What the Auto Editor does for you
A raw take is slow: long pauses before you start a thought, "um" and "uh" between phrases, and one flat shot for minutes at a time. The Auto Editor fixes those in a single run, and shows you every change before anything is rendered.
- Cut pauses. It measures the loudness of the audio, works out where the quiet is for your recording, and trims the gaps. Natural only shortens long gaps, Tight removes short pauses too, and Fast leaves almost no air between phrases.
- Cut filler words. A speech model transcribes the audio on your device, and the filler words it finds (um, uh, erm, er, ah, hmm and mm) are cut with a few hundredths of a second of padding.
- Jump-cut zoom. The picture alternates between 100% and 110% at every cut, centered a little above the middle where faces usually sit. It is the trick that makes jump cuts look like deliberate changes of shot.
- Captions. Optionally burn in captions, re-timed to the edited video, in the same styles as Auto Subtitles. You can save the captions as an SRT file too.
- Loudness. The sound is leveled to -14 LUFS, the loudness YouTube and Spotify play at, with the peaks kept under -1 dB.
Filler words: what to expect
Finding filler words is harder than finding silence, because a computer has to understand what was said. ClipLime uses an open-source speech model, Whisper, and cuts the filler words that show up in its transcript.
Here is the honest part: speech models are trained to write clean text, so they often leave "um" and "uh" out of the transcript. A filler that is not in the transcript can't be cut automatically, so a few may slip through. Open What the speech model heard to read the transcript, with the fillers it found highlighted.
- Scrub the timeline and play the edit with Skip removed parts on. It takes a minute and catches anything that is left.
- The model's word timings can be off by a few tenths of a second, so each cut is placed on the actual sound of the filler. A filler said right inside a phrase, with no gap around it, is left in rather than risk cutting a real word. Click any cut to keep it.
- Filler words are English (plus a few common French and German sounds). Other languages are transcribed for captions, and pauses are cut in any language.
The speech model is downloaded once and your browser keeps it. It is about 41 MB for the Fast model on a processor, and larger for the Balanced and Accurate models or when your graphics card is used. The tool shows the exact size before you start. Your audio is processed in this tab and is never uploaded.
Cut, zoom and captions in one pass
Many tools cut the video, save a file, then open that file again to add zoom and captions. Every save re-encodes the picture, and every re-encode costs a little quality and a lot of time. The Auto Editor does it in one pass: each kept part of your video is decoded once, the frame is zoomed and captioned on a canvas, and the result is encoded straight into the MP4.
The sound is cut and leveled first, then encoded alongside the picture, with a short fade at every join so cuts never click. Everything is rendered by your own computer, using your hardware video encoder when the browser offers one.
Finish the edit in Premiere Pro or DaVinci Resolve
If you edit in a full suite, you don't need the MP4. Download the EDL and import it into Premiere Pro, DaVinci Resolve or Avid, link it to your original file, and every cut appears on your timeline. The cut list is a plain-text list of every pause and filler word that was removed, with its time.
The EDL and the cut list contain the cuts only. The zoom, captions and loudness are applied in the exported MP4. For a different shape, continue with Resize for Social, or shrink the file with the Video Compressor.