中文 · Local screen/video recording for Chrome and Microsoft Edge.
Load this directory as an unpacked extension from your browser's extension developer page, or extract store/wescreen-1.28.0.zip and load it. Click WeScreen to open the recorder. Update/reload the extension after changing files; close old recorder pages before the database upgrade.
- Telegram channel profiles retain links, independent numbering, a message recording plan and download subfolders. Completed recordings advance to the next planned item; the library filters by channel and links back to the source message. Screen sharing still requires the browser picker.
- Record a screen, window or tab. MP4 requires native H.264 + AAC encoder support; WebM is available otherwise. The source picker controls shared audio.
- Use Library for playback, search, sorting, course records, imports and last export location. Moving to Trash is reversible; permanent deletion does not remove exported disk files.
- Recordings are stored in extension IndexedDB, with unlimitedStorage and a persistent-storage request. There is no fixed 10 GB library cap. Disk space is finite. Long recordings automatically rotate near 800 MB to bound memory; no single in-memory segment should reach the 1.6 GB ceiling. A small boundary overlap is possible.
- Optional camera composition uses a worker frame pipeline rather than visible-page animation. Unsupported browsers fail clearly; direct screen capture remains available without the camera.
- Telegram preset: pause the video, choose the Telegram tab with shared audio, then play after the countdown. Audio availability for a desktop window depends on the operating system. Use crop/trim to keep only the player and remove waiting time.
Download/extract the separate helper package or use the repository. Install Python 3.10–3.13 and FFmpeg, then run bash enhancement/start.sh. The extension connects automatically; manual token entry remains under Advanced connection settings. On macOS you can open enhancement/start.command.
Natural restoration is the default. FSRCNN offers lightweight 2× enhancement; SeedVR2 is optional and requires bash enhancement/install-seedvr.sh (about 7 GB of weights). Preview 3, 5 or 10 seconds from the current playback position, compare synchronized playback or native-size detail, then process a full video. AI may invent details; output is SDR, not HDR.
Crop/trim saves a new MP4. Full results remain in enhancement/.runtime/jobs until explicitly deleted. Processing tasks can be reopened, saved to Library again after a failure, or streamed directly to a file. Completed results survive helper restarts; interrupted tasks must be rerun. One task runs at a time. Completed results have no fixed count cap; available disk space limits retention. No videos go to an external processing service. See helper guide and privacy policy.
node --test tests/*.test.js
enhancement/.venv/bin/python -m unittest discover -s tests -p '*_test.py'
WESCREEN_SEEDVR_TEST=1 enhancement/.venv/bin/python -m unittest discover -s tests -p '*_test.py'
python3 scripts/build-package.pyBrowser tests require Playwright and Chromium; if absent those tests are skipped. Tests include cross-page recovery, slow storage, worker composition, native media playback, trash/restore, quota failures and result retention. Synthetic fixtures verify the pipeline, not restoration quality on real Telegram content. Validate long sessions, audio sync and OS/browser support separately before release.
New controls stay inside the existing advanced menus: recording → advanced settings → Professional capture & audio; player → More → Professional playback tools; library → File management. See feature coverage and verification.
Select a recording folder to store each completed segment once on disk (256 MB rotation threshold), with a file index in the library. Recovery chunks remain in browser storage until the disk file closes and its size is verified; disk failures retain a full library copy. Keep the selected files in place and folder permission granted. An optional dual-copy mode keeps both disk and library copies. This does not trigger browser downloads. Monitoring reports preview frame delivery/drops, not encoder frame statistics. Independent audio is optional and follows recording pauses; download or delete it in File management.
Select one continuous recording and use Merge for compatible-stream copying, or Compatible merge for explicit re-encoding. Results stay in the helper until manually saved or exported. Originals remain available. Large merged files can be exported directly without first importing them into the library; library import still has a 1.6 GB per-file limit.
Continuous recordings also have a Merge & download (compressed) button beside Play. It processes all parts in sequence into one H.264/AAC MP4, preserving the aspect ratio within 1080p bounds without enlarging smaller sources (up to 30 fps, 2.2 Mbps video and 96 kbps audio). On macOS it prefers hardware encoding, with software fallback. When processing completes, a save dialog opens; the result streams from the local helper instead of loading the whole video into browser memory. Originals are retained. Keep a WeScreen page open; if closed, automatic export resumes when a page is reopened and the helper reconnects. A cancelled or failed download can be retried using Export result in Video processing. Update both the extension and helper to enable this feature.
On macOS, run the helper's install-launcher.command once after dependencies are installed. The extension's Start local helper link then launches or updates the managed helper without interrupting an active processing job. The installer retains existing models and recordings. The helper package is separate from the Edge Store extension package.
Version 1.17 adds a persistent recent-play playlist, three clear quality choices, release-once seeking with bounded thumbnails, signed audio sync, hold-to-compare, subtitle appearance and recovery feedback. Resolution caps (4K / 1440p / 1080p / 720p) are separate from encoding bitrate; the actual captured pixels are reported. Source mode can retain 8K only when the source, device and browser support it.
Automatic static-tail cleanup is enabled by default and requires a terminal freeze plus silence longer than one minute. A new trimmed version is saved; originals stay intact. Still lectures with sound, a wholly static video, and internal holds followed by motion are preserved. Local fixed-watermark filling requires preview of the current region before saving. See the helper guide for limitations.
Update the macOS launcher with install-launcher.command. A new extension ID is approved in a native local confirmation window; previously approved IDs remain trusted. Background reconnect never opens the confirmation window. No token needs to be copied for this flow.
Channel workbench supports reorder, skip, rerecord and duplicate links. Recording diagnostics show preview fps/dropped frames, last completed writes and missing audio signals. A persisted sequential enhancement queue supports pause after the current job, resume and retries. Editing supports clockwise/counterclockwise 90° and arbitrary rotation, plus a 1920×1080 landscape canvas, preserving originals and audio. Border suggestions use the current frame only and require visual checking. Channel crop preferences and synchronized detail zoom are available. Token-file import simplifies pairing. Existing Homebrew installations can install missing FFmpeg/Python when starting; other systems still need prerequisites.
Check the shared picture and audio before starting. Open the separate control window to pause, resume or stop. Save the current video and prepare the next to retain the shared source while paused. Copy the next message link and open it in the original shared tab before resuming. Review start/middle/end thumbnails, inspect suggested black-screen trims, and remember crop/rotation/landscape preferences for matching dimensions. The control window is not always-on-top and may appear in whole-screen capture.
Open WeScreen from the playing video tab, select Original video tab (compatibility capture), then grant the optional tabCapture permission on Start. This bypasses the sharing picker while preserving MP4 H.264/AAC output and local audio playback. Native API tests verify changing decoded frames from a real playing video tab. Device-specific Telegram freezes still require user verification.
Normal updates retain the wescreen IndexedDB database under the same extension ID. The schema migration only adds missing stores; it does not delete or rebuild existing stores. Release versions and database schema versions are separate.
- Edge Store: update the existing listing. Do not uninstall before updating or switch to a different extension listing.
- Unpacked: close recording/processing pages after saving the active recording, extract new extension files over the existing extension directory, then Reload that existing entry in
edge://extensions. Keep the same browser profile, extension entry and directory. Do not Remove it and load another folder. - Switching between store and unpacked builds can change the extension ID and does not transfer the library automatically. Download recordings before switching.
- Clearing browser/extension data or uninstalling can remove browser-local recordings. Separately exported MP4 files are not deleted by updating the extension.
- Helper updates should also replace files in the existing helper directory, preserving
.runtimeandmodels.
Upgrade verification: NODE_PATH=... node scripts/test-upgrade.cjs uses an isolated Chromium profile, verifies legacy schema 3→4 migration, updates 1.13.0 to the current package in the same directory, and compares video bytes/SHA-256, metadata, recovery chunks, tasks, playback state and preferences. It does not access the user's browser data.
Resolution limits follow source orientation: 1080p means up to 1920×1080 landscape or 1080×1920 portrait, without upscaling. Frame rate can follow the capture source or use 24/25/30/50/60 fps. Video bitrate follows pixel count, frame rate and desktop/motion content. The recorder separates target bitrate, the encoder-reported bitrate and measured file bitrate including audio; these are different measurements.
Professional audio offers 192 kbps speech or 256 kbps music for both the main recording and optional independent tracks. Disable microphone voice processing for music. System audio is not noise reduced. Peak protection is optional. Playback loudness balancing is a bounded, smoothed RMS adjustment; offline audio repair retains two-pass loudness normalization.
Visual processing copies compatible audio packets and preserves SDR color tags. Explicit HDR-to-SDR conversion requires FFmpeg's zscale/tonemap filters; on Mac use brew install ffmpeg-full and restart the helper. The original stays in the library. Realtime AI uses lossless PNG with the new helper and retains the chosen mode across parts of the same recording; slower machines fall back to original frames.
Upgrade the existing extension in its original directory. Do not uninstall it first or change its extension ID: recordings and playback history stay in the existing database.
Experimental face restoration is available under Player settings → Tools → Face restoration. Install the optional local model with enhancement/install-face.command, preview and compare, then confirm full processing. It is generative and may alter facial features; heavy masking cannot be reliably reversed. Original media and audio are retained. See helper setup and CodeFormer licensing.
The five workspaces share a compact charcoal and champagne interface. Recording settings sit beside the source view; the library defaults to rows, with related versions expandable beneath the original and an optional grid view. Restoration uses a parameter sidebar and comparison workspace. Chinese uses bundled Source Han Sans (SIL OFL); English prefers system SF Pro on macOS and falls back to the platform font elsewhere. Update the existing extension in place to retain recordings.
Use Open local video, or drop a video onto the library/player, for direct File-backed playback without importing. Supported codecs depend on the browser; use Import video to keep a persistent library copy.
The dedicated Player tab opens local files without importing and plays recordings selected in the library. Recording completion keeps the review and offers Play now. Telegram setup lives in its dedicated tab.
Privacy password is off by default. Set a custom password under About → Privacy password to protect library/player access. Unlock lasts for the current page; reload or Lock now requires the password again. Passwords are stored only as a salted PBKDF2 verifier. This is an interface lock, not encryption of video files. Changing or disabling protection requires the current password; forgotten passwords cannot be recovered from this UI. Existing recordings are unchanged.
Realtime RGB supports 1920×1080 or 1080×1920 without input downsampling. When the displayed image is larger, the helper can return genuine 2× output with width/height headers; older helpers retain the previous protocol. Text edges use feathered protection and conservative sharpening. OpenCL is attempted when available with CPU fallback; performance depends on the driver and hardware, and slow frames leave the original video visible. Native launcher approval is persisted before restarting the helper, so approved extensions can reconnect without a timely challenge exchange.
智能最佳低清修复：本机分段缓存、运动对齐降噪、可选 Real-ESRGAN GPU 修复。启动本机程序后，可运行 enhancement/install-smart.command 安装并验证 GPU 模型。播放器新增退出播放与 5 分钟无操作自动退出。
New and migrated recording settings default to 1080p, 30 fps and space-saving bitrate (about 1.4 GB/hour for 16:9 text content, encoder dependent). Migration runs once; later explicit quality choices remain unchanged. The setup screen estimates size before capture and updates it using actual capture dimensions.