Guide, updated October 19, 2026
Auto captions offline on Windows, with Whisper on your PC
Most automatic captions are made by sending your video to a server. For a product demo that shows an internal tool, a customer name or a pre-release screen, that upload is the part you would rather skip. Ciakko writes the captions of your screen recordings with Whisper on your own computer. Here is what it needs, how long it takes, and what you get.
In one glance
- Where: on your PC, with Whisper. The model is downloaded once, on request.
- Speed: much faster on an NVIDIA card, still usable on the processor.
- Output: captions drawn in the video in six styles, and an .srt file with every export.
- Limit: it captions recordings made in Ciakko, on Windows 10 and 11.
What runs where
Whisper is the open speech model released by OpenAI. Ciakko runs it locally through Whisper.net. The first time, you choose a quality and press Download model; after that, captions are written with no connection to any speech service. Only the stretches where somebody speaks are passed to the model, which saves time on a demo with long silent parts. One pass gives both the captions and the transcript, with the time of every word, so the same work also feeds the cleanup and the Script.
One thing does leave the PC, and only if you use it: the voice over. Its text goes to Microsoft Azure Speech to be read aloud. Captions laid on a voice over need no model at all, because the words are already known.
How long it takes
With an NVIDIA graphics card Whisper runs on the card and is much faster. Measured on a GTX 1650 Ti, a 30 second window of speech takes about 13 seconds with the large-v3-turbo model. Without an NVIDIA card it runs on the processor: choose Fast or Balanced and it stays practical for tutorials of a few minutes.
Six caption styles
Classic is a plain line. Karaoke lights up each word as it is spoken. Pill and Box put the words on a background. Gradient colors the text. Short shows three big words at a time, in capitals, the way vertical videos do. Size, highlight color, word pop and width are settings; you can also drag the captions on the preview and double click one to change its words.
In the picture, or only in the file
Turn the captions on and they are drawn into the video. Turn them off and they stay out of the picture. Either way the export writes an .srt file beside the video, the kind YouTube reads, so viewers can switch captions on and off and search engines can read what you said.
How to do it
- Open the take and go to the Captions tab.
- Pick a Quality and press Download model the first time.
- Set the Language you spoke, or Auto detect.
- Press Generate captions, then choose a Style.
- Export: the video and its .srt file land side by side.
When another tool is the better fit
If the video already exists and was not recorded in Ciakko, use a tool made for files: Subtitle Edit can run Whisper on any video, and YouTube Studio writes captions after upload. If you need captions translated into other languages, that is not in Ciakko today.
Questions about offline captions
Can I make captions offline on Windows?
Yes. Ciakko writes captions with Whisper running on your PC. After the model is downloaded once, the recording is never uploaded to be transcribed.
Do I need an NVIDIA graphics card?
No. With an NVIDIA card the captions are written much faster; without one Whisper runs on the processor. On a GTX 1650 Ti, 30 seconds of speech take about 13 seconds with the turbo model.
Which Whisper model does it use?
You pick a quality, Fast, Balanced or Best, and the product picks the model that fits your PC among tiny, base, small, medium and large-v3-turbo. Manage models lists what is on disk, to add or delete one.
Does it make an .srt file for YouTube?
Yes. Every export writes an .srt file next to the video, also when the captions are kept out of the picture.
Can I caption a video recorded with another program?
No. Ciakko captions its own recordings. If you need captions for an existing file, a general tool such as Subtitle Edit with Whisper or the captions of YouTube Studio is the better fit.
Which languages are supported?
The languages Whisper writes. Choose the one you spoke or Auto detect; the quality depends on the language and on the model.