As a content creator, especially if you work on interviews, podcasts, or shareable videos, you've almost certainly run into this frustrating workflow:

AI-transcribed subtitles have typos that need fixing; you have to delete all the "umms" and filler words from speech; if you want bilingual subtitles, you also need to adjust the timeline to match reading speed. After transcription, you have to jump to another tool for proofreading, then export it back to your editing software again. Changing a single word requires realigning the entire timeline, and tinkering with a few spots can easily eat up half an hour.

Recently, developer Baoyu released BaoCut, built specifically to solve this problem. It is currently only available for Mac platforms.


Its core design concept is very simple: it treats text as the single source of truth, and you edit subtitles just like editing a document: press Enter to split a line, press Backspace to merge two lines. After you fix a typo, the timeline automatically adjusts to match the text, no manual adjustment required.

The entire workflow from raw footage to finished subtitles is split into three steps, and you can manually edit and intervene at any step:

### Step 1: Local transcription with automatic speaker diarization

All speech-to-text conversion and speaker recognition runs locally on your Mac, using either the Qwen3-ASR or Whisper model, running on Apple Silicon's MLX framework. Your audio and video files are never uploaded to third-party servers, so your privacy is fully protected. Transcription results can provide timestamps accurate to each individual word, far more precise than results that estimate timestamps per full sentence.

### Step 2: Generate aligned bilingual subtitles

Translations are aligned sentence by sentence, you can edit one pair of source text and translation at a time. The tool also automatically controls the character speed per subtitle line, preventing subtitles from appearing too fast to read, and avoids splitting a single sentence into overly fragmented segments that hurt readability. After generation, you can review each sentence line by line before applying changes, and it will never overwrite your original content directly.


### Step 3: Editing text equals editing video

When you delete filler words and repeated content from the transcribed text, the software will automatically cut the corresponding segments from the video. It also supports reversible soft cutting, so you can restore the original footage at any time if you change your mind. It also supports adding B-roll to cover the original footage. Finally, you can directly export a finished video with embedded subtitles, or export subtitle files in SRT, VTT, ASS, Markdown and other formats for archiving.

Beyond manual editing, BaoCut's biggest feature is its compatibility with modern AI Agent workflows.

Many people use agents like Claude Code and Codex to process content. Previously, after agents finished transcribing subtitles, there was no user-friendly interface for people to do secondary proofreading and editing. BaoCut has a custom-built Agent Skill, allowing agents to call BaoCut's capabilities via the command line to automatically complete transcription, polishing, and translation. The progress is synced to the GUI in real time, and you just preview and make final edits directly in the interface when it's done.

After installing the app and the skill, you just enter a command in your agent to trigger it. For example, saying "/baocut transcribe and translate this video" is enough to start the process.

Many users have pointed out that this tool hits exactly on real-world needs: when clients ask for subtitle edits, the worst part is getting stuck in the final five minutes—if a name is wrong or you deleted too many filler words, fixing it messes up the timeline and you have to rewatch the whole video. Letting AI do the foundational work and leaving users with an interface that supports previewing and undoing changes is far more practical than chasing a single perfect generation.

Currently known information:

- Only supports macOS 15+ on Apple Silicon; Windows support will be decided based on user demand in the future

- Translation speed is relatively slow, but translation quality is good

- Free to use, no account registration required, all content is stored locally

- Supports exporting subtitles in Markdown format, which is very convenient for archiving video content

Tool official website and download link: [https://baocut.app/](https://baocut.app/)

Agent Skill open source repository: [https://github.com/jimliu/baocut](https://github.com/jimliu/baocut)

Installing the Agent Skill can be done with a single command, and it is compatible with Node.js environments. If you don't have Node.js installed, you can also clone and install it manually. Please refer to the instructions in the GitHub repository for detailed steps.