Claude Code Recreates Someone Else's Video Editing Style on Your Footage for 1.1% of a Weekly Limit
Editing a video by hand eats an evening: you watch someone else's clip whose pacing you like, then try to match the fonts, caption bars and cuts. Tekhnokafe handed the job to Claude Code, which extracts the style from a reference video and edits your footage the same way in four steps. An eight-slide video carousel used 1.1% of a weekly subscription limit, or about $11 at API prices. You can test the first two steps in an evening.
AI-processed from TG: @tekhnocafe (Технокафе / Вайбкодинг / AI-агенты); edited by Hamidun News
Editing short videos usually takes more time than shooting them. Say you like someone else's video: it has a recognizable rhythm, distinctive fonts, caption bars and cuts in the right places. Recreating all of that on your own footage by hand means an evening on the timeline.
The Tekhnokafe team has published a recipe for handing that work to Claude Code: it breaks down another creator's video, writes down its style and applies it to your material. The recipe has four steps, and all of Claude Code's work on a carousel of eight video slides took about 1.1% of a weekly subscription limit (roughly $11 at API prices).
If you make videos regularly, this is a way to turn manual editing into a repeatable pipeline.
Four steps: from reference to finished edit
Step 1. Analyzing the reference. The video is downloaded with yt-dlp and split into frames. The sampling rate varies: where the picture is static, it takes one frame every 2 seconds, and in sections with active editing it takes one every 0.2 seconds, so it catches every cut and every graphic that appears. Identical frames are discarded, so the model doesn't have to work through hundreds of repeats.
Step 2. The style guide. The remaining frames produce a description: which fonts and colors are used, how often cuts occur, how caption bars enter the screen. The reference speaker's face is removed, while the graphics stay as a template. The result is a text-based "taste map" that you can apply to any other video.
Step 3. The rough cut. It takes your footage and a word-level transcript, meaning text in which every word has a timecode. Using it, the model flags pauses longer than 0.8 seconds, duplicate takes and broken-off phrases. It does not try to cut the video itself: a script does that from the finished timecodes. This split is more reliable than asking a neural network to "fix the video": Claude makes the decisions, and ordinary code carries them out.
Step 4. Caption bars and rendering. Claude finds the places in the speech where the reference's style calls for a caption bar and puts your text there. The reference template had two items; your footage will have four, and the template adjusts. The final render runs in HyperFrames, and Claude Code ties all the steps together.
What came out of it, and what it cost
The authors went through a nine-minute video about editing in Claude frame by frame and noticed that a running check appears in it six times. That technique grew into their reel "A Check for 110 Videos." The toolset is the same: yt-dlp, ffmpeg, a word-level transcript, HyperFrames, with Claude Code tying the steps together.
They also calculated separately what the carousel of eight video slides cost. Layout and motion were done in code: HTML and frame-by-frame rendering run on their own computer. Sounds came from their own library, the reel's frame came from a finished video, and no image or video generators were used. Kira's voice was synthesized with ElevenLabs: 858 characters cost about 10 cents.
Two figures matter in practice. If you have a Claude Code subscription, the whole carousel used only 1.1% of the weekly limit, so you would barely notice it. At API prices it comes to about $11 for the piece, which is the calculation you need if you are building a paid service on this.
Limitations and what to do
This is the authors' own report, not an independent test. The post contains no editing-quality metrics: there is no comparison with a human editor's work and no viewer ratings. All we know is that the recipe produced working videos and a carousel.
There is also an ethical question. The recipe extracts style: fonts, colors, rhythm. But copying a specific creator's recognizable branded graphics one-to-one is risky, so it makes more sense to use several references and blend their common techniques.
The team has published its full skill in the guide: a SKILL.md file, scripts, a prompt for each step and a list of the places where they stumbled during testing. Start with the first two steps, frame extraction and the style guide: you can test them in an evening without rendering. If the guide turns out accurate, the rest is mechanics: a word-level transcript, script-based cutting and caption-bar layout.
- Nov 18, 2026JetBrains' Mellum Runs Locally in an IDE: 5 Out of 10 on a MacBook M3 Pro
- Nov 18, 2026OpenAI Releases Legal AI: 95% Fewer Fabricated Facts Than Claude Opus 5.5
- Nov 18, 2026Claude SEO Plugin Runs Up to 17 Agents and Audits a Site with One Command
- Nov 18, 2026ZuBrIQ gateway hides cards and UNPs from GPT and Claude, swapping in lookalike doubles even when streaming
Key Questions Answered
What are the key details on Four steps: from reference to finished edit?
Step 1. Analyzing the reference. The video is downloaded with yt-dlp and split into frames. The sampling rate varies: where the picture is static, it takes one frame every 2 seconds, and in sections with active editing it takes one every 0.2 seconds, so it catches every cut and every graphic that appears. Identical frames are discarded, so the model doesn't have to work through hundreds of repeats. Step 2. The style guide.
What came out of it, and what it cost?
The authors went through a nine-minute video about editing in Claude frame by frame and noticed that a running check appears in it six times. That technique grew into their reel "A Check for 110 Videos." The toolset is the same: yt-dlp, ffmpeg, a word-level transcript, HyperFrames, with Claude Code tying the steps together. They also calculated separately what the carousel of eight video slides cost. Layout and motion were done in code: HTML and frame-by-frame rendering run on their own computer.
What are the key details on Limitations and what to do?
This is the authors' own report, not an independent test. The post contains no editing-quality metrics: there is no comparison with a human editor's work and no viewer ratings. All we know is that the recipe produced working videos and a carousel. There is also an ethical question. The recipe extracts style: fonts, colors, rhythm. But copying a specific creator's recognizable branded graphics one-to-one is risky, so it makes more sense to use several references and blend their common techniques.
Build & Ship Production AI Agents Without Boilerplate Code
Start your free 72-hour trial at Hamidun Academy (academy.hamidun.com) — interactive lessons on Claude Code & Cursor, live Tech Radar, and production agent templates.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.