Local AI and speech

#What runs where

#Set up Local AI

• After sign-in and workspace setup, the app offers Local AI. Choose Download Local AI or Set up later; editing and imports work either way.
• The optional download is about 944 MB, with progress, cancel and retry. Completed files are reused.
• Come back to it through Local AI in Project media or Set up Local AI in Storage settings.
• With Prepare new footage automatically on and Local AI downloaded, small imports (up to 10 visual files, 500 MB and five minutes) are prepared in the background. Larger batches ask first.
• Pause imports or analysis from Project media. Unfinished jobs wait for Resume when you reopen a project.

#Transcription

• Transcribe from the Media library or the Transcript tab. Formats sends the clip audio to its hosted speech service and returns an editable transcript with word-level timing.
• It supports 90+ languages. Choose the audio language, then whether captions keep the original script, keep English words, use Latin letters or translate to English.
• The transcript is saved with the project in your workspace and powers captions, search and filler-word removal.

#Speech from a text clip

1. 1Right-click a text clip on the timeline and choose Generate Audio From Text.
2. 2Pick an engine, then a voice and, where the engine supports it, speed and language.
3. 3Preview the result, then insert it into the timeline or save it to the library.

For studio voices in 16 languages, voice cloning and AI avatars, use hosted Voiceover and Avatar videos from the Library, which need the Creator plan. See Generate with AI.

#Manage models

• Open Storage settings to manage Local AI.
• Use Local AI Model Cache to inspect or clear downloaded models, for example if a download is corrupt.
• Use Unload Local Models to release memory.
• Long jobs can be cancelled if you no longer need the result.