Opus 5.5の動画編集力がやばい。撮って投げるだけでYouTube始められます【この動画も全て自動編集】

Summary
Japanese content creator and community leader Shuhei (借金しゅうへい) demonstrates how Anthropic’s Claude Opus 5.5 model, driven through the CLI tool Claude Code, can execute fully automated, end-to-end YouTube video editing without opening traditional NLE editing software. He shows raw 7-minute camera footage and screen recordings being processed into a polished 4-minute video with jump cuts, subtitles (telop), b-roll cutaways, and picture-in-picture framing, while revealing that the video the viewer is watching was also edited autonomously by Opus 5.5.

---

What is shown

• [00:00 - 00:30] Opening teaser showing the resulting edited video with Japanese title badges and subtitles, alongside Claude Code logs indicating only 9 minutes of execution time and 2% weekly quota usage.
• [00:58 - 01:36] Demonstration of the output video: an unboxing and review test clip of an “M6 Mac mini,” featuring automated picture-in-picture framing, full transcription captions, slide transitions, and corner badges (“実物はこんなに小さい”).
• [02:18 - 03:02] Inspection of the raw source files: 7 minutes of unedited footage containing dead space, setup fidgeting, and an intentional test interruption where Shuhei pretended to answer a phone call, which Opus 5.5 automatically detected and cut.
• [03:20 - 03:37] Explanation of Claude account settings, selecting the standard $20/month Pro tier and unchecking the AI model data improvement toggle for privacy.
• [03:38 - 04:30] Hardware showcase: demonstrating that the task was executed on a base M2 MacBook Air with 24GB of unified memory rather than a high-end desktop workstation.
• [05:04 - 05:58] Step-by-step workflow: granting workspace permissions in macOS terminal, copying file paths directly from Finder (Option + Right Click -> Copy Pathname), and submitting them to Claude Code.
• [06:01 - 06:45] Claude Code CLI outputs: Opus 5.5 first generating a 40-second test preview video and cut list before rendering the final 4-minute MP4 via FFmpeg/Remotion scripts.
• [07:50 - 08:30] Demonstration that the entire YouTube video itself was assembled, cut, and titled by Opus 5.5.
• [08:50 - 10:57] Information on Shuhei’s “Freelance no Gakko” (フリーランスの学校) free community and upcoming AI video editing seminars.

---

Claims & numbers

• Hardware used: Presenter emphasizes this ran on an older M2 MacBook Air with 24GB RAM [00:15, 03:38].
• Subscription tier: Run using a standard $20/month (approx. 3,000 yen/month) Claude Pro plan [00:13, 03:31].
• Processing and edit times:
  • Raw footage duration: 7 minutes [00:20, 02:22].
  • Output video duration: 4 minutes [00:20, 06:18].
  • Total pipeline time: ~9 minutes (16:37 to 16:46 in terminal logs) [00:24, 06:29].
  • Machine processing breakdown: cutting/trimming ~2 min 45 sec, encoding/rendering ~1 min 30 sec, Whisper transcription ~45 sec [00:20, 06:27].
• Model token & limit consumption:
  • Context size used: ~155,000 tokens (16% of a 1M token context window) [00:20, 06:27].
  • 5-hour quota consumption: 8% [00:20, 06:27].
  • Weekly quota consumption: 2% of total Pro plan weekly allocation [00:26, 07:20].
  • The presenter estimates a creator could produce approximately twenty 10-minute videos per week on a single $20/month plan [07:35].
• Webinar signups: The presenter claims 1,200 people registered for his upcoming 60-minute free AI video seminar [09:00].
• Community scale: Claims over 7,000 members have joined his free online community [10:38].

---

Notable quotes

• [00:06] “実はこれ使ってないんです、編集アプリ全く使っておりません。実はClaude Codeだけに、えーOpus 5.5に投げました” (“Actually, I didn’t use any editing app at all. I literally just threw it to Claude Code with Opus 5.5.”)
• [06:28] “7分の動画を4分の動画に編集。それがですねなんと、本編の作業時間たった9分です” (“Editing a 7-minute video down to 4 minutes. Incredibly, the total processing time was just 9 minutes.”)
• [08:02] “この動画もまさに、Opus 5.5に作らせています。これができちゃうんですね” (“This video you are watching right now was also created by Opus 5.5. It can actually do this.”)

---

Assessment
This is a genuine practitioner tutorial and technical demonstration showcasing automated programmatic video editing (orchestrating FFmpeg and Remotion scripts via Claude Code CLI powered by Claude Opus 5.5). The claims are directly corroborated on screen by actual terminal outputs, file structures, raw footage comparisons, and the automated captioning/layout style of the uploaded video itself.

---

Lyrics & themes
(Spoken narration / technical walkthrough)

• Introduction & Proof of Concept [00:00 - 01:50]: Announcing that Claude Opus 5.5 can handle end-to-end video editing without human intervention in NLE software like Premiere Pro.
• The Experiment Setup [01:51 - 03:19]: How deliberate raw errors (phone calls, long silences) were fed into the model to test its editorial judgment.
• Hardware & Cost Efficiency [03:20 - 05:03]: Proving that expensive cloud rigs or high-end Mac Studios are unnecessary; an M2 MacBook Air and a base $20/mo Claude Pro tier suffice.
• Execution & Results [05:04 - 08:08]: Demonstrating path-copying, preview generation, full rendering, and the sub-10-minute turnaround.
• Industry Impact & Education [08:09 - 10:57]: How AI video editing is shifting video creation from manual assembly to prompt orchestration.

---

Lore & references

• Claude Opus 5.5: Anthropic’s flagship frontier reasoning model released in September 2026.
• Claude Code: Anthropic’s official agentic terminal coding tool, used here to programmatically run local CLI utilities (FFmpeg, local transcription scripts, and Remotion video-in-code rendering).
• M6 Mac mini gag: Shuhei reviews a fictional future “M6 Mac mini” on his test screen to demonstrate technical commentary handling.
• Data Opt-out toggle: Showing Anthropic’s “Help improve AI models” toggle in account settings to protect raw creator footage from training datasets.
• Weekly Usage Limit (週の利用枠): Referencing Anthropic’s subscription rate-limiting architecture introduced for heavy agentic coding workloads.

---

Visual style & craft

• AI-Generated Editing Pass: The editing format follows a recognizable Remotion / code-rendered layout: programmatic round avatar picture-in-picture in the bottom-left corner during screen shares, dynamically placed white subtitle boxes with black text (“フルテロップ”), and automated red/black section banners in the upper corners.
• Jump Cut Pacing: Tight conversational cuts matched to speech boundaries generated automatically from audio transcripts.
• Studio Setup: Filmed in a well-lit home library/office with wooden bookshelves, an external Shure microphone, and an M2 MacBook Air on a stand.

Described by gemini-3.8-flash on 2026-10-16 from the video’s audio and frames.