Turn Long Videos into Short Clips: Transcript Mining, B-Roll & Narration
Extract high-value segments from long interviews or webinars using Claude, generate visual b-roll in Runway, and polish audio with ElevenLabs.
Workflow Execution Pipeline
4 sequential stages from initial input to final verified deliverable.
Mine the Full Transcript for Self-Contained Insights
Clean Audio and Record Context Bridges
Generate Supplemental B-Roll Cutaways
Format Short-Line Captions and Platform Copy
Target Outcome & Deliverable
This workflow converts a 30-to-60 minute webinar, tutorial, or podcast recording into 3–5 self-contained 45-second vertical clips. Instead of relying on automated random cuts that chop sentences mid-thought, you use Claude to analyze the full timestamped transcript for complete standalone insights, Runway to generate supplemental visual b-roll, and ElevenLabs to clean audio or record tight bridge intros.
Step-by-Step Execution Instructions
Mine the Full Transcript for Self-Contained Insights
Upload the full timestamped transcript (`.srt` or `.txt`) into Claude. Ask it to extract exact start/end timestamps where the speaker explains a complete actionable concept.
Read this 45-minute webinar transcript. Identify 4 standalone segments (each 90–140 words long) that answer a specific practical question. Return a table: [Clip # | Start–End Timestamp | Core Question Answered | Verbatim Transcript].
Clean Audio and Record Context Bridges
Run noisy audio tracks through ElevenLabs Voice Isolator to remove laptop fan or room echo, or generate a 1-sentence intro voiceover to frame the clip.
Bridge script (3 seconds): "Here is how to prevent CSS cascade overrides when migrating to Tailwind v4."
Generate Supplemental B-Roll Cutaways
For each 45-second clip, identify one 5-second section where a visual cutaway illustrates the point. Use Runway Image-to-Video to animate a clean diagram or concept frame.
Gentle zoom-in on the workflow diagram card, subtle depth of field, clean steady motion.
Format Short-Line Captions and Platform Copy
Paste the final clip transcript into ChatGPT and ask it to split captions into natural 4–6 word phrase lines and write a 2-sentence post caption.
Split this 110-word clip transcript into caption lines of max 32 characters each, breaking at natural grammatical pauses.
Quality Assurance & Optimization Tips
- Sample Input: A 42-minute recorded developer workshop on API caching + exported SRT subtitle file.
- Sample Deliverable: Four 45-second 9:16 vertical MP4 clips with cleaned voice audio, 5-second b-roll visual inserts, and proofread captions.
- AI transcript analysis cannot judge visual framing errors (such as a speaker looking off-screen); always preview the video cut.
- Voice cloning or synthetic bridge audio should never be used to attribute statements a speaker did not endorse.
Common Workflow Pitfalls to Avoid
Pitfall: Workflow Bottleneck #1
Solution: Selecting clips based on loud vocal energy rather than whether the segment actually teaches a complete tip.
Pitfall: Workflow Bottleneck #2
Solution: Placing captions too close to the bottom or right edge of a 9:16 frame where platform UI buttons cover the text.