How to use a transcript while editing a short video

The sentence you want may be buried in the middle of a recording. A transcript helps you find it, but the words on a page do not show a breath, a laugh or the glance that makes the moment work on video. Use the text as a map, then make the cut by watching and listening.

The Studio preview combines a short multitrack timeline with timed transcription and reviewable cut suggestions. It is currently bounded to a ten-minute composition. For a full podcast episode whose main output is text, see the podcast transcript guide; the workflow here is for editing a video or audio excerpt.

Transcribe the range you are editing

Select the source clip and request transcription for a source range. Studio sends the selected audio to its transcription service and speech provider. Transcription usage and the supported speech-to-text API-key setting apply. This is distinct from local audio matching, which compares source sound in your browser without an upload.

When the provider supplies word timings, click a word to seek in the recording. Scan for the point where the speaker starts answering the question, then play several seconds before it. A good opening often needs a little context that does not look important in the transcript. Listen through the end as well; a text selection can stop before the speaker finishes the thought.

If the provider returns text without usable word timings, Studio keeps it untimed rather than inventing positions. You can still read it, but caption corrections and timing-based actions need reliable timed words. Names, acronyms and overlapping speakers deserve a proofread against the sound even when timings are present.

Separate caption corrections from source speech

Studio lets you correct caption text while preserving the raw transcript. That distinction is helpful when the speech recogniser spells a name incorrectly: the subtitle can be fixed without pretending the audio said something different. Export SRT or VTT sidecars mapped to the current timeline after you have settled the edit. A selected range rebases captions to the exported range's start.

Keep captions readable in playback. A line that is technically aligned can still disappear before someone has time to read it. Seek through the beginning and end of each caption around your cuts, especially if you have moved or trimmed clips since transcription. The Studio guide shows the caption and export controls.

Treat cuts as proposals

Literal filler search runs locally. You can also request AI cut proposals from timed transcript words. Studio sends transcript words, a source label and your instruction to the configured model provider. Preview each proposed edit, accept it individually or accept a reviewed batch, and use Undo if the new rhythm is wrong. A removed “um” may leave an audible jump or shorten a pause that gave the answer its shape.

AI can suggest highlight excerpts from timed words as well. Preview a suggested range across all tracks; exporting that range rebases its captions and does not alter the saved composition. These suggestions use a server-selected model. A speech-to-text API key does not select the suggestion model. A cancelled or invalid response leaves the edit alone.

Finish on the timeline

After a text-led cut, play across it with picture and sound together. Check the room tone, the position of any separate microphone and the first frame after a camera change. Adjust clip positions, source in and source out, fades or track audio as needed. The multitrack editor guide walks through those controls.

Export a whole edit or selected range as video WebM with the saved mix. The preview renders 720p WebM in real time with the tab visible. Supported audio-only WAV exports have a separate 60-second limit and source-format restrictions. Save a project backup and retain the original media before you close the browser. The transcript will still help you locate the next excerpt, but the last decision belongs to the playback.