I spent a week using video editing Skills to build a video editing Agent.
It feels amazing!
It can automatically edit a 30-minute video in just 10 minutes.
Video editing Agent demo: automatically editing a 30-minute video in 10 minutes.
I often use CapCut to edit talking-head videos, but after using it for a long time, I found several problems.
Problem 1: Smart talking-head editing does not understand meaning
Because it cannot understand the meaning, it sometimes fails to identify repeated sections. If I speak continuously for 20 or 30 minutes, editing the video myself becomes exhausting.
Problem 2: The subtitle quality is poor
The automatically generated subtitles contain many incorrect words and typos.
So I used the Skills feature in Claude Code to build a video editing Agent.
The fundamental difference is simple:
CapCut vs. Agent: a fixed tool vs. an adaptive assistant.
The key difference is:
CapCut = fixed tool + manual operation
Agent = adaptive system + automatic learning
I am not replacing CapCut with a better algorithm. I am replacing it with a system that can continuously improve itself.
But that is not even the most impressive part.
The most impressive part is this: the more I use it, the better it understands me, and the faster it becomes.
Three Core Designs
1. Agent Logic
It only takes four steps.
Video editing Agent workflow: from the video file to the final video.
2. The Skills System
At first, I put every function into one large Skill. I had to add instructions to distinguish between different tasks, which was very inconvenient.
Now I have separated the five core video editing tasks into five independent Skills and placed them in the .claude/skills/ directory. This makes the structure clearer and the tasks easier to select.
When I enter /v, Claude Code automatically lists the five available Skills.
The list of five independent Skills.
I select one, and the AI runs that Skill. Simple, right? A manual task that used to take 10 minutes now only requires selecting an item from the menu.
All of the methods are written into the Skills, so I do not need to explain them again every time.
The main reason for separating them into independent Skills is that people need to inspect the output from each stage. For example, I need to check whether the review draft is correct before confirming that the Agent should perform the actual edit.
3. A Self-Updating System: It Understands You Better the More You Use It
This is the design I am most proud of.
After each task, I can give the AI feedback, and it permanently saves that feedback into the Skills.
The self-update loop: a feedback-learning system that understands you better over time.
The key point is that the Skills gradually change from a general set of rules into a customized solution made specifically for you.
Use it 10 times, and it will understand 80% of your habits. Use it 50 times, and it will fully match your requirements.
The more you use it, the more personalized the Skills become and the better they understand you. That is the power of self-updating.
How to Use It
Step 1: Download the Skills
Open a new folder. You can use this folder as a dedicated workspace for video editing in the future, which makes everything more convenient.
Ask Claude Code:
Help me download the video editing Skills:
- Repository: https://github.com/Ceeon/videocut-skills
- Destination: .claude/skills/
- Clone them directly into this directory without creating another subfolder.
It will download the Skills automatically, and the process is very quick.
The Skills being downloaded automatically.
Restart Claude Code, and you will be able to see the Skills.
All Skills displayed in Claude Code after restarting.
Step 2: Install the Environment
Enter /v and select videocut:install.
Selecting the videocut:install command.
The AI will automatically install the dependencies and download the models, which are about 5 GB in total:
- FunASR: used to identify verbal mistakes.
- Whisper: used to generate subtitles.
The two models have different strengths. FunASR is suitable for word-level editing, such as identifying verbal mistakes and filler words. Whisper produces better subtitles.
Supported AI models can also be accessed through WisGate.
Step 3: Edit the Talking-Head Video
Enter /v, select videocut:edit-talking-head, and then give the AI the path to your video file.
Selecting videocut:edit-talking-head and entering the video path.
The AI will automatically:
- Transcribe the video.
- Identify verbal mistakes by checking every sentence without missing any.
- Identify filler sounds such as "um," "ah," and "eh."
- Identify silences of at least one second.
- Generate a review draft.
An example of the review draft generated for a talking-head video.
In the review draft, the pause symbol indicates a silent section that needs to be deleted. A red dot followed by a wavy line indicates a repeated section that also needs to be deleted. If you prefer another format, you can ask the AI to change it to match your requirements.
The key step is to review the draft after it has been generated.
If anything is unsatisfactory, such as wanting to keep more filler words or correcting a word that was recognized incorrectly, continue talking with the AI until the result is right.
Then enter /v, select videocut:self-update, and the AI will save the adjustments from this task into the Skills.
After you confirm the draft, it automatically performs the edit.
Step 4: Add Subtitles
The key point is this: a custom dictionary can double subtitle accuracy.
The Skills include a dictionary file:
.claude/skills/videocut:subtitles/dictionary.txt
Open the file, and you will see that it already contains some predefined terms.
You only need to add your own proper nouns, such as your company name, product name, and channel name:
Cheng Feng's Channel
AI Product Freedom
My Brand Name
By adding the terms to the dictionary in advance, the AI can use it to correct recognition errors automatically when generating subtitles. The subtitles can be accurate from the first run.
Then enter /v and select videocut:subtitles.
The AI will automatically:
- Transcribe the video with Whisper.
- Use your custom dictionary to correct incorrect words.
- Generate a subtitle file.
- Ask you to confirm whether the subtitles are correct.
- Burn the subtitles into the video after confirmation.
If you later discover that a term is missing from the dictionary, you can select videocut:self-update at any time to add it.
Four steps, and the entire process is automated.
Editing Result
The final result is almost the same as editing the video manually in CapCut!
Summary
If you often edit talking-head videos, I strongly recommend giving it a try. Make sure to use the self-update system, because that is how the Agent comes to understand you better over time.
If you have more complex requirements or run into problems while using it, you are welcome to join the group and discuss them.
Join the discussion on Discord: https://discord.gg/GjqHPC4U2t.
Top comments (0)