docs01
Whisper models
Choose your model and language at project import. You can re-transcribe anytime using a larger model without re-importing.
ModelSpeedBest for
BaseFastestQuick rough drafts and clear solo speech
SmallBalancedRecommended default for high accuracy in 19 languages
MediumMost accurateAccented speakers, jargon, and complex audio
02
MossFormer2 Studio Voice
The FP16 neural model (~111 MB) runs on Apple MLX with direct Apple Silicon GPU acceleration at 48 kHz. It is bundled ready to use from the first app launch.
03
Offline capabilities
LocalVideo, audio, transcription, cuts, and exports
OnlineSign-in, license/trial verification, updates