What is text-based video editing?

Text-based video editing means editing a video through its transcript instead of a timeline. Speech is transcribed with a start and end time for every word, so a sentence in the text points at an exact range of footage. You choose words to cut or keep, and the video follows.

Can I edit a video by editing the transcript in Valmera?

You edit against the transcript, by describing the change. Name the words ("cut the sentence about the old price") and the agent finds them in the word-level transcript and cuts them on word boundaries. Deleting text in the transcript panel does not cut footage: the panel is for reading, jumping to a moment and correcting misheard words, which updates the captions.

How do I remove specific words from a video?

Quote them. "Remove 'to be honest' every time I say it" cuts every instance; "cut the first 'basically' at 1:20" cuts one. The agent locates the words in the transcript, snaps each cut to word edges, renders a preview, and any cut can be restored.

How is this different from Descript?

Descript's core workflow is a document: you select and delete words and the media follows, and its Underlord assistant also takes conversational direction. In Valmera you describe outcomes and an agent performs them, which scales to requests like "every um in a two-hour recording" and covers non-text edits like zooms, music and reframing in the same chat. If you want to hand-pick every word in a document view, Descript is built for that.

How accurate are cuts made from text?

Cuts snap to word boundaries from the word-level transcript, so words are not clipped mid-syllable. Transcription can mishear names or overlapping speech; if a cut lands wrong, say so ("you cut too much, bring back the dog story") and the agent restores it.

Can I still edit on a timeline?

Yes. The Studio timeline lets you split, trim and delete segments, and retime or remove text by hand. Each change is saved as a new version alongside the agent's edits.

Is text-based editing free in Valmera?

Creating an account and uploading are free. Editing requires a subscription. Exports are full-resolution H.264 MP4s and end with a brief Valmera end card.

Text-Based Video Editing: Edit Video by Its Transcript

Text-based video editing means cutting a video through its transcript instead of a timeline. In Valmera the word-level transcript is the map and the chat is the instruction: name the words to remove, such as “cut the sentence about the old price,” and the AI agent finds them, cuts them on word boundaries, and renders a preview.

The transcript panel shows every word, jumps the player when you click, dims the words your edit has cut, and lets you fix a misheard line. Deleting text there does not cut footage; cuts come from what you ask for in chat, or from the timeline if you prefer to edit by hand.

Edit Your Video by Its Words

Upload, read the transcript, describe the cut. Creating an account and uploading are free; editing requires a subscription.

Create an account →

See pricing →

How to Edit a Video by Its Transcript in Valmera

1
Upload the video
MP4, MOV, M4V, MKV or WebM, up to 14 GB and 3 hours. Indexing builds a word-level transcript with speaker labels, plus shots, silences and frames.
2
Read the transcript
Skim it like a document and click any word to jump the player there. Fix any misheard name so captions come out right.
3
Describe the cut
Quote a few words to anchor it: “cut the sentence that starts ‘so before we get into it’,” or describe the passage: “remove the tangent about my cat.”
4
Review, restore, export
Watch the preview; cut words now appear dimmed in the transcript. Ask for anything back, then export a full-resolution MP4.

PROMPTS TO PASTE

Cut the sentence that starts “so before we get into it”.

Remove the words “to be honest” every time I say them.

Cut everything before I say “welcome back”.

Move the story about the launch to the start of the video.

What Happens Where

You want to	Where	What happens
Find a moment	Transcript panel	Click any word and the player jumps there.
See what is cut	Transcript panel	Words the current edit removes appear dimmed.
Fix a misheard word	Transcript panel	Correct the line; captions use the new text. The audio is unchanged.
Cut, keep or reorder speech	Chat	Name the words or passage; the agent edits on word boundaries.
Split, trim or delete a segment	Timeline	Direct edits by hand, each saved as a new version.
Bring something back	Chat or version history	“Restore the joke at 2:10” returns that range only.

Deleting Words vs Describing the Edit

Text-based editors fall into two schools. In the document school, made popular by Descript and now also in Premiere Pro and DaVinci Resolve Studio, you select words and delete them, and the linked footage follows. It is precise for hand-picking, and it is still manual: removing every “um” in a two-hour recording means finding every “um.”

In the instruction school, which Valmera uses, you describe the outcome and an agent performs it against the same word-timed index. One sentence covers the whole file (“remove every ‘to be honest’”), and the same chat handles what text cannot express: a punch-in on a word, music under a section, a 9:16 reframe. Full head-to-head: Valmera vs Descript.

Settings That Matter

Every instance or one. “Every time I say it” cuts all of them; “the first one at 1:20” cuts one.
Exact words beat paraphrase. Quoting three or four words pins the passage; a summary may match more than one moment.
Which clock. Times you see in the preview are edited-video time. Say “in the original recording” if you mean the source.
Word-edge snapping. Transcript-based cuts snap outward to whole words so nothing is clipped mid-syllable.

Common Mistakes

Expecting a transcript correction to change the audio. Fixing “Val Mira” to “Valmera” corrects captions; the speaker still says what they said.
Deleting text in the panel to cut. Ask in chat instead, or cut on the timeline.
Reordering without cover. Moving a sentence splices two moments that were never adjacent, so the picture jumps. Cover the join with b-roll or a punch-in.

Limits

A transcript only maps speech. Silent b-roll, music and visual moments are not in it, so describe those by what is on screen (the agent watches the frames) or by time. Captions are burned in, with no SRT or VTT import or export. English is the best-tested transcription language.

Related Tools

The most common transcript edits have their own pages: remove filler words, remove bad takes, remove silence and cut by timestamp. Learn how the transcript and timeline fit together in the timeline and transcript docs and the cuts and cleanup docs, see the podcasters and YouTubers workflows, or take the AI video editor tour.

Assistants connected to Valmera's MCP server edit the same way, with search_transcript, get_words, cut_range, keep_segments and get_kept_transcript.