Edit Video by Talking to Claude: Text-Based Timeline Cuts with Whisper & DaVinci Resolve EDLs (2026)

The 4-Stage Text-Based Editing Pipeline

Here is how the entire system works from raw camera card to finished social reel:

Stage 1: Extracting Word-Level Timecodes with Whisper

To give Claude surgical precision, you need timestamped transcriptions. You can generate this locally for free using whisper-cli or the open-source faster-whisper Python library:

# Extract high-efficiency audio from your raw video
ffmpeg -i raw_podcast_4k.mp4 -vn -acodec libmp3lame -q:a 4 podcast_audio.mp3

# Run faster-whisper to generate detailed JSON timestamps
pip install faster-whisper

# transcribe_for_claude.py
from faster_whisper import WhisperModel
import json

model = WhisperModel("large-v3", device="auto", compute_type="float16")
segments, info = model.transcribe("podcast_audio.mp3", word_timestamps=True)

transcript_data = []
for segment in segments:
    transcript_data.append({
        "start": round(segment.start, 2),
        "end": round(segment.end, 2),
        "text": segment.text.strip()
    })

with open("transcript_timeline.json", "w") as f:
    json.dump(transcript_data, f, indent=2)

print("✅ Saved timestamped transcript for Claude analysis!")

Stage 2: The Master Narrative Editing Prompt

Upload the transcript_timeline.json into Claude (Claude 3.7 / Claude Sonnet handles hundreds of thousands of words with ease). Give Claude this specialized directive:

You are a senior television story editor and documentary post-production specialist.
Attached is a timestamped transcript from a raw interview.

Your goal:
1. Identify the 3 most compelling, viral, or intellectually insightful segments (between 45 and 90 seconds each).
2. For each segment, establish a strong opening hook within the first 3 seconds.
3. Eliminate verbal filler, false starts, and trailing tangents while ensuring the speaker's thoughts flow logically.
4. Output the exact IN and OUT timecodes in SMPTE format (HH:MM:SS:FF at 24fps or 30fps).
5. Generate a standard CMX 3600 EDL (Edit Decision List) for the top segment.

Stage 3: What an Edit Decision List (EDL) Looks Like

Claude analyzes the conversation structure, discards the awkward rambling, and outputs a CMX 3600 EDL block:

TITLE: PODCAST_VIRAL_HOOK_01
FCM: NON-DROP FRAME

001  AX       V     C        00:04:12:10 00:04:28:18 00:00:00:00 00:00:16:08
* FROM CLIP: RAW_CAMERA_A.MOV
* HOOK: "THE REAL REASON MOST AI AGENTS FAIL IN PRODUCTION"

002  AX       V     C        00:04:31:02 00:04:52:14 00:00:16:08 00:00:37:20
* FROM CLIP: RAW_CAMERA_A.MOV
* CORE EXPLANATION: CONTEXT DRIFT AND RETRY LOOPS

003  AX       V     C        00:05:01:18 00:05:22:04 00:00:37:20 00:00:58:06
* FROM CLIP: RAW_CAMERA_A.MOV
* PUNCHLINE & SUMMARY: THE 3 RULES OF DETERMINISTIC AGENT DESIGN

Stage 4: 2-Click Import into DaVinci Resolve or Premiere Pro

Save Claude's EDL output as cut_timeline.edl. Now bring it into your professional video suite:

In DaVinci Resolve:

1. Open your project where RAW_CAMERA_A.MOV is loaded in the Media Pool.
2. Navigate to File > Import > Timeline... (or press Ctrl+I / Cmd+I).
3. Select your cut_timeline.edl file.
4. Resolve will ask: "Which media clip corresponds to reel AX?" Select your raw 4K footage.
5. Instant result: DaVinci instantly generates a brand-new timeline containing only the exact narrative cut points selected by Claude!

In Adobe Premiere Pro:

1. Navigate to File > Import, choose cut_timeline.edl.
2. Premiere creates a new sequence with the cuts already spliced.
3. Right-click the sequence and choose Link Media to bind the cuts back to your high-res original camera files.

Claude vs. Dedicated SaaS Tools (Descript, OpusClip, Klap)

Pro-Tips for Seamless Dialogue Edits

• Padding Safety Margins: Always instruct Claude to add 3 to 6 frames of buffer before the first spoken word and after the last syllable. This prevents clipped consonants.
• J-Cuts and L-Cuts: Ask Claude to separate audio timecodes from video timecodes so the guest's voice leads into the cut before the camera switches.
• Room Tone Preservation: In DaVinci Resolve, keep your ambient room tone on an independent audio track so the cuts between distant timecodes don't have sudden background silence drops.

Frequently asked questions