Edit Interview Video by Speaker with AI
Valmera edits interviews from a transcript that knows who is speaking. Upload a two-person interview and describe the edit in plain English: keep the guest's answers, drop repeated questions, frame whoever is talking in a vertical cut, add captions. The agent makes the cuts, renders a preview it checks itself, and you revise by chatting.
A tight main interview, speaker-framed vertical clips, and a version history of every change. Your original file is never modified, so any cut can be restored.
How Editing by Speaker Works
When you upload, Valmera transcribes the interview word by word and labels each line by speaker: S0, S1 and so on. Tell the agent who is who once (“S0 is the host, Maya; S1 is the guest, Dev”) and you can edit by person instead of by timestamp. Speaker labels come from the audio, so check the joins where people talk over each other.
Keep or cut by speaker
“Remove the host's questions unless the answer needs them” or “cut everything S1 says before minute three.” The agent finds the lines in the transcript and cuts on word boundaries.
Build the story from complete answers
Give the structure and target length, and the agent keeps the answers that serve it and cuts the rest, in recorded order. Ask it to keep answers whole rather than splicing words into new sentences.
Tighten without flattening
Dead air, ums, false starts and repeated takes come out. Laughter, breaths and pauses that carry meaning stay when you say so, and every cut can be restored.
Related tools: Text-based video editing; Remove repeated takes; Remove silence; Transcript and timeline docs.
Two-Person and Two-Camera Setups
Framing is chosen section by section: in a vertical cut the crop moves to the next speaker at the turn rather than continuously following movement. Pick the setup that matches your recording.
One wide camera
The agent aims a 9:16 crop at whoever is speaking in each section, using the speaker turns and the faces it finds in the frame. For reactions, it keeps the full two-shot inside the vertical frame with room for a headline and captions.
Side-by-side recording
A Zoom, Riverside or StreamYard layout puts both people in one frame. The agent crops to the half that is talking, section by section, so each person fills the vertical frame on their own answers.
Two cameras
Upload camera A as the main video and camera B as a clip, and give the sync point, such as a clap at 0:04 on A and 0:07 on B. The agent can cut to camera B for the sections you choose while the main audio keeps playing, or build a side-by-side panel of up to 20 seconds.
Framing guides: Auto reframe; Reframing docs; Multi-speaker video podcasts.
What Makes This Editing Job Difficult
Meaning can change at a cut
Removing a qualifier, setup or follow-up can make a quote stronger and less true. Every structural cut is an editorial decision.
Two people, one vertical frame
A 9:16 crop holds one face comfortably. The edit has to decide who is on screen in each section, and when a reaction matters more than the answer.
The conversation isn't the story
Questions arrive in recording order. The finished piece needs a hook, an argument and a close, which means choosing the answers that build it and cutting everything else.
What to Prepare Before You Upload
Most of the editorial context lives in your head. Give the agent the names, the story you want and the rules it must not break.
- The interview recording (MP4, MOV, M4V, MKV or WebM, up to 14 GB and 3 hours) and any second-camera file, with a sync point if you have one.
- Who is who (“S0 is the host, Maya Chen; S1 is the guest, Dev Patel”), plus spellings of names, companies and terms.
- Target length, audience and where the cut will be published.
- Must-keep quotes, anything off the record, and claims that need exact wording.
- Caption style, logo, b-roll and music you want to use.
How to Auto-Edit a Two-Person Interview in Valmera
Each step is one message in the chat. The agent renders a preview after every change, so approve the story before you spend time on polish.
- 1
Upload and name the speakers
Upload the interview. When indexing finishes, check the transcript, tell the agent who S0 and S1 are, and correct any misheard names.
- 2
Ask for the story cut
Describe the structure and target length. The agent selects complete answers, keeps the questions that give them context, and shows you the cut before adding polish.
- 3
Clean up the dialogue
Remove dead air, ums, false starts and repeated takes. Protect pauses, laughter and qualifiers that change meaning, then listen across the joins.
- 4
Frame the speakers
Keep 16:9 for the main cut, or ask for 9:16 framed on whoever is speaking, section by section. Use the full two-shot for reactions. With two cameras, give the sync point and the sections for camera B.
- 5
Add captions, names and sound
Add word-timed captions and a name title on each person's first appearance. Duck any music under speech and master the mix to -14 LUFS.
- 6
Make clips and export
Turn complete answers into vertical shorts, review each preview, then export full-resolution MP4s.
Prompts You Can Copy
These are production briefs, not magic words. Replace bracketed details with the real audience, duration, platform, claims, and assets for your project.
Limits
Speaker labels come from transcription and can slip when voices overlap or sound alike; English is the best-tested language. The main recording plays in recorded order: to open on a later moment, upload that moment as a separate clip. Framing is chosen per section, not tracked continuously, and Valmera doesn't sync cameras or switch angles on its own. It doesn't remove background noise or balance two voices recorded into one track, captions are burned in (no SRT or VTT files), and you post the finished MP4 yourself. There are no team seats yet.
Learn more: Valmera docs; Automate with MCP; How the AI video editor works.
Edit the Interview by Speaker
Upload the interview, name the speakers, and ask for the cut you want.
Direct the first edit →The Review Gate: What a Human Must Check
Quote integrity
Compare every shortened answer with the source. Certainty, chronology and conditions must survive the cut.
Who is on screen
In vertical cuts, check each speaker turn: the right face, no one cut in half, and reactions shown where they matter.
Speech and captions
Listen on headphones and a phone speaker. Check names, numbers and terms in the captions against the final audio.
Deliverables to Request
- Main interview MP4 in 16:9 with captions and name titles.
- Speaker-framed 9:16 clips, each built around one complete answer.
- A caption-free version if your platform adds its own captions.
- A short teaser cut from the approved interview.
When an Agentic Editor Fits—and When It Does Not
A strong fit
- Speech-led interviews with clear audio and identifiable speakers.
- Teams that know the story they want and the quotes that must survive.
- Shows that need both a long cut and vertical clips from one recording.
Use another workflow
- Multicam productions that need automatic sync and angle switching.
- Recordings that need noise repair before anyone can understand them.
- Investigative or legally sensitive work without a human editor reviewing every cut.
The Operating Principle
Decide what the conversation says first, then how viewers see and hear it. A transcript with speaker labels makes the first decision fast: you edit people and answers, not timestamps. The agent handles the cutting, framing and captions; you own the meaning.
New to the model? Read the complete guide to AI video editing or start with production-ready editing prompts.
Frequently Asked Questions
Direct the Interview Edit
Give the agent the footage, the speakers and the story. Review a rendered cut instead of starting from an empty timeline.
Edit this workflow with Valmera →