Auto-captions for screen recordings, without the cloud

Short answer: you don't need a cloud service, or a per-minute credit meter, to caption a screen recording. Modern speech models run on your own machine: Girato downloads a ~40MB model once, then transcribes your mic or voiceover on-device, free and unlimited, producing timed cues you can edit like text. Your audio never uploads.

Why captions matter more than ever

Most feeds autoplay muted. A demo without captions is a silent film for the majority of first-time viewers. LinkedIn, X, Reels, docs pages. Captions aren't accessibility garnish; they're the difference between "watched" and "scrolled past".

The two ways tools do this

Cloud transcription (Tella, FocuSee's AI tools, Descript): audio uploads to their servers. Quality is good; the costs are per-minute pricing, credit meters, and your voice on someone else's machine. On-device transcription (Girato): the model runs locally. First run needs internet once to fetch the model; after that it works on a plane. No meter, no upload.

How to do it in Girato

1. Record with the mic on (or upload a voiceover afterwards).
2. Open Captions → Auto-captions from mic. First run downloads the model (~40MB, once).
3. Proofread the cue list, every cue is editable text with its own timing.
4. Export. Captions render into the video, in every aspect ratio.

Tips for cleaner transcripts

• Speak a touch slower than feels natural; pause at sentence ends, the model uses silence to time cues.
• Name your product once, clearly, then correct any mishears with one edit per cue.
• Cut dead air first (Girato's silence auto-cut) so cue timing stays tight.

Frequently asked questions

Can captions really be generated without uploading audio?

• Yes. Speech models like Whisper run on ordinary laptops.
• Girato downloads a ~40MB model once.
• It transcribes your mic or voiceover entirely on-device.
• The audio never leaves your machine.
• It works offline afterwards.

Are local captions accurate enough?

• For clear speech, on-device Whisper is close to cloud services.
• Every tool, cloud or local, still needs a quick proofread.
• Girato makes each caption cue editable text.
• Fixing a word takes seconds.

Why do some tools charge credits for captions?

• They run transcription on their servers, which costs per minute.
• So they meter it.
• On-device transcription costs the vendor nothing per use.
• That is why Girato's captions are free and unlimited.