3.6.1

• Popup redesign: the master (FIR) and EQ modes are now two labelled toggles in
the header (green FIR, blue EQ) instead of a toggle plus a checkbox.
• Diagnostics reorganized to lead with the current mode, in a clearer order and
plainer wording.
• The picture-delay readout now reads as a simple sum (FIR + pipeline + trim),
and the "Run on" site line is tidier (e.g. "youtube.com").
• Lighter resource use: the YouTube quality check backs off on idle tabs, and
status updates pause while a video is paused.
• Clearer "loading your settings…" message on a slow first load, instead of a
momentary "no impulse responses loaded".

3.6.0

Reliability
• Fixed the extension sometimes showing "No video found" or "No impulse
responses loaded" on a fresh link, even though the video was playing
(heavy pages like YouTube). Media detection and settings/impulse-response
loading now run up front and retry until ready, so the filter reliably
engages on first load.

Subtitles
• Fixed a delayed caption occasionally sticking on screen at the start of a
video (a stale first line in a corner or off to one side). Abandoned caption
windows are now detected and cleared within a couple of seconds, and captions
reset correctly on seek, resume-from-middle, ad→video transitions and reloads.

Popup
• The header now has two toggles — green FIR (the full filter + A/V sync) and
blue EQ (low-latency tonal-only mode) — each labelled on the switch itself.
The separate EQ checkbox is gone.
• The status line now explains the current mode in one plain sentence.

Other
• Backup & restore (Settings): save every setting plus your imported impulse
response(s) to a single file, and restore it in one click after an update
or reinstall.

3.5.0

Reliability & sync fixes:

• Fixed the extension sometimes showing "No video found" or "No impulse
responses loaded" on a fresh link even though the video was playing.
Media detection and settings loading are now resilient on heavy pages
(e.g. YouTube), with automatic retry so the filter always engages.

• Fixed a stale subtitle line occasionally sticking on screen (top-left /
side) on YouTube. Delayed captions now clear correctly when the player
swaps caption windows or when you resume a video from the middle.

• Added Backup & restore (Settings page): save every setting plus your
imported impulse response(s) to one file, and restore it in one click
after an update or reinstall.

• Various startup and first-load stability improvements.

v3.1.2 — playback reliability

• Fixed the first-load freeze on YouTube when the video switches from its initial low quality to your set quality — the delayed picture now follows the switch instead of getting stuck.
• The delayed picture can no longer freeze indefinitely: if a stream can't be mirrored it cleanly falls back to the standard picture path.
• YouTube quality lock now forces your target quality earlier so less of the low-quality start plays.
• Continued stability fixes for seeking, buffering, long videos, and livestreams.

v3.1.1 — stability & bug fixes

• Native-smooth delayed picture: fixed first-load freezes, and improved handling of seeking, buffering, looping, long videos, and livestreams.
• Audio now waits for the picture on first load — no more "audio before the video."
• YouTube quality lock: sets your target quality earlier (starts in-range, no disruptive mid-load switch) and leaves Shorts on native quality.
• Memory: the video mirror is now bounded, so long / high-quality videos don't climb memory or freeze; extreme/unmirrorable streams cleanly fall back to the canvas path.

v3.1.0
• New: YouTube quality lock — force and pin a quality (matches your screen by default) so YouTube can't downgrade mid-video. Auto-skips Shorts.
• Native-smooth delayed picture is now far more reliable: fixes for reels/looping, seeking/timestamp jumps, long videos, buffering, and self-hosted MSE streams.
• Audio now waits for the picture on first load — no more "audio before the video."
• Fixed a memory leak that could slow down or crash the tab over long sessions.

2.4.0

New — EQ mode: a low-latency (~0 ms) per-channel EQ derived from your FIR's own
magnitude response (a bank of biquads, extracted once when the IR loads). It needs
no picture delay, so it runs on players the full convolution can't reach — cross-
origin embedded players — and, with the master switch off, on every capturable
player. On by default; toggle in the popup and in Settings.

New — Automatic per-player support detection: players that genuinely can't be
processed by any extension (DRM; cross-origin media without CORS) are detected and
left completely untouched, and flagged on the toolbar.

New — Clearer toolbar badge states, each explained in a legend in Settings:
ON – full FIR convolution running
EQ – low-latency EQ mode running
HDR – full convolution + HDR passthrough (delayed picture is a native 10-bit shadow)
NOT – this player can't be processed; left untouched
OFF – idle / not enabled here

New — Live EQ verification in the popup: per-side band counts and gain ranges read
directly from the running filters, so you can confirm the EQ is actually applied.

Improved — Redesigned Settings page with inline help on every control; assorted
A/V-sync and stability fixes.

v2.2.0

• New: subtitles are now delayed to land on the delayed picture even on players
  that render their own captions (e.g. YouTube), not only standard <track>
  subtitles. It mirrors the site's own caption, delayed, matching its position
  and style.
• New: a manual "Subtitles" offset in the popup to fine-tune caption timing.
• Fixed: captions could appear slightly early, briefly flicker on streaming
  sites, or jump to the corner for a moment during a silence.
• Fixed: a periodic stutter caused by subtitle re-timing.
• Performance: the caption and picture-delay loops now run at ~60 Hz instead of
  the monitor's full refresh rate, and avoid a per-frame style flush — lower CPU/GPU use.
• Cleaner popup: per-control help is now behind "?" tooltips.

v2.1.0

• HDR passthrough (new): HDR / 10-bit video is now delayed using a hidden
native <video> ("shadow") fed the same media, instead of an 8-bit canvas —
so HDR is preserved while the picture is still delayed to match the FIR.
Works on local/direct files and on MSE streaming (e.g. YouTube) by mirroring
the page's own video segments into a private MediaSource. DRM-protected
video (Netflix etc.) cannot be processed by any extension and is unchanged.

• Reduce FIR latency is now a factor: 1× / 2× / 4× / 8× / 16× (keep taps ÷ factor
around the impulse peak), replacing the old millisecond field.

• New "filter impact" panel in Settings: shows, original → after reduction,
per channel — taps, latency, the EQ's magnitude range, the deviation from the
original EQ response (max/RMS dB), and group-delay variation, so the exact
cost of a reduction is visible before you commit to it.

• Fix: fullscreen HDR no longer dims when the player's on-screen controls hide.

• Toolbar badge shows off / on / HDR per tab.

Version 2.0.0

Smoothness & performance
• Fixed the main cause of periodic video stutter — a recurring main-thread layout
flush in the overlay geometry check. On high-refresh displays the delayed picture
now runs with no lost or dropped frames.
• Hidden/background video tabs now idle their video pipeline entirely (no capture,
buffering, or drawing) while audio keeps playing — so running several video tabs
and switching between them no longer loads the machine.
• The frame buffer is strictly bounded and self-cleaning: each new frame evicts the
oldest and frees its GPU memory immediately, so memory stays constant over time
and a hidden tab drops to zero.

Simplification
• Removed three audio options that were shown not to affect quality or smoothness
(convolution rate, audio buffer size, silent-tail trim). The convolution now always
runs at the filter's native rate.

Version 1.3.0 — big filters, performance, and presets.

Large & low-latency filters
• Full support for very large linear-phase filters (multi-second group delay, e.g. subwoofer FIRs) — the picture-delay cap now scales to the filter instead of truncating it, so audio and video stay locked.
• Reduce FIR latency: optionally window a huge filter down to near real-time, with a one-click Auto.
• Trim silent tail (lossless): drops filter taps below a chosen dB floor — inaudible, but cuts convolution CPU several times over.

Performance / smoothness
• Convolution rate option (48 / 96 kHz) to lighten a heavy filter, native by default.
• Audio buffer option (large) so a heavy convolution stops stuttering the video; added latency is auto-compensated.
• Off-main-thread video presenter that now runs on more sites.

Presets
• Three full settings profiles (1/2/3), switchable from the popup and Settings — each remembers its own impulse responses and every setting.

Sync & controls
• Offset and Device A/V latency now adjust up to ±15000 ms for large filters.
• Instant resume after pause (the delayed picture no longer refills from scratch).
• Import one stereo WAV and it's split into L/R automatically.
• Various A/V-timing and device-latency accuracy fixes.

Version 1.2.0 — major sync, smoothness and stability update.

Sync accuracy
• Real-time output-clock drift compensation — measures the audio-clock vs wall-clock gap and corrects it silently and continuously, so sync no longer wanders over time.
• Per-video player A/V skew correction (mediaTime vs element clock) — fixes the "in sync on the test clip but off on YouTube/streaming" gap on heavy MSE/4K sources.
• Device A/V latency: a one-time calibration (via the built-in Sync calibrator) for your display + DSP/HDMI chain, which no browser API can see. Applies globally; the per-video offset stays free for fine-tuning.
• Audio context is pinned to the impulse response's native sample rate, so the convolution is bit-exact regardless of the device's rate.

Smoothness
• GPU zero-copy picture pipeline (createImageBitmap capture + transferFromImageBitmap present) — full source quality, never downscaled.
• Optional off-main-thread presenter (OffscreenCanvas worker) immune to page jank, with automatic fallback where a site's CSP forbids workers.
• Cadence phase-lock + miss-tolerant frame timing removes high-refresh judder.
• The old automatic quality downgrade is gone — video always plays at source resolution.

Output device binding (speakers vs. headphones)
• Optionally process only on your chosen output device and bypass cleanly on others (FIR speaker correction is wrong on headphones).
• Optional Windows helper (downloaded from Settings, installed manually) reports the current output device for instant, reliable switching.

Audio quality & stability
• Dual mono convolvers with synchronized re-prime — no L/R drift, no dropouts on large filters.
• Sample-accurate corrections applied via inaudible rate-capped glides (no clicks or pitch artifacts).
• Auto-resync safety net (default 15 min) flushes browser-internal audio-queue bloat that can otherwise push audio seconds late on very long sessions.

Convenience
• Import one stereo WAV and it is split into L + R automatically.
• Redesigned popup and Settings, live diagnostics, and a one-click Resync.