Marketing and Content · Blog · 🇨🇳 China · via agentusecases.ai
Instructor uses Claude Code with Kdenlive and cuts a 6-hour training recording down to about a day of work
Feng Yu built a local pipeline with videocut skills, cli-anything-kdenlive and faster-whisper; work that took about 3 days fell to a little more than 1.
- Profession
- Developer, Content Creator
- Tools
- Claude Code, Kdenlive, faster-whisper
- Source
- dteam.top ↗
- Country
- 🇨🇳 China
- Found via
- agentusecases.ai (Alex Finn) ↗
- Report date
- April 17, 2026
- Verified on
- November 5, 2026
See more: AI for developers · AI for content creators
What they did
Feng Yu records online trainings of about 6 hours and needs an edited version without the interaction and review segments. Instead of only rendering a final video, he asked Claude Code for a flow that outputs a Kdenlive project for review. He combined the videocut skill (speech cutting) with cli-anything-kdenlive, replaced a third-party transcription API with local faster-whisper and fixed a bug in the project generator that blocked opening the file in Kdenlive (a fix that went into the official repository).
How it went
On a limited machine, he used the faster-whisper small model in int8 and split the audio into chunks to transcribe in parallel. Claude Code delivered the rough Kdenlive project; he opened it, listened through and finished the polish. According to the author, what used to take about 3 days now gets done in a little more than 1 day, with the slowest part (marking the timeline) handled by the AI.
Good to know
This is one person's account. Too many clips remain (Kdenlive gets heavy), the small int8 model's accuracy is limited and many speech mistakes still need review. The gain is an editable draft, not a final video without oversight.
The prompt
I have a training recording of [duration] at [path to the video or audio file]. Use a local flow: transcribe with faster-whisper (model [small or other]), mark long pauses, hesitations and segments of [interaction, Q&A or what to cut], build the cut timeline and export a Kdenlive project for me to review. Do not render the final video. Before cutting, show me the clip plan and where transcription confidence is low.
Replace what is in [brackets]. Prompt written by us from the account, not a literal copy of the source. Do not paste other people’s personal data (privacy laws such as LGPD and GDPR apply).
Summary in our own words. Results as stated in the original account; we have not tested or audited them. Read the source ↗ Case found via agentusecases.ai (Alex Finn); text and prompt rewritten by us. Found a mistake?
Similar cases
01Letting AI actually edit videos in Kdenlive
A dev created an open source MCP server that lets agents edit video in Kdenlive without touching the interface.
02Teaching AI to make your site findable by people and by other AIs
A developer created a skill for Claude Code to review the SEO and GEO of websites.
03A blog that learns the author's voice and tracks Search Console
Chudi Nnorukam uses five skills: voice profile, ranking opportunities, drafting and review.
04Weekly AI newsletter for marketers, assembled by an agent
Every Monday at 9 a.m., the agent researches the week, writes the issue, saves it in Airtable and emails the draft.
05Generate posts for real estate agents with AI
A developer created Regenc, an app that generates posts and captions for real estate agents, and ran into the Play Store's 12-tester rule.
06A requirements doc shelved for a year becomes a live website in one day with Claude Code
A tech consultant pulled out the spec for an aggregator of his own posts and shipped the site in a single day with Claude Code.
07When your AI limit runs out in the middle of the day
A dev running four subagents across four repositories ran out of tokens and reflected on their dependence on AI.
08Giving an agent a server and the mission of building a business
A dev let Claude Code run every 30 minutes on a VPS, with memory kept in files, asking for help via Telegram.