How to Make Hormozi-Style Captions
Big, heavy, few words at a time, with the key word hit in a second color. Here is why the look works and how to make it without keyframing a single word.
Read the post
Guides
Captixa Team
1 min read
Free Offline Subtitles with Local Whisper
Whisper can run entirely on your own computer: no upload, no API key, no internet. Here is which model to pick and what to expect.
Captixa Team
1 min read You can make subtitles completely offline by running a Whisper speech-to-text model on your own computer. Download the model once, and from then on transcription needs no internet, no account and no API key, and your video never leaves your machine. The trade-off is speed: your computer does the work a cloud service would.
What Whisper is
Whisper is an open speech-recognition model released by OpenAI. It is available in several sizes, and optimized versions such as whisper.cpp run on ordinary computers without a special graphics card.
Which model to pick
| Model | Download | Good for |
|---|---|---|
| Tiny | about 75 MB | Quick drafts and testing |
| Base | about 142 MB | Clear speech, short clips |
| Small | about 466 MB | A good everyday balance |
| Medium | about 1.5 GB | Accents and noisier audio |
| Large-v3 Turbo | about 1.6 GB | Close to Large-v3, several times faster |
| Large-v3 | about 3 GB | The most accurate |
If you are not sure, start with Large-v3 Turbo. Drop to Small if transcription is too slow on your computer.
When offline makes sense
- Private footage: client work, unreleased episodes, classroom recordings.
- No reliable internet: travelling, on location, or on a metered connection.
- No accounts: no provider sign-up and no key to manage.
When a cloud engine is better
On a slower computer, or for long videos, a cloud engine such as Groq is much faster, and Groq has a free tier. The trade-off is that the extracted audio (not the video) is sent to the provider.
Offline subtitles in Captixa
- Open Settings and set the transcription engine to Local Whisper.
- Click Download next to a model.
- Load a video and click Transcribe. Then style it and export an MP4 with captions, or an SRT.
Everything after the model download, from transcription to export, runs without an internet connection.
Is local Whisper really free?
Yes. The models are free to download and run on your own computer. There is no key and no per-minute cost.
Is it as accurate as cloud transcription?
The larger local models are close. Large-v3 is the most accurate local model; Large-v3 Turbo is close to it and several times faster. Smaller models trade accuracy for speed.
Does anything need the internet?
Only the one-time model download. After that, transcribing, styling, editing and exporting all work offline. Translation and Roman-script conversion still need a connection.