How Tofucut works

New Project

New Project asks one thing: Record or Blank Project. Record opens the recorder page; ‹ New Project at its top takes you back. Blank Project starts empty, for your own videos or just your voice (see below). Under the two, “Already recorded? Start from your videos” brings in videos made with another app. ⇧⌘R still starts recording straight away.

The recorder page

The recorder is one column: two switches, your brand, the screen and camera cards, the inputs and a round red Record button. Press Record (or ⇧⌘R, or Return) and teach. When a sentence comes out wrong, say it again and keep going; the edit takes care of it. Save as Default keeps your sources, devices and brand, so every launch starts from your setup.

Screens and camera

Each display card shows a live picture of the screen it records. Add a second or third display with the + tile; each one goes to its own file. The screens record at 30 or 60 fps. The camera card shows your camera live and records at 1080p, or 4K when the camera offers it. The live camera, the meters and the screen previews run only while Tofucut's window is open and in front: switch to another app, minimize or close the window and the camera light and the microphone go off. Flip mirrors the previews and the camera in new projects; the recorded file is never mirrored.

A four-second camera check tells you when your camera's frame rate won't look smooth, and how to fix it (often by setting the camera to 30 fps).

Microphone, system audio, clicks and keys

Your voice and your Mac's sound record on separate tracks, so you can fix one without touching the other. Level meters move before you record, so a silent microphone shows up before the take and not after it.

Clicks and keys records your mouse and keyboard. With it on, the screen is recorded without the pointer, and the editor draws a smooth cursor that zooms can follow. Key presses are not noted while a password field has the focus.

Show Tofucut, and the floating recorder

Show Tofucut in the recording decides whether Tofucut's own windows and sound appear in the video; it's off by default. Floating recorder hides the main window while you record and leaves a small tag at the edge of the screen with the time, your camera and the controls. Neither the tag nor the thin red lines that mark what's being recorded ever appear in the video.

Notes while you record

Notes opens a small window with your talking points. It floats above every app on the recorder page and while you record, and it is never in the video. Write holds your notes as Markdown; Read shows them styled, and a task (- [ ]) can be ticked off there as you go. Text in Read can be selected and copied. The two letter buttons change the text size, the top bar drags the window, and any edge or the corner grip resizes it. Your notes, the size and the place are remembered on your Mac. While recording, ⇧⌘N or the note button on the floating recorder shows or hides it.

Pause, restart, discard

While recording, Pause (⇧⌘P) stops every file until you Resume; the paused time isn't in the files and everything stays in sync. Restart puts the take in the Trash and starts again with the same setup. Discard puts it in the Trash and stops. Both ask first. Stop opens the new project in the editor.

What a recording writes

Each recording is a folder in your Movies folder: one video file per screen, the camera, the microphone and system audio as their own files, your clicks and keys, and a manifest that keeps them on one clock (in sync within 40 ms). Tofucut never modifies, re-encodes or deletes these files. Your edit lives next to them in a small edit.json, so the original is always there to go back to.

If your Mac sleeps, the main display is unplugged or the disk drops under 2 GB, the recording stops and every file written so far stays playable. Tofucut then plays an alert sound and shows a "Recording stopped" window with the reason, above every app on the screen you are using, until you click OK. Ending the recording from macOS's own Stop sharing control finishes the take normally; the project shows it as interrupted, not failed.

If the camera or microphone file can't be made, the recording goes on without it and a warning row says so. A microphone that only starts delivering sound late is still recorded. Record never waits more than about 8 seconds for a camera or microphone that hangs.

While you record, the project list shows the take as a “Recording…” row. It can't be opened, selected or moved to the Trash until you press Stop.

Start from videos you already have

Recorded with another app? Choose File › New Project from Files, click “Already recorded? Start from your videos” on the New Project page, or drop the videos on the project list. Bring your screen video and, if it is a separate file, your camera video, and say which is which. A Loom or Screen Studio video with you already in the picture comes on its own as “Screen + camera”.

Tofucut lines the two up by the sound both of them heard, so a camera started a few seconds late still plays in sync. If only one has sound, or they don't match, it asks how much later the camera started. Your voice comes from the clearer of the two, or the one you pick. The videos are copied into a new project and the originals stay where they are; from there it works like a recording: Edit for Me, layouts, zooms from what you say, export. Click zooms and the smooth cursor need a recording made in Tofucut. It opens .mov, .mp4 and .m4v; from OBS, record or remux to MP4.

A video with no recording

Blank Project (or File › New Blank Project, ⌥⌘N) opens an empty project with three tracks, each for one thing: Screen (your screen recording, slides or pictures), Camera (you on camera) and Voice. An icon on each track says which is which. Drag your own files from the Finder onto their track (they are copied into the project) and they play that part, so the layouts put your camera video where the camera goes. A blank project keeps one Screen and one Camera track.

No video at all? Click the record button on the Voice track, say what the video is about and press Space to stop. Then, in the Animate tab, Make the Animations from Your Voice transcribes it, splits it into scenes that cover the whole video and makes one animation for each, in your brand, standing where the screen would be. The AI shows the big moments full frame and the rest in your brand's frame; turn off AI framing per scene in the Brand tab's layouts to keep every scene in the frame. Edit for Me does the same as its “Animate from your voice” step. Both are offered only while the Screen and Camera tracks are empty.

The AI panel

Type what you want the way you'd brief an editor: “Edit this with the Tofucut branding”, “cut the retakes and the long pauses”, “put a name card when I introduce myself”, “add an animation of the three steps at 2:10”. Return sends. Paste a picture with ⌘V to show what you mean. A message takes at most 10 pictures, and each pasted picture is shrunk to 2000 pixels on its long side.

The box starts with the playhead's time, like [3:14], and it follows the playhead as you move it or play the video. It is sent as the start of your message, so you can park the playhead somewhere and write “make this part shorter” or “zoom in here” and the AI knows where you mean. You can't type over it or delete it; switch Keep time off below the box and it is gone, and nothing is added to your messages.

Asking for the whole edit

One request can cover the whole first cut: silences, filler words, retakes, layouts, zooms, lower thirds and animations. The AI works on the real project through Tofucut's own tools, and each change appears on the timeline within a second, where you can keep it, change it or undo it. You can keep editing while it works; your changes and the AI's are saved together. Each project keeps its own chats.

⌘Z undoes your own edits only. It doesn't step back over what the AI or a command changed. When a change can't be merged with yours, Tofucut shows the other version and keeps yours in Snapshots as “Your edit before the merge”.

The AI works inside the project: it can read the project, its recordings and Tofucut's animation kit, not your other folders, and it can't start a recording. On ChatGPT, a custom server or the Tofucut plan it also runs without network access, so the steps that call the AI or transcription themselves (Edit for Me's AI steps, Zoom, Highlight and Lower thirds with AI, building animations) run from the app's own buttons, not from a request in the AI tab.

Which AI it uses

You choose the AI once, in Settings › AI, and every “with AI” button in the app uses that choice. The first choice is the Tofucut plan, which is coming: Tofucut will run the AI and the transcription for you, with nothing to connect and no other bill. It opens with the launch.

Or connect the AI you already pay for, on the Free plan. Claude runs through Claude Code on your Claude plan or an Anthropic API key. ChatGPT runs through OpenAI's Codex on your ChatGPT plan or an OpenAI key. Custom works with any server that speaks OpenAI's Responses API, such as OpenRouter, Ollama or LM Studio, local models included. Those companies bill you for what the AI uses. A custom server's address must start with https:// (http works only for a server on this Mac), and after you change the address the key is asked for again.

Tofucut checks that claude and codex are signed by Anthropic and OpenAI before it runs them. Another copy, for example one installed with npm, shows “Found, but not verified” in Settings › AI with a Use This Program button, which you press once. The first Install asks before it downloads and runs Anthropic's installer.

Every step, one click

Edit for Me, in the window's toolbar, runs every step of the edit in order, as if you clicked each one yourself: cut silences, transcribe, find fillers, find pauses, find retakes, apply the cuts, clean up the voice, lay out, zoom, highlight and touch up the camera. Pick the steps once (it remembers them) and press Start.

While it runs, the button reads “3 of 11 · Find retakes” and the log shows what each step did, how long it took and where its result lives. A failed step doesn't stop the run, and Undo All puts the edit back as it was, in one step. An AI step can also end with “The edit changed while this ran. Run it again.” That happens when you cut the timeline while it was working and its result no longer fits.

Silences

Silences are found on your Mac the moment the tab opens. Cut silences removes them all in one step, before transcription, so less audio is sent and billed. Restore silences puts them back. The silence settings hold the threshold, the shortest silence that counts (0.7 s by default) and how much room to keep around speech.

Transcribe

Transcribe sends only the voice audio that's still on the timeline and has no transcript yet to ElevenLabs Scribe v2, with your own key (about US$0.22 per hour of audio). The language is detected. The transcript is saved in the project and never billed twice.

Fillers, pauses and retakes

Three switches mark what to cut. Fillers marks um, uh, er and hmm. Pauses marks every pause longer than your limit, from 150 to 1,000 ms. Retakes asks the AI to read the transcript and cross out repeated takes and false starts. Filler words show in violet, long pauses as green chips and retakes in red. Nothing is cut until you Apply, and a click on any mark keeps that word or pause.

Apply, and bring anything back

Apply removes everything marked in one undo step. Each removed part becomes a small red mark in the transcript and on the timeline; hover it to see what went and click Bring back. Restoring every removed part, in any order, gives you the whole recording back.

Click a sentence to select its footage and jump to it, double-click it to cut or keep it, and watch the word being heard turn yellow while the video plays. It works the other way too: click a clip of the transcribed track on the timeline and the transcript scrolls to its sentence.

The five templates

• Face only: you talking to the viewer while nothing on the screen matters yet. The intro, an opinion, the wrap-up.
• Screen + camera: the screen fills the frame with you in a corner, for clicking through, typing and showing a result.
• Side by side: a large screen panel with your face beside it, for long walkthroughs, code reviews and plans.
• Screen only: the screen alone.
• Screen + cutout: you, cut out of your background on the Mac with Apple Vision, standing on the screen.

Lay out with AI

Lay out with AI reads the transcript and what happened on the screen (clicks and typing) and picks a template for each part. Switches between layouts morph in 0.6 seconds.

The Layout lane, and Customize

Every switch is a block on the Layout lane: drag it, drag its edges, change its template or remove it. Drag an edge all the way to the next layout and the two meet, with nothing of the default left between them. Customize… sets where the camera sits (eight spots), its size, margin, rounding, shadow and framing, on a live preview of your own footage: click where you want the camera. For Screen + cutout you pick the backdrop colour and tune the key, or let Fix with AI do it. Save as default keeps the look for new projects.

Zoom with AI

Zoom with AI reads what you say for the things you name on the screen, finds each one in a screenshot taken at that moment, and zooms on it calmly: lighter and quicker during a walkthrough, closer with a slow push on one thing, back out before the screen changes. It also adds a slow push-in on your face where only the camera shows.

Each zoom, and push-ins

Every zoom is a block on the Zoom lane. Select one to set how close it gets (1.2× to 4×), where it aims (click the picture), whether it holds still or follows your cursor, and how closely. Aim with AI re-aims a single zoom. Without AI, Zoom on clicks zooms where you clicked.

A push-in is a slow zoom in or out that keeps a still shot alive, or a punch-in that hides a jump cut, on the whole picture, the screen or the camera.

Highlight with AI

Highlight with AI finds the small things you name, a button, a field, a number, and draws a red ring or a yellow highlighter on them in your brand. A highlight stays on its thing through zooms, and never plays while the screen is hidden or a zoom is moving.

The cursor

For screens recorded with Clicks and keys, Tofucut draws the pointer: its size, a dark or light arrow, a smooth glide that still lands exactly on every click, a glide across cuts, and fading out after five seconds without moving.

Name cards and topic titles

Lower thirds are the name cards and topic titles at the bottom of the picture, in the Brand tab under the layouts. Enter your name and role once. Create with AI reads the transcript and places a few where they belong, never in the last ten seconds. Add one at the playhead, pick one of six positions and edit its text: lower thirds are drawn live in your brand, so a change shows at once, with nothing to render.

Describe it, and it's made

Sweep the Animate lane or press + to make an empty block, then describe the animation in its chat: “a three-step diagram of the login flow”. An AI agent writes it and Tofucut renders it on your Mac at the block's length, in your brand, and places it on the timeline. Paste pictures as references (up to 10 per message), keep several chats per animation, and render again at a new length without asking the AI again. Over Video draws it on top of the picture; Full Frame replaces the screen while it plays. A revision renders only the moments that changed, and each animation keeps its last two renders (Undo goes back one); older ones go to the Trash to save disk space unless a snapshot you saved or a 9:16 / 1:1 version uses them. Turn that off in Settings › General to keep every version.

A cut through an animation, or a new length, doesn't change it by itself. Tofucut says so right away, and the Animate tab lists what no longer fits until you choose: Update to Match has the AI fit each one to the words now under it, and Just Fit the Length renders the same scene again at its new length with no AI. Until then it plays as it was made, cut short.

The first animation on a Mac sets up what renders it, once: Tofucut downloads Node.js (from nodejs.org), ffmpeg (from ffmpeg.martin-riedl.de) and its own browser engine (Chrome for Testing, from Google, about 100 MB) into its own folder and shows the progress where you asked for the animation. It uses its own engine even when Chrome is installed, and renders run with the network blocked. There is nothing to install by hand, and if a download fails it says so and you can try again. After the update that added this, the animation tools download once more (about 190 MB).

When an animation needs more time than the words under it, a list that can't be read in time or an end card after your last sentence, the agent makes room: everything after the block moves later, your voice pauses there, and the block grows to fit. The chat says how many seconds it added. Tell it not to add pauses and it won't.

Touch up with AI

The Camera tab grades your camera's light and colour. Touch up with AI shows the AI frames of your camera and close-ups of you. It grades like a colourist (true blacks, natural skin, a well-lit face, your background's colour kept), checks its result next to the original and refines it, up to six rounds. Or set it yourself: Exposure, Contrast, Highlights, Shadows, Whites, Blacks, Temperature, Tint, Saturation, Vibrance and Clean blacks. A before and after line on the picture shows the difference.

Clean voice

The wand on an audio track plays the voice on it cleaned on your Mac (your microphone, a voice-over take or a voice file you added): Apple's voice isolation, an EQ that measures your microphone and brings it to one clear tone (a dark mic gets its clarity back, a bright one loses its harshness), an even level, quieter gaps between words and broadcast loudness (−16 LUFS). It's made in the background, about a minute for an 88-minute recording, and the export uses it.

To say something again, click the record button on any audio track's header. The video plays from the playhead while your microphone records; click it again or press Space to stop. If you quit while a take is running, Tofucut stops it and places it on the timeline first. The take lands on that track where you started, over what was there (still restorable), and plays cleaned when the track's wand is on. Choose the microphone in Settings › General. To use only the good part of a take, mark it with I and O and drag the take onto your voice track: only that part moves (hold⌥ to copy it and keep the take whole). Said a sentence again to fix one word? Right-click the take and choose Sync with Main Track: it moves, on its own track, to where the same words are in your recording, to the millisecond, by comparing the sound. With In and Out set, only that part is compared.

Summary, and every track

The timeline opens as a Summary for watching the edit: the Layout, Zoom, Lower 3rds and Animate lanes over one video strip, one waveform and one sound row. Turn Summary off to see every track at full height and edit: screen and camera on video tracks, microphone and system audio on audio tracks, linked so they move together.

A real editor

Split at the playhead (K), ripple delete (G) or delete and leave the gap (B), trim edges, drag clips between tracks, copy and paste, link and unlink, and add tracks or any recorded file from Media. A file you drag from the Finder onto the timeline is copied into the project's media folder, so the project stays whole; if the disk has no room, a notice says so. Scrub the ruler to see and hear any moment. Snapping is always on. Reframe… on a video track trims the picture's edges for every clip, to hide a menu bar, the Dock or a notification without recording again.

Every cut is kept

The Cuts lane shows every removed part as a red tick, whether it came from the transcript, a silence or a trim. Click one to restore it. Nothing you remove is lost: restoring every part gives you the whole project back. If one of the cut's tracks is locked, bringing it back is refused with a note such as “Unlock A1 to bring this back.”

Save a version, go back to it

Save Snapshot keeps this exact version of the edit, transcript marks included. Go To returns to any snapshot and keeps your current edit first as “Before …”, so one ⌘Z swaps back. Tofucut also takes snapshots for you before big changes. When a change from the AI or a command can't be merged with yours, the other version is shown and yours is kept here as “Your edit before the merge”.

Back to Original

The eraser in the toolbar takes the project back to the moment you pressed Stop, after saving a snapshot of where you were.

Export

Export (⌘M) opens the export sheet: 4K, 1440p or 1080p presets with YouTube's recommended upload settings (H.264, AAC 320 kbps, 45 Mbps at 4K), or Custom with HEVC and your own bitrate. It shows the file's size and how long the export will take on your Mac, and warns when the disk is too full. Drag the handles to export only a part. The export runs in the background while you keep working. Its bar moves, and Cancel works, even while your cleaned voice is still being made. Quitting while an export runs asks first: “An export is still running. Quit anyway?” with Keep Working or Quit.

Ready to post

When it finishes, “Ready to post.” opens from the Export button with the file, its size and Show in Finder. The MP4 is saved in the project's folder.

Separate files exports every track as its own file with your cuts and nothing else, each starting at zero and as long as the timeline, so they line up in Premiere, Final Cut or DaVinci Resolve. With the background removed you also get your cutout on transparency.

Vertical 9:16 and square 1:1

Right-click a project › Create in New Format makes a 9:16 or 1:1 version next to the original, from the same recordings, with the same cut, lower thirds, zooms and touch-up, and three layouts made for the shape. Its AI pass frames the screen on what you're talking about, so a phone shows it whole and readable (about US$1 for an 8-minute video), and it can make the animations again for the shape. The original isn't changed.

The project list

The project list shows every recording with its picture, length, size and state. After an AI run, a project still named by its date gets a short title from the transcript; Rename… sets your own. Click a project to select it, and double-click it (or right-click › Open) to open it; the open one has a small bar at its left edge. ⌘-click or ⇧-click selects several, and a right-click acts on all of them. Select several recordings and Combine makes one video from several takes, edited and exported as one. Delete Cache… frees the files Tofucut can make again by moving them to the Trash, and Move to Trash keeps everything recoverable.

A project someone sent you can use only files inside its own folder and its recordings. A clip that pointed at any other file shows as missing until you add that file yourself.

While a project is open, the list steps aside so the editor gets the whole window. The Projects button at the top left, or ⌃⌘S, slides it in on top of the editor without squeezing it; click outside it or press Esc to put it away.

Your style, everywhere

A brand is a style: the colours, type and shapes of your layouts, lower thirds, highlights and new animations. Tofucut comes with two, Tofucut and Excalidraw. The Brands button at the top right of the window opens the Brands page from anywhere. You pick a project's brand on the recorder page, or later at the top of its Brand tab (right-click a project › Brand… opens it there).

Layouts and lower thirds are your brand's templates, never made for one video. Pick another brand and every layout, lower third and highlight changes to it at once, with your texts kept; ⌘Z undoes the switch. Animations are made for each video in the brand you had, so the Brand tab tells you how many were made in another brand; ask for them again in the Animate tab to match.

Make your own

New Style… starts one, and its brand chat builds it from your words, your logo or your website. Import… brings in a style made elsewhere, Export… shares yours as a zip, with its learnings log inside (the export says so), and Get the Claude Skill… lets Claude make one in exactly the format Tofucut reads. Tofucut checks a zip before it unpacks it. Replacing a brand keeps the old one in the Trash.

Four panes

Settings (⌘,) has five panes. Account: who is signed in, with Sign Out; your plan, with a button that opens tofucut.com in your browser, already signed in: the Tofucut plan's page on the free plan, your subscription (card, invoices, cancelling) on the Tofucut plan. The plan and the payment are handled there, never in the app; and on the Tofucut plan, a bar showing how much of the month's AI is used and the day it renews. General: the appearance, the transcript's font (serif or sans), the voice-over microphone, whether older animation renders go to the Trash (on by default) and What leaves your Mac (see below). AI: the AI everything uses (see Which AI it uses). Transcription: your ElevenLabs key, kept in your login keychain. Shortcuts: change any editor shortcut.

Help › Take the Tour runs the first-run tour again: a one-minute practice take, edited by the AI, with each control ringed as you go.

To use Tofucut you sign in once, with a free account. While Tofucut is in closed beta, signing in on tofucut.com puts your account on the waitlist, and the download and the app's sign-in open once it is let in. The first window says “Sign in to use Tofucut” with one button, Sign In with Google. It opens your browser on tofucut.com; you click Google's button there (one click when your browser already knows your Google account), check that the code on the page is the one the app shows, and choose Connect. The page then offers Back to Tofucut, and the app carries on by itself. After that it works without a connection. When it has one, the app checks your plan and the month's AI again whenever it comes to the front.

What each AI feature sends

“Your AI” is the one you connected on the free plan. The Tofucut plan is coming; when it opens, the same things will go through Tofucut's server to the AI and transcription it runs for you, and nothing more.

Before the first upload to each service, the app asks once. A sheet says what will be sent and to whom: for ElevenLabs, the voice track's audio; for your own Claude or ChatGPT, a custom server or the Tofucut plan, the transcript, still frames of the screen and camera, pasted pictures and what the AI reads in the project. Choose Allow or Not Now. Settings › General has a section, What leaves your Mac, with a row per destination (Allowed or Not asked yet) and a Reset button.

Everything else runs on your Mac: recording, silence detection, fillers and pauses, background removal, Clean voice, rendering animations, the preview and every export. Tofucut has no telemetry or analytics, and API keys stay in your login keychain. Signing in sends nothing from your Mac but the Mac's name, which your account page lists; Tofucut's server keeps your name and email address from Google, your plan, and for the Tofucut plan how much AI each request used, never what was in it. The first animation downloads the tools that render it: Node.js from nodejs.org, ffmpeg from ffmpeg.martin-riedl.de and a browser engine (Chrome for Testing) from Google, each checked before use. Connecting an AI downloads its program too: Claude Code from claude.ai, after asking you, and for ChatGPT, OpenAI's Codex from OpenAI's releases (releases.openai.com, or its GitHub releases as a fallback). Nothing of yours is sent to get any of them.

The ones you'll use most

Change any editor shortcut in Settings › Shortcuts.

⌘Q takes two presses to quit, so a slip doesn't close the app. With the main window closed or minimized, one press quits. When work is running, such as an export, Tofucut asks first: Keep Working or Quit.

The preposter command and Claude Code

The preposter command-line tool ships inside the app and does everything the editor does, by command, on any project by the name the app shows (“13 Oct 2026 at 15:51”). preposter agent brief prints the script as heard with every removed part, preposter agent auto runs Edit for Me without the window, and preposter export exports. The open app picks up each change within a second.

preposter agent skill --install installs the tofucut-edit skill, so Claude Code can edit a project end to end from one sentence: “edit project 13 Oct 2026 at 15:51 using the Tofucut branding”. The same skill is a download of its own on the Download page, beside the app, to hand to your agent.

What you need, and what it doesn't do

• macOS 26 or later on a Mac with Apple Silicon. There is no Windows or Intel version.
• On the Free plan, transcription needs an ElevenLabs key and the AI features need Claude, ChatGPT or a custom AI set up in Settings. The Tofucut plan, when it opens, needs neither.
• Finding retakes and filler words works best in English today.
• macOS asks you to confirm screen recording permission about once a month. It's a system setting for every screen recorder.
• Not included: music libraries, effects packs, transitions, captions, or uploading to YouTube.

How long does the AI edit take?

A few minutes for a typical video. Transcription and each AI step take about a minute or two, and you can keep working while they run.

Does it work with my camera and microphone?

It works with any camera and microphone your Mac sees, built in or plugged in. The camera check tells you if a camera won't look smooth, and the level meters show your microphone working before you record.

What does the AI cost to run?

On the Tofucut plan, once it opens, nothing extra: the AI and the transcription will be included. On the Free plan you pay those companies directly. Transcription is about US$0.22 per hour of audio on your ElevenLabs account, and the AI steps run on the Claude or ChatGPT plan you already pay for, or your own key. With a key, a 9:16 version's AI pass is about US$1 for an 8-minute video, and making an animation again costs about US$0.30 to US$0.50.

Do I need to know video editing?

No. Ask the AI panel for the edit, watch it, and bring back anything you wanted to keep. The timeline is there when you want to change something by hand.

Is this the same Tofucut as the YouTube research tool?

Same maker, different product. Tofucut started as a web tool for YouTube research and packaging. That tool is no longer developed, and Tofucut is now this Mac app for editing your videos.

Recent changes

• 9 Nov 2026 The app asks once before the first upload to each service, and Settings › General shows what you allowed. ⌘Z undoes only your own edits, and quitting asks first when work is running.
• 30 Oct 2026 New Project: Record or Blank Project, an empty project with a Screen, a Camera and a Voice track for your own files or just your voice. Over an open project the project list slides in on top.
• 27 Oct 2026 Voice-over: record onto any audio track while the video plays, and Clean voice on every audio track.
• 20 Oct 2026 Clean voice: your voice cleaned on the Mac, from a wand on the microphone's track.
• 19 Oct 2026 The recorder shows microphone and system audio levels before you record.
• 19 Oct 2026 A first-run tour that makes your first video with you.
• 19 Oct 2026 Vertical 9:16 and square 1:1 versions, framed on what you say.
• 19 Oct 2026 Highlights: a ring or a marker on the thing you name.
• 19 Oct 2026 Zoom with AI works from what you say and what's on the screen; the timeline opens as a Summary.
• 18 Oct 2026 The AI panel in every project, on the AI you already pay for.
• 17 Oct 2026 Snapshots, and one panel with every tab.