Do I have to turn on video recording in my tests first?

For the best output, yes. The preferred input is a test-results folder or a trace.zip whose sibling is a .webm recorded by Playwright's recordVideo, which gives native frame rate and the sharpest zooms. A standalone trace.zip still works - the source video is assembled from screencast JPEG frames inside the trace - but at variable cadence.

Does wiring this into my suite make the tests crawl?

Not if you drive it from the step helpers. The README describes narrate() with autoWait padding, pace() holds, and waitForNarration() to hold a beat until a line finishes, and notes that with TTS you can skip autoWait entirely and run the test at full speed while the renderer freezes frames to stay in sync.

What does the Jev director replace in my code?

Manual direction calls. .direct(JevDirector(), { goal }) watches the rendered recording using pixel changes and local OCR, then selects camera moves, emphasis and pacing from a bounded candidate set, so you can drop explicit zoom() and highlight() calls. The manual helpers stay available.

Can I point it at a recording that Playwright did not make?

No. The inputs are a trace.zip or a test-results directory, so the frames come from Playwright either way. What you can skip is the pipeline rather than the source: every stage is optional, so the trace parser or the subtitle generator can be used alone.

What kinds of artifacts can the pipeline produce?

TTS voiceover through OpenAI, ElevenLabs or Amazon Polly; SRT, WebVTT and ASS subtitles with optional burn-in; animated click highlights and cursor overlay; auto-zoom with easing; URL bar pills; background music with ducking; intro and outro clips; frame interpolation; and a toggleable embedded subtitle track.

What do I need installed, and how do I run it?

npm install playwright-recast, with ffmpeg and ffprobe on PATH; on macOS brew install ffmpeg-full gives drawtext for URL bars and libass for burned subtitles. Run npx playwright-recast -i ./test-results/trace.zip -o demo.mp4, adding flags such as --speed-idle, --srt or --provider.

Which trace layouts does it accept?

Either a test-results directory, which is preferred and uses a sibling .webm recorded by recordVideo for native frame rate, or a standalone trace.zip, in which case the pipeline assembles the source video from screencast JPEG frames stored inside the trace at a variable cadence.

CommunityJev Demos & Experiments

Playwright Recast

Turns a Playwright trace into a narrated demo video through a chainable pipeline - speed control, subtitles, TTS, click and cursor overlays, zoom - with ffmpeg doing the render.

See highlights

Open github.com

Stars: 64
Language: TypeScript
License: MIT
Published by: Community
Author: ThePatriczek
Last verified: Oct 2026

ThePatriczek/playwright-recast

User repository on GitHub

View on GitHub

Fluent pipeline library for processing Playwright traces into polished demo videos with TTS voiceover, subtitles, speed control, and zoom.

Forks: 6
Last pushed: Updated Oct 2026

Repo stats from the GitHub API, cached Oct 2026.

About this project

Transform Playwright traces into stunning demo videos — automatically. Website · Documentation · Release notes · Changelog. Snowflake Cortex Neo4j Agent Integration — see a real agent workflow spanning Snowflake Cortex and Neo4j presented as a polished product video. Recording product demos is painful. Every UI change means re-recording. Manual voiceover and subtitling takes hours. Timing is always off.

Quoted from the project's README. Read the full README.

Tagscommunity jevbest

Works forJev for Classification Jev for Routing

Used inJev for Software & DevTools

Languages: TypeScript

Highlights

The Jev director watches rendered frames with pixel diffs and local OCR, picking camera moves and pacing from bounded candidates.
Speed control fast-forwards idle time, network waits and navigation while leaving user actions at normal speed.
Subtitles from an external SRT or BDD step titles, in SRT, WebVTT or ASS, burned in or embedded as a track.
A suite reporter turns one Playwright run into one video, one clip per test, and exits with Playwright's own status.

Watch out

Quality depends on what Playwright actually recorded: the preferred input is a trace.zip with a sibling .webm from recordVideo, while a standalone trace.zip is assembled from screencast JPEG frames at roughly the capture rate, so zooms can look softer on that second path. ffmpeg and ffprobe must be on your PATH, and frame interpolation is an ffmpeg pass. The Jev director is the least specified part - the README says it inspects the rendered recording and chooses from bounded candidates, but the capture does not show which model runs it, how it is configured, or what it costs. The captured README stops inside the playwright-bdd examples, so MCP server setup and the studio tooling were not visible. Copy reviewed against the repository README on 2026-10-04, not from running it.

Questions about this entry

Short answers drawn from the linked project or page.

More Jev resources like this

jevry

123

Electron desktop browser where Jev picks the next executable action from a page's observed controls while a separate text model plans; deterministic code builds the action space, validates responses, dispatches input and keeps receipts.

Jev Demos & Experiments#community#jevbest

Community

michaelswissa/jevryTypeScript

Pi-Jev-Guide

Pi guard plugin that checks inputs and tool calls with local rules plus Jev semantic judgements, with redaction and verification reminders.