Playwright and FFmpeg, or a take: three ways to script a demo video
Playwright records what a script did. FFmpeg can cut and zoom it if you tell it where and when. A take keeps the clicks beside the video, so the zooms plan themselves and stay editable.
By Hongbin Li · Updated November 2026
Ask an assistant how to have a coding agent record a product demo and the usual answer is: drive the page with Playwright, record the video, edit it with FFmpeg. That works, and for some jobs it is the right answer. This page says what each piece gives you, what is left for you to do, and how the same Playwright run can become an edited video without the hand work. (Disclosure: vosso is our product.)
What each gives you
Sources, read 18 November 2026: Playwright (videos, recordVideo options, trace viewer); FFmpeg (about, filters).
What Playwright records
Turn the video option on and Playwright records the page’s viewport, at 25 frames a second unless you raise it, scaled to fit 800 by 800 unless you set a size. The file is written when the browser context closes. It is a faithful record of what the script did, which is what a test wants.
Recent versions add an opt-in that highlights each action with an outline and a title, and draws a pointer that moves from one action to the next. That helps a reader follow a test. It is not an edit: the frame stays at full width, every wait is still there, and nothing tells the viewer where to look.
A trace is a different thing. It is a record you open in Playwright’s Trace Viewer: each action, DOM snapshots, the network, and a film strip of screenshots. It is not a video. But it knows exactly where and when every click landed, and that turns out to matter below.
What FFmpeg adds, and what it costs
FFmpeg can crop, scale, zoom and pan, and join clips, each with a filter. What it cannot know is your product. For every zoom you give it a place and a time; for every cut, two timestamps. A thirty-second demo with eight zooms and four cuts is a long command, and when the interface moves by forty pixels the numbers are wrong. An agent can write that command. It writes it again after every change.
Keep your Playwright run, and get the edit
If you already have the run, you do not have to choose. Record the video and the trace in the same run, then:
vos ingest video.webm --cursor trace.zip --out take
The video becomes the take’s footage, and the clicks in the trace, with their points and times, become its cursor track. From there the planner has clicks to zoom on, and the take opens in the same editor as any other. Without a trace there is no cursor track and nothing is planned; you place zooms by eye.
Or let the recorder drive the page itself. The script is a list of steps, close to what a test already says, and the take keeps each step’s time and place:
vos record --actions actions.json --url http://localhost:3000 --out take
See record with the CLI and the script’s reference.
When Playwright and FFmpeg are the right choice
- The video is evidence, not marketing. A failing test for a bug report, a run attached to a pull request. Nobody needs a zoom.
- You need Firefox or WebKit. Playwright records all three engines; vosso records Chromium.
- You want nothing new in the repo. If your tests already use Playwright, its video costs one option.
- The cut is one of a kind. For a single film you will never make again, hand-tuned FFmpeg is as good as anything.
A take earns its place when the video is shown to people and has to be made again: every release, every changed screen. More on that in a video for every release, and the agent tools that do this are compared on demo videos made by AI agents.
Frequently asked questions
Can Playwright record a video of my app?
Yes. Playwright records the page’s viewport when you turn the video option on; it is off by default, and the file is written when the browser context closes. Its docs give a default of 25 frames a second and a size scaled to fit 800 by 800 unless you set one.
Does a Playwright video show the mouse cursor?
It can. Recent versions have an opt-in that highlights each action with an outline and a title and draws a pointer moving between actions. It is off unless you ask for it, so a plain recording shows the page changing with no pointer.
How do I turn a Playwright recording into a product demo?
Two ways. By hand: use FFmpeg’s crop, scale, zoompan and concat filters, working out each zoom’s place and time yourself. Or hand the video and its trace to vos ingest, which reads the clicks out of the trace and plans a zoom for each, as an edit you can change.
Is a Playwright trace a video?
No. A trace is a debugging record you open in the Trace Viewer: the actions, DOM snapshots, network requests and a film strip of screenshots. It is the best record of what a test did and where it clicked, which is exactly what a zoom needs to know.
When is Playwright with FFmpeg the right choice?
When the video is evidence, not marketing: a failed test for a bug report, a run attached to a pull request. Also when you need Firefox or WebKit, or want no dependency beyond what your tests already use. It asks more work of you for every second that should look edited.
Something else? hello@vos.so