Cut long podcasts into vertical clips with Whisper, Claude scoring, YOLO reframing and n8n
An architecture write-up of an in-house media tool that transcribes podcasts with Whisper, has Claude score the best moments for each brand's audience, reframes video to vertical, and keeps humans in the loop.
Evidence: The source shows its work: steps, screenshots, code or data you can inspect.
The business problem
A media team wants short vertical clips from hours of podcast video without paying an editor to find and cut each moment.
What was tried
An uploaded file or YouTube link has audio extracted, and n8n runs transcription with Whisper on a serverless GPU and scoring. Claude rates candidate moments by fit to each brand's audience, selected moments are cut and reframed from 16:9 to 9:16 with YOLO person detection, and captions are rendered in a house style. A web editor lets people change captions and music before export, a human approves the opening hook for full episodes, and an AI reviewer checks rendered frames for caption problems. Jobs run through a state machine with a watchdog, and source video is kept until clips are safely stored.
What was reported (not reported)
The repository reports no accuracy, throughput or time-saved figures. It states the platform is in production for internal use with more than 200 commits.
Limitations
The code is private and the repository documents design only, with no licence stated. It runs one processing job at a time on a single server with SQLite, which suits internal volume but not multi-tenant use. No costs, benchmarks or lessons learned are given beyond design decisions. It was built by a media company, so a solo creator would need to rebuild it.
What you need
FFmpeg, n8n, Whisper on a GPU service, a Claude API key and a web server. Costs are not stated.
Sources
- GitHub (kleniobraz/sb-clips-case-study) ↗ Code repository, publication date unknown
Source published: unknown. Last reviewed here: October 11, 2026. Spot a mistake? Tell us.
Tools in this workflow
- Claude: Scores candidate momentsTry Claude
- n8n: Orchestrates the pipelineTry n8n
- Whisper: TranscriptionTry Whisper
Related workflows
Send a Monday business metrics email from Stripe, Notion and Sheets with Claude writing the narrative in n8n
A shared n8n template pulls last week's Stripe revenue, Notion deal pipeline and a Sheets row of support numbers, has Claude write a three-paragraph summary, and emails and posts it to Slack.
Research sponsors, build a sponsorship CRM and schedule checks with the Manus agent
A content creator used Manus to find companies advertising in AI newsletters, build a web-app CRM for YouTube sponsorships, and schedule recurring lead and trend checks.
Draft help-centre-cited support replies and propose refunds that a person must approve
An open-source support workbench classifies tickets into three tiers, drafts replies that cite help-centre articles, reads billing data in read-only test mode, and only proposes refunds that need a named approver.
Install a free pack of six Claude skills for positioning, customer research, pricing and go-to-market
An open-source set of Claude skills that builds a shared context file, then produces positioning, customer research, competitive briefs, pricing and packaging and launch plans that build on each other.