Edit a recording with AI

Everything the video editor can do, you can also ask for in words. Open a recording in your Library, click “Edit video”, then the AI button at the top of the rail. Type what should change: cut the first ten seconds, blur the email field, add captions, zoom in on each click, add a title card that says Quarterly review. The AI proposes, the preview plays the result, and you keep it or undo it. Nothing is saved until you click Save, like any other edit.

AI editing is part of the Web app plan and is off until you turn it on in Settings › AI editing on catu.app.

What it can do#

• Cut and trim: cut from 0:12 to 0:20, tighten the start and end, remove the silences, remove the filler words.
• Point at things: add an arrow at the Save button, highlight the pricing table, zoom in on each click.
• Hide things: blur the email address whenever it shows.
• Say things: add a title card, add a tooltip that says Click here, add a button that opens catu.app, captions from the speech.
• Dress it: add a small confetti GIF at 0:10, a thumbs-up sticker in the corner, a party emoji when the upload finishes, a photo of a city skyline behind the title. GIFs and stickers come from GIPHY, photos from Pexels.
• Stock video: an ocean stock video under the title card, a small city clip in the top-right corner. Videos come from Pexels. A stock video fills the picture unless you say where it goes, and it is silent unless you ask for a volume, so it never talks over the recording.
• Make a picture: make a sticker of a smiling coffee cup, create a picture of a rocket for the intro. The AI makes a picture or a sticker (OpenAI, through catu) only when you ask for one; otherwise it looks for a GIF, sticker or photo first.
• The picture: crop it to a square, make it vertical for Reels, shrink the recording to the left half, make the recording half see-through.
• The look of things: make the captions larger, yellow on black, put the title in a serif font, extra large, round the button's corners, put the stock video behind the recording, play the voice-over a bit faster. The AI sets the same things the editor's panels do, including the seven sizes from XX-small to XX-large, colours, backgrounds and their transparency, shadows, outlines, rotation, layer order, volume, speed and ducking.
• Narrate: add a voice-over at the start saying Welcome, narrate the captions. It speaks in the voice you chose in the editor's Audio panel.
• Music and sounds: add calm background music, upbeat music under the intro, quiet, a click sound at 0:04. The AI picks from catu's own library first (cleared for any video); when nothing fits it makes a piece to your words (ElevenLabs), instrumental, up to 3 minutes, or a sound effect of a few seconds. In the editor, both are in the Audio panel.

It can't add your own files: add those in the editor's panels.

The chips under the field are the usual asks; the ones about speech and clicks appear when the recording has the notes for them (below).

If something it chose could not be fetched or made, the rest of its proposal is still applied, and its answer lists what was skipped, and why, after Left out:.

What the AI sees#

Only when you type an ask, your browser sends: your words, the recording’s edit list, its length and size, a few small still pictures it takes from the video (one every few seconds and around the moments your ask names), and, when the recording has them, the notes your Mac made at recording time. Never the video file, the audio, the title or the key. When it needs a GIF, sticker, photo, stock video or sound, it searches GIPHY, Pexels or catu's library with a few words of its own; your browser fetches what it picks and keeps it with the recording, encrypted, like a file you add yourself. The result comes back as a list of small operations the editor applies itself; the AI cannot save, delete or touch anything.

Each ask counts against a daily budget. The panel shows how much of today’s budget is used; it resets within 24 hours.

Notes your Mac makes#

With “Notes for AI editing” on (Settings › Privacy in the Mac app; on by default), the Mac writes down two things after a recording is shared, on the Mac itself:

• a transcript of the speech with the time of each word, made with Apple’s on-device recognition (macOS 26 or later; the first time, a speech model may download once);
• the clicks, scrolls and pointer positions inside the recorded area while you recorded. Never which keys you pressed.

Both are encrypted with the recording’s own key and stored with it, so the editor can make captions from the speech, the AI can remove silences and filler words exactly, and “zoom in on each click” knows where you clicked. Videos uploaded from the catu menu get the transcript too; videos uploaded from the web have no notes. For those, the Captions card offers “Generate from the video” (the audio goes to OpenAI through catu once; see privacy).

Keep or undo#

A proposal is applied to the preview at once and the player jumps to the first change. Keep adds it to your edit history; Undo throws it away. Either way, the recording is unchanged until you click Save. Follow-up asks understand the last one: a bit larger, not the second one.

The Transcript tab#

Below the player, next to Comments, a Transcript tab shows the recording’s speech in paragraphs with timestamps. Click a paragraph to jump there; copy one, copy all, or download it as text, SRT or WebVTT. The small switch in the tab decides whether people who open your link see it too; it is off until you turn it on.