The picker belongs to the browser, and so does the permission
Press Record a Tab in the drop zone, or the compact Record tab chip in the clip strip once you already have a clip, and what opens is not our interface. It is the browser's own sharing dialog, the same one a video call puts in front of you, offering a tab, a window, or a whole screen. Chrome, Edge, Firefox and Safari each draw it a little differently and offer a slightly different set of surfaces. You choose one, the browser hands this page a stream, and an indicator sits in the tab strip until you stop.
That arrangement is the whole point. No extension asking for permission to read every site you visit. No menu-bar recorder. No installer that needs admin rights on a locked-down work laptop. The button hides itself where the underlying API is missing, which in practice means phones and most tablets, so this is a desktop feature and labelled as one.
Ten seconds of footage costs ten seconds
This is real-time capture. There is no fast path through it: the recording runs at wall-clock speed and ends when you press stop, or automatically at five minutes. A three-minute walkthrough takes three minutes to record and will take you two attempts, which is worth knowing before you start rather than at minute two of take four.
So spend the discipline up front. Know the two clicks you are demonstrating. Put the page in its starting state before you press record, not after. Quit the chat app, silence calendar alerts, and close the tab with your personal email in the title, because notifications land in the capture and a banner sliding across frame is not something the trim can remove. Record roughly twice what you need and no more.
What arrives is a clip, not a finished file
The recording drops into the sequence as an ordinary clip, which means every control applies to it exactly as it would to a file you dragged in. Frame-accurate trim with single-frame nudges. Crop locked to 1:1, 9:16, 4:5, 4:3 or 16:9. Per-clip speed between 0.25x and 4x, which is how you get past the eight seconds you spent reading a dialog. Reverse and bounce. Split, at the playhead, when one take contains two demos. Up to three captions with their own timing, and up to ten clips stitched into one output.
Then Play edit runs the assembled thing through the same pipeline that will encode it, so the cuts, the captions and the loop are what you see before you commit to anything. The Output Format switch takes that same edit out as a GIF, an MP4, or an APNG without re-recording.
It records the screen, which is not the same as recording the source
Worth being blunt about, because it decides when this is the right tool. A tab capture gives you the pixels the compositor drew, at the rate it managed to draw them. If the machine was busy, the stutter is now in your recording. If the window was scaled down to fit beside your notes, the text is soft and no export setting will sharpen it. The cursor is in frame whether you wanted it or not, and so is the scrollbar.
So when a real file exists, use the file. An export from a design tool, a downloaded Loom, a QuickTime screen recording, anything your operating system's recorder produced: all of them are cleaner sources, and all of them drag straight in. The screen recording page covers that route. Record a Tab earns its place where no file exists or getting one is not worth the trouble: a web animation with no export button, a bug that only reproduces in the browser, a dashboard that will look different tomorrow.
One practical trick that costs nothing: resize the browser window to roughly the dimensions you want to export at before you record. Capturing at 800 px and scaling down is sharp. Capturing at 400 px and scaling up is mush, every time.
The recording never leaves the tab that made it
A recorder that uploads is an entire category of product, and this is not one of them. The stream from the share picker goes into this page's memory, the frames get encoded by your own processor, and the finished file lands in Downloads. There is no server anywhere in that path, which is why the whole flow still works with the network off once the editor has loaded.
Which matters most for the exact things people record: a staging build, a customer's account open on screen, an internal admin tool, a bug in a system nobody outside the company is meant to see. Converting without uploading is the same principle applied to files you already have.