows that go against Sablate first. Five of the seven run on Linux and Sablate does not. Four are free. Subtitle Edit handles more than 380 subtitle formats against Sablate's handful. And on a Mac, MacWhisper adds Parakeet models and the Neural Engine, which Sablate does not use. If one of those decides it for you, it decides it.
Is an online AI subtitle generator better than a desktop one?
Not better, different: both run AI speech recognition, and the split is where it runs and what it is built for. Most "best subtitle generator" lists in search results come from online caption services such as Submagic, VEED, Descript and Happy Scribe, so they compare browser tools and leave desktop apps out.
| Desktop subtitle generators | Online AI subtitle generators | |
|---|---|---|
| Where the video goes | Stays on your computer | Uploaded to the service |
| How you usually pay | Free or one-time | Monthly or annual plan |
| What limits a long recording | Your hardware | Plan minutes and upload size |
| Works on a phone or Chromebook | No | Yes |
| Shared projects and team review | Rare | Standard |
The online tools win where speed of setup and teamwork matter: nothing to install, templates for short social clips, and a project a colleague can open in the browser. The desktop tools on this list win on long or confidential footage, on many languages from one recording, and on cost once you subtitle every week. If you are weighing the online side, our comparisons of VEED, Submagic, CapCut's auto captions and Descript go through what each one charges and where it stops.
Offline subtitle generators for Windows 10 and 11
Six of the seven run on a Windows 10 or 11 PC; only MacWhisper does not. On Windows the deciding hardware question is an NVIDIA card. Most of the Whisper-based tools here can run on one through CUDA, and that is the difference between a long recording taking minutes or most of an hour; how much setup it takes is what separates them.
- Sablate detects an NVIDIA GPU on its own and downloads the CUDA runtime the first time it is needed, so the installer stays small for people without one. Without a GPU it falls back to the CPU.
- Subtitle Edit lets you choose the engine, including GPU builds such as Purfview's Faster-Whisper-XXL, which gives the most control and asks the most decisions of you.
- Buzz uses CUDA on NVIDIA cards and Vulkan on other GPUs.
If you have no dedicated GPU, pick the model rather than the tool: small or large-v3-turbo keep a CPU-only laptop practical, as the model comparison shows.
Offline subtitle generators for Mac
On a Mac the hardware question is different. Apple Silicon has a capable GPU, but only tools built on whisper.cpp or on Apple's own frameworks use it; faster-whisper runs on the CPU there.
- MacWhisper uses the Mac's GPU, which makes it the fastest option on this list for long recordings on Apple Silicon.
- Buzz runs on Apple Silicon; its last release for Intel Macs is 1.4.5.
- Sablate ships separate builds for Apple Silicon and Intel Macs; on Apple Silicon it transcribes on the GPU through whisper.cpp and Metal, on Intel Macs on the CPU. What it adds on a Mac is the same as on Windows: styling, burn-in at 9:16, 1:1 or 16:9, and several languages from one transcription.
- DaVinci Resolve Studio, Kdenlive, Subtitle Edit and the Whisper command line also run on macOS.
So on a Mac the honest split is speed against what happens after the transcript: MacWhisper for the fastest text, Sablate when the text has to become designed or multilingual subtitles. For meetings and interviews rather than subtitles, transcription software for Mac covers the note-taking and cloud apps.
How to make subtitles offline in five minutes
Whichever tool you choose, the first run follows the same shape. In Sablate it is:
- Install the Windows installer or the Mac build for your chip. No account or sign-up is needed to use the free version. Pro needs a connection once, to activate the license key from your purchase email; after that the app checks the license about once a day when it is online, and Pro keeps working when it is not.
- Drop in a video that is already on your disk, and pick the spoken language or leave it on auto-detect.
- Pick a model tier. Fast and Balanced run on the free plan. The model downloads once on first use, about 78 MB for Fast and 486 MB for Balanced, and is reused after that.
- Correct the transcript in the editor. Find and replace fixes a misspelled name everywhere at once, and the timeline lets you retime, split and merge cues.
- Export an
.srtor.vtt, or burn the subtitles into an.mp4at the aspect ratio you need.
Which one should you pick?
"I want free, and I don't mind learning the tool." Subtitle Edit. Nothing free goes deeper on subtitle files.
"I need an SRT of an interview, today." Buzz on any system, MacWhisper on a Mac.
"I already edit in Resolve Studio or Kdenlive." Use the built-in recognition. Moving subtitles between apps costs more time than it saves.
"The captions are part of the look." Sablate: presets, word-by-word highlights and a libass burn-in at 9:16, 1:1 or 16:9. The burn-in guide covers the quality side of that last step.
"One video, several languages." Sablate. One transcription and a translation layer per language, each with its own style — translating an SRT without breaking the timing explains why that order matters.
"Hundreds of files and nobody watching." The Whisper command line, or Buzz's watch folder.
"Long recordings on a Mac." MacWhisper. Sablate also transcribes on the Apple Silicon GPU, but MacWhisper adds the Neural Engine and the faster Parakeet models.
The bottom line
Offline transcription is no longer the compromise it was: the Whisper models are the same ones the cloud Whisper services run, and a mid-range GPU turns an hour of speech into a coffee-break job. Choose by what you need after the transcript. For free subtitle files, Subtitle Edit or Buzz. For subtitles that have to look designed or ship in several languages, download Sablate for Windows or Mac and run one real video through the free plan, or see what Pro adds for a single $29 payment.
Frequently asked questions
What is the best offline subtitle generator for Windows?
Subtitle Edit if you want free and thorough, Buzz if you only need a quick SRT, and Sablate when the subtitles have to be styled, burned in or translated into several languages. All three run on Windows 10 and 11 and use an NVIDIA GPU when there is one, which is the biggest speed-up available on Windows.
What is the best offline subtitle generator for Mac?
MacWhisper for fast transcripts, because it uses the Mac's GPU, and Buzz as the free alternative on Apple Silicon. Choose Sablate when the subtitles need styling, burn-in or several languages; it runs on both Apple Silicon and Intel Macs and transcribes on the GPU on Apple Silicon.
What is the best offline subtitle generator?
It depends on what happens after the transcript. Subtitle Edit is the best free all-rounder, Buzz the simplest free transcriber, MacWhisper the most polished on a Mac, and DaVinci Resolve Studio the natural choice inside an edit. Sablate fits when the subtitles have to be styled, burned into the video or delivered in several languages.
Can I generate subtitles without uploading my video?
Yes. Every tool on this list runs speech recognition on your own computer, so the video file stays there. Several of them also offer optional cloud engines, which do send the audio out when you select them, so check which engine is active before you process confidential footage. Browser-based caption services upload the video by design.
Is offline transcription as accurate as cloud transcription?
Close, but not always equal. Local tools run OpenAI's open Whisper models, and OpenAI's own Whisper API serves large-v2, an older version than the large-v3 you can run locally. Newer proprietary cloud models do report lower error rates than Whisper large-v2 and large-v3, so on difficult audio a cloud model can still be measurably better.
Do I need a GPU to transcribe video offline?
No, but it helps a lot. In faster-whisper's published benchmark, 13 minutes of audio took about a minute with the large-v2 model on an NVIDIA RTX 3070 Ti, and under two minutes with the small model on an eight-thread desktop CPU. CPU-only machines are practical with small or turbo; the largest model on a CPU is slow.
Do I need an account to use an offline subtitle generator?
Usually not. Subtitle Edit, Buzz, Kdenlive and the Whisper command line need no account, and Sablate's free version runs without an account or sign-up. Paid tiers differ: Sablate Pro needs an internet connection once, to activate the license key from your purchase email, and keeps working offline after that.
Which offline subtitle generators are free?
Subtitle Edit, Buzz, Kdenlive and the Whisper command-line tool are free and open source, and MacWhisper has a free version. Sablate has a permanent free plan limited to ten-minute videos and three jobs a month. The free version of DaVinci Resolve has no automatic subtitles; that feature is part of the paid Studio version.
What is the difference between transcription software and a subtitle generator?
Transcription software produces text with timestamps. A subtitle generator produces cues a viewer can actually read: short lines, a sensible reading speed, breaks at phrase boundaries, and a file a player or platform accepts. Most tools on this list start from the same Whisper transcript; what separates them is how much of the subtitle work comes after it.