Major Update! My Claude Skills Video Editing Agent Gets Better Recognition and UI

Hi everyone, I'm Cheng Feng.

Editing Skills has been updated.

After the last version was released, I received a lot of feedback. The two most common issues were:

• Downloading a 5GB model took dozens of minutes, causing many people to give up.
• Finding verbal slips line by line in a document was too exhausting.

So I made a major overhaul this time:

I replaced the Volcengine audio/video recognition model and upgraded the interaction interface.

How's the effect? Let's compare it directly with CapCut (Jianying):

Editing Skills:

CapCut:

In the same video, the AI identified all the verbal slips that CapCut missed, and it has a better display effect.

Based on user feedback, this update mainly focuses on three aspects.

1. Faster: API Replaces Local Models

The previous version required downloading a 5GB local model, which many reported was too slow.

This time, I switched directly to API calls.

• Fast Download: No need to wait hours for the model.
• Fast Execution: API servers are much faster than your computer.

2. More Accurate: Refactored Verbal Slip Recognition

The previous verbal slip recognition was a bit crude; now the entire system has been refactored.

Model Replacement

Doubao's audio/video subtitling is the most accurate recognition I've experienced.

(Thanks to @极客杰尼 for the tip)

AI Semantic Understanding

Now the AI analyzes segment by segment:

It judges each segment based on rules: silence, repeated sentences, stuttering words, filler words...

Transparent User Habits

All recognition rules are placed in an independent folder:

If you think a rule is wrong, just edit the file. If you want to add a new rule, just create a new .md file.

3. More Intuitive: Visual Review Interface

This is the biggest improvement.

Previously, after editing, the AI would output a review draft in the command line. You had to find issues in the text and tell the AI what to change.

Now, after starting the edit, a review service starts automatically.

Open your browser, and you'll see an interface similar to CapCut's talking-head editing:

Two extra features have been added:

Feature 1: Speed Button

You can listen to the entire video at double speed.

Feature 2: Copy Deletion List

This feature is for AI learning. After confirming the review results, click "Copy Deletion List," copy the deleted segment info, and tell the AI. The AI will analyze your expression habits to identify them more accurately next time.

This is the self-updating mechanism I've always emphasized—the more you use it, the better it understands you.

The process is much simpler now, taking only 15 minutes.

Step 1: Download Skills (1 minute)

Help me download Editing Skills:

- Address:

https://github.com/Ceeon/videocut-skills

- Storage path: .claude/skills/ (Clone directly to this directory, do not create subfolders)

Step 2: Configure API Key (3 minutes)

This time we use Volcengine's "Audio/Video Subtitle Generation" capability.

It has a trial with 20 free hours:

Log in to Volcengine Model Management and enable "Audio/Video Subtitle Generation":

Get the API Key:

Enter /v, select videocut:install, enter the API Key, and complete the configuration.

Step 3: Edit Video (10 minutes)

Enter /v, select videocut:edit, and enter the video path.

The AI will automatically:

1. Call the API for transcription
2. Start the local review service
3. Provide a link to the review interface

Open the link to see the review interface:

We can check the AI's recognition accuracy here.

Operation: Shift + Drag to select or deselect segments.

Four Function Buttons:

1. Play — Like CapCut, automatically skips selected segments.
2. Speed Play — A feature CapCut lacks; you can speed up slow-talking videos.
3. Copy Deletion List — If there's a big difference, copy your deletion results for the AI to learn your habits.
4. Execute Edit — After confirming, click to automatically complete the edit.

Once finished, a video with "cut" at the end of its name will appear.

Step 4: Add Subtitles (3 minutes)

As before, enter /v and select videocut:subtitles.

Remember to add your proper nouns to the dictionary file for higher subtitle accuracy.

Four steps, fully visualized.

If you found it troublesome before, try it again this time.

Welcome to let me know in the comments.