#Transcription options

CutZal finds the cuts by reading the transcript. A transcript is what you said, written down word by word with timing. If a video has no transcript when you open it, the Transcription needed notice appears. It lists three methods side by side, with their cost and how long they take. Choose one and click the button below them. The method that suits you best right now is marked "Recommended" and already selected. A method you can't use shows the reason in its place.

#At a glance

CutZal transcription (credits or your own key) is the most accurate. Free local transcription is less accurate, and imported subtitles only let you cut line by line.

CutZal transcription uses ElevenLabs speech recognition (Scribe). With credits, it uses your CutZal credits. With your own ElevenLabs key, it transcribes with that key and ElevenLabs bills you. ElevenLabs detects the video language on its own.

#Import subtitles

If you already have a subtitle file, choose Import subtitles and click Choose subtitle file to pick it. This works well when another app or a transcription service already made subtitles for you.

CutZal treats each subtitle line as one piece. So cut boundaries fall on subtitle lines rather than words, and cutting a single word inside a line is hard. Cuts between lines, such as silences and retakes, work well. The shorter the subtitle lines, the more precisely you can cut.

#CutZal transcription credits

1. In the notice, choose CutZal transcription and click Sign in with Google. You sign in through your default browser, and you're taken back to CutZal when you're done. If you aren't taken back automatically, paste the sign-in code shown in the browser. If you closed the browser window, open it again with Reopen browser in the same place. To continue without signing in, click Cancel sign-in.
2. Once you're signed in, the CutZal transcription row shows, in minutes, the credits this transcription will use and the credits you have left.
3. When you've checked them, click Transcribe with CutZal. The audio goes to ElevenLabs through the CutZal server, which doesn't store it.

Signing in gives you free transcription credits every month. If you don't have enough credits, you can't choose that row, and it tells you when new credits arrive. The only thing sent is the video's audio. You can edit and transcribe (free local transcription, imported subtitles) without signing in. You need to sign in to use credits and to export a video.

• If transcription fails, the credits it used come back.
• Once transcription starts, you can't stop it partway, and you can't close the app until it finishes. Very long videos can't be transcribed in one go, so import them in parts.
• If the service's total allowance for the month runs out, transcription can be blocked until the next grant date, even if you have credits left. The notice shows the date it opens again.

#Your own ElevenLabs key

This is for developers with an ElevenLabs account and API key. With a key in place, CutZal transcribes with your key instead of CutZal credits, and ElevenLabs bills your account.

1. Create a .env file in the .cutzal folder in your home folder and add one line: ELEVENLABS_API_KEY=your-key. For CutZal Beta, the folder is .cutzal-beta.
2. Reopen CutZal.
3. When the CutZal transcription row says "charged to your ElevenLabs key", click Transcribe with CutZal.

If a key is present, CutZal always uses it first, whether you're signed in or not. Keep the key file on this Mac only. Don't share it or put it in a repository.

#Free local transcription

This transcribes on this Mac at no cost. It needs no sign-in or credits, and the audio never leaves your Mac. It works on Apple silicon Macs.

1. In the notice, choose Free local transcription. It also shows roughly how long it takes on this Mac.
2. The Video language field holds the language spoken in the video. If you already have the model, CutZal listens to a few spoken parts of the video on this Mac, detects the language, and fills it in ("Detected from the video" appears next to the field). If it isn't sure, it leaves the field empty for you to choose. If it's wrong, change it right there. A language you choose yourself is selected in advance for your next video. It transcribes 14 languages, including Korean, English, Japanese, and Chinese.
3. The first time, it downloads the local transcription model (about 1.9 GB) from Hugging Face. No audio or account information is sent. The button below reads Download and transcribe. Click it to start the download, and transcription begins as soon as the download finishes. If you didn't choose a language, CutZal detects it after the download and then transcribes. If it can't tell, it asks you to choose one.
4. While it downloads, the row shows the progress, and you can stop with Cancel download. Click again and it resumes from where it stopped.

Once the model is downloaded, click Start free local transcription to transcribe right away.

After you have the model, it transcribes even without internet. Once transcription starts, you can't stop it partway.

It's less accurate than CutZal transcription. It misses some fillers and doesn't mark coughs or keyboard sounds. So the default cut can't find those sounds, and you need to check them yourself as you play. If you have plenty of credits or your own key, CutZal transcription is marked "Recommended" and selected first. If you're short on credits or not signed in, free local transcription is. You always decide which one to use, and nothing starts until you click the button.

You can see the size of the downloaded model in the Local transcription model field under Models in Settings, and remove it with Delete. If you remove it, it downloads again the next time you choose free local transcription.

#When money is spent

CutZal only starts work that costs money when you click for it yourself.

• Opening or importing a video doesn't start transcription. It only runs checks that cost nothing (length, sync, looking for an existing transcript).
• If the same video already has a transcript in the edit/transcripts/ folder next to it, CutZal loads that instead of transcribing again. Once a video is transcribed, making a new project from it doesn't cost money again.
• Credit transcription and transcription with your own key only happen at the moment you click Transcribe with CutZal.
• Free local transcription also starts only when you click the button. The model downloads only when you click Download and transcribe, never on its own when you install or first open the app.
• If transcription fails, CutZal doesn't switch to another method on its own. It stops and tells you why.
• On the screen where you choose how to transcribe, Start Auto cut when transcription finishes is off by default. Only if you turn it on does Auto cut start as soon as transcription finishes and use your Claude subscription allowance.

Besides transcription, the things that cost money or allowance are Auto cut and chat (your Claude Code or Codex subscription allowance) and creating AI sound effects. Chat and your agent and Layout, B-roll and sound effects cover them.