Captions, voices and cutout

Nothing uploads. Captions, voices, cutout and enhance all run on your machine, on models that download from Settings the first time a feature needs one and never need the network again. Your footage never leaves your disk. Concat ships no model weights inside its installers: a cutout network, a voice bank and a caption model together outweigh the editor many times over, and most people never ask for all of them.

Auto-captions

Local Whisper transcribes the audio and puts styled captions on the timeline. Pick a size: bigger is more accurate and slower, and the larger ones want 16 GB of RAM. Each size comes in an English-only variant and a multilingual one that detects the language on its own.

Model	English	Multilingual
Tiny	78 MB	78 MB
Base	148 MB	148 MB
Small	488 MB	488 MB

Text-to-speech and voice cloning

Three families of voice, one Speech panel.

Model	What it does	Size
Kokoro	Built-in speakers in several languages. Quick.	132 MB (int8) or 350 MB
Pocket TTS	Kyutai’s English voice. No speakers of its own: every voice is a few seconds of a recording. Two recordings come with it, and any clip in the bin can be the third.	98 MB
Chatterbox Turbo	Resemble AI’s voice, the closest to a studio read. Every voice is a recording, and it reads [laugh], [chuckle] and [sigh] as what they say. Slower, and the biggest download.	About 1.1 GB, in nine files

Background removal and enhance

Model	What it does	Size
Person cutout	Cuts a person out of the frame	15 MB
Object cutout	Cuts an object out of the frame	179 MB
Cutout brush	Paint the mask yourself	40 MB
Enhance	Restores and upscales a clip’s picture, one clip at a time	5 MB

Voice cleanup

One switch to denoise, enhance the voice and level the loudness. Plus chipmunk, robot, telephone and friends. These need no download.