Claude Code: Tạo Skill Edit Video Tự Động, Hướng Dẫn Từ A-Z Hoàn Toàn Miễn Phí

Summary

The video explains how to set up and customize an automated video-editing skill for Claude using an open-source GitHub project rather than buying a pre-made skill. The presenter argues that paid skills may not suit every creator’s footage or editing style, and that a polished demo does not guarantee good results.

How the setup works

• A Claude “skill” is a guide file containing structured, step-by-step instructions. It can be invoked by name or selected by Claude based on a request.
• The tutorial uses a video-editing project found on GitHub, which the presenter says has more than 28,000 stars and is open source.
• The presenter assumes viewers already have the Claude desktop app and a paid plan. They select Claude’s coding mode, create a project folder, and paste in the repository’s setup prompt.
• FFmpeg is installed as a dependency. The setup requests an ElevenLabs API key, but the presenter skips it because they do not need AI voice generation.
• The presenter recommends checking GitHub skills’ source code for safety, since malicious code could be included.

Editing workflow and results

1. Basic rough cut: The presenter uploads a roughly two-minute raw clip and asks Claude to remove long pauses. The tool analyzes the video’s duration, frame rate, silence, and audio level, then estimates a shorter output. This approach detects silence but does not understand what was said, so it cannot reliably identify stutters or repeated phrases.
2. Speech-to-text integration: To help identify verbal repetitions, the presenter uses Vbee’s speech-to-text API as a substitute for ElevenLabs. They create a Vbee app ID and token and add them to the project’s .env file. The tutorial warns that the token can authorize paid API features, so it should be kept private; a time-limited token is suggested as a safer option.
3. Automated edit: The presenter asks Claude to remove pauses, filler sounds, and repeated sections. The tool processes the approximately two-minute clip for over 12 minutes, with a further estimate of four to five minutes while processing. The resulting rough cut is reported to be around 35 seconds.
4. More advanced additions: The presenter then requests subtitles, B-roll, graphics, and sound effects. Claude asks for clarification about the B-roll and offers recommendations. The final demo includes subtitles, relevant B-roll inserts, graphics, and sound effects.

Review and practical takeaways

• The presenter is pleased with the first edit, saying it removes long silences and makes cuts around stumbles and repetitions feel smoother than a manual rough cut.
• The more advanced edit is also received positively, particularly for its subtitles and B-roll related to the spoken content. The presenter notes that knowing video-editing terminology can help users request more specific effects.
• Processing time may be substantial even for a short clip, so longer videos—such as 20–30 minutes—could take longer.
• The skill is presented as customizable rather than universally suitable: users should specify their footage and desired style, and adapt the instructions to their needs.
• Although the skill itself is free to obtain, the workflow is not necessarily cost-free: the presenter assumes a paid Claude plan, and API services such as Vbee may consume tokens.

Main speakers and sources

• Main speaker: The presenter from the Học AI channel, demonstrating the setup and reviewing the edits.
• Tools and project discussed: Claude/Claude Code, the GitHub video-editing skill repository, FFmpeg, Vbee speech-to-text, and ElevenLabs as an optional integration.