You filmed an event. While filming, you commented out loud on what you want done in the edit: "cut there", "keep that", "this is the highlight", "redo". These instructions, aligned with the video's timecode, can be turned into an EDL importable into Premiere Pro or DaVinci Resolve.

Editing is when the instructions given in the field come into their own. But between the shoot and the editing desk, the details fade: which take was the good one, at exactly which second, why you said to keep that one. Timecoded voice instructions skip that dead zone.

Prerequisites

Before you start, you need:

1. A timecoded transcript of your video, in SRT or VTT format. Generate it from your recording with Whisper, CapCut, or any transcription tool. If you want to dictate your editing instructions by voice — saying "cut there" at the exact moment — a timecoded video mode is planned in KontexVoX (see below).
2. An AI agent with file-system access: Claude Code, Codex CLI, or Gemini. Not a web chatbot.
3. The transcript exported to your computer.
A web chatbot doesn't read your local files and doesn't generate technical files in a precise format. To produce an EDL and markers, you need an agent working inside your folder.

The timecoded video mode (coming)

KontexVoX today enriches photos: you look at an image, you speak, the description is engraved into the EXIF. For video, the principle is the same but the medium changes: you watch your recording, you dictate your editing instructions as they come to you, and the app generates a timecoded file (SRT or JSON) linked to the video — not a per-image EXIF description. This mode doesn't exist yet, it's on the backlog. In the meantime, transcribe your video with any tool (Whisper, CapCut, Premiere) and run the prompt below. The result is the same.

What you dictated: an example

Here's what the SRT transcript of a 12-minute video contains:

00:01:23,400 → "This is the opening speech, keep all of it." 00:03:47,200 → "Cut there, the guy walking in front ruins it." 00:05:12,800 → "Highlight: she cries reading the letter. Keep the wide shot." 00:07:30,000 → "Redo, I stuttered. Start again after." 00:09:15,600 → "This is the finale, keep the music, cut on the wide shot."

These instructions, you said while watching the video. The AI agent will parse these entries, identify the editing instructions ("cut", "keep", "highlight", "redo"), and turn them into EDL entries with precise timecodes.

The prompt

Read the timecoded transcript file (SRT or VTT). Parse each entry to extract the natural-language editing instructions ("cut", "keep", "highlight", "redo", etc.). Generate an EDL file in CMX 3600 format, the standard for Premiere Pro and DaVinci Resolve. Each EDL entry must contain: start timecode, end timecode, event type (video, audio), and a note with the original voice comment. Also include a markers file in CSV format for DaVinci Resolve.

What the AI produces with this prompt

Two files:

1. An EDL file in CMX 3600 format, directly importable into Premiere Pro or DaVinci Resolve. Each entry contains the start timecode, end timecode, event type, and a note with your original voice comment.
2. A CSV markers file for DaVinci Resolve, placing markers on the timeline at the moments you flagged.

The editor starts from a pre-built timeline instead of an empty project. The instructions you gave while watching the video are already there, positioned at the right timecode, with the context of your voice. All that's left is to execute the cuts and adjust the rhythm.

Recommended tool: Codex CLI

Codex CLI is excellent at structured parsing and generating technical files in precise formats. CMX 3600 EDL is a strict format, and Codex handles this kind of generation well.

Frequently asked questions

Does it work with any transcript format?

The prompt asks for SRT or VTT, the standard timecoded transcript formats. If your transcript is in another format (TXT without timecodes, JSON), the agent can read it but the result will be less precise — timecodes are essential for positioning the instructions on the timeline.

Does the AI guess the end timecodes?

No. The prompt extracts the EDL's start and end timecodes from the voice instructions. For cuts ("cut there"), the AI takes the instruction's timecode as the cut point. For segments to keep ("keep all of it"), it takes the instruction's timecode as the start and the next instruction's timecode as the end. It's a working base, not a finished edit — the editor adjusts afterwards.

Can I edit the EDL by hand afterwards?

Yes, and it's recommended. The generated EDL is a starting point, not a final edit. The timecodes are precise but editing choices get refined at the desk. The advantage is starting from an already-built structure instead of repositioning everything manually.

Does it work if I have several videos?

Yes, but you need one transcript per video. Run the prompt separately for each transcript, then merge the EDLs in Premiere or DaVinci. The prompt doesn't handle multi-camera — it processes one linear timeline per video.

Going further

Automatic rough cut with an AI agent Extract the highlights from a 3-hour live stream Which AI agent for your use case? Video page