feat: Advanced Video Editor Implementation, Native WGC Integration, and AI Auto-Captions#122
Merged
webadderall merged 12 commits intowebadderall:mainfrom Mar 28, 2026
Merged
Conversation
… introduce video editor UI components.
- Hardened IPC with safety guards for all webContents.send calls - Scrubbed sensitive absolute paths from caption logs - Realigned Windows artifact naming logic
…veforms, and export capabilities.
…o playback, timeline editing, and event handling.
Contributor
|
|
|
Warning
|
| Cohort / File(s) | Summary |
|---|---|
Whisper IPC & Types electron/electron-env.d.ts, electron/preload.ts, electron/ipc/handlers.ts |
Replaced single-model Whisper APIs with model-parameterized endpoints; added model to progress payloads; added auto-caption progress/chunk events and new generate options (startTimeMs, durationMs); generalized handlers and safe-send IPC wrapper. |
Auto-caption chunking & flow electron/ipc/handlers.ts |
Segments video into 5-minute chunks with overlap, extracts per-chunk audio, runs Whisper with progress parsing and fallbacks, offsets/merges cues, emits auto-caption-chunk and auto-caption-progress, and cleans temp files. |
Native capture & monitor lookup electron/native/wgc-capture/src/monitor_utils.*, electron/native/wgc-capture/src/main.cpp, electron/native/bin/win32-x64/whisper-runtime.json |
Added findMonitorByBounds and bounds fields in capture config; findMonitorByDisplayId now returns nullptr on no exact match; added Windows whisper runtime metadata. |
Build scripts & gitignore scripts/*build-*.mjs, .gitignore, electron-builder.json5, package.json |
Prefer project-local .cmake_ext CMake, adjust extraction paths, normalize .gitignore and add caches/docs ignores, add vite.config.* ignores; electron-builder artifact templating and copyright; add npm typecheck. |
Audio editing UI & engine src/components/video-editor/AudioSettingsPanel.tsx, src/components/video-editor/VideoEditor.tsx, src/components/video-editor/SettingsPanel.tsx, src/components/video-editor/types.ts, src/components/video-editor/editorPreferences.ts, src/components/video-editor/projectPersistence.ts |
New AudioSettingsPanel, master/track mute/solo/volume/fade state and persistence; UI wiring for audio regions, master audio controls, time-selection generation range, Whisper model selection. |
Timeline, items, selection & playback src/components/video-editor/timeline/TimelineEditor.tsx, src/components/video-editor/timeline/TimelineWrapper.tsx, src/components/video-editor/timeline/Item.tsx, src/components/video-editor/timeline/KeyframeMarkers.tsx, src/components/video-editor/timeline/Row.tsx, src/components/video-editor/timeline/ItemGlass.module.css, src/components/video-editor/VideoPlayback.tsx, src/components/video-editor/videoPlayback/videoEventHandlers.ts |
Introduced timeSelection and timelineMode (move/select), background drag-to-select, caption rows/items, caption span routing with rowId, item variants (caption/caption-range), audio waveform/fade overlays, cyan glass styles, and playback seek API plus selection-boundary pause. |
Caption editing & merge src/components/video-editor/SettingsPanel.tsx, src/components/video-editor/VideoEditor.tsx |
Caption cue editor UI, per-chunk merging and ID assignment, seek/edit/delete for cues, generate button progress UI, and model-driven caption generation controls. |
Export & mixing (renderer/exporter) src/lib/exporter/audioEncoder.ts, src/lib/exporter/videoExporter.ts, src/lib/exporter/types.ts |
Extended export types/config to accept master/audio track controls; AudioProcessor and renderMixedTimelineAudio now handle master solo/mute/volume, per-region fades, and solo logic during mixing/export. |
Blur annotation support src/components/video-editor/AnnotationOverlay.tsx, src/components/video-editor/AnnotationSettingsPanel.tsx, src/lib/exporter/annotationRenderer.ts, src/components/video-editor/types.ts |
Added blur annotation type and blurIntensity, UI slider and handler, canvas export blur renderer that captures, blurs, and redraws clipped regions. |
Waveform utility & styles src/utils/audioWaveform.ts, src/index.css, src/components/video-editor/timeline/ItemGlass.module.css |
New generateWaveform utility with in-memory cache; added subtle-scrollbar utility and minor timeline CSS variants. |
Sequence Diagrams
sequenceDiagram
autonumber
participant UI as Client (UI)
participant Pre as Preload
participant IPC as Main IPC
participant FS as FileSystem
participant Whisper as Whisper Runner
participant Renderer as Renderer (webContents)
UI->>Pre: request generateAutoCaptions(opts)
Pre->>IPC: invoke generate-auto-captions(opts)
IPC->>Whisper: segment video, for each chunk extract audio
Whisper->>FS: write/read temp audio chunk
Whisper->>Whisper: runWhisperWithProgress() (parse stderr %)
Whisper->>Renderer: safeSend('auto-caption-chunk', {cues, model})
Whisper->>Renderer: safeSend('auto-caption-progress', {progress})
Renderer-->>UI: onAutoCaptionChunk / onAutoCaptionProgress events
alt all chunks done
IPC->>Renderer: final result {success, cues}
end
sequenceDiagram
autonumber
participant UI as Timeline UI
participant Player as Video Element
participant Engine as WebAudio (AudioContext)
participant Track as Audio Region Node
participant Master as Master Gain
UI->>Engine: createAudioContext + nodes
Engine->>Player: createMediaElementSource(video)
loop per audio region
UI->>Track: createGain + set volume/mute/solo/fade
Track->>Master: connect
end
Engine->>Master: set master gain (masterVolume / muted / solo)
Master->>Engine: connect destination
loop playback ticks
UI->>Engine: compute fadeMultiplier & effective gains
Engine->>Track: setTargetAtTime(gain)
Engine->>Master: setTargetAtTime(masterGain)
end
Estimated code review effort
🎯 4 (Complex) | ⏱️ ~75 minutes
Possibly related PRs
- Add custom cursor packs and harden native postinstall build flow #111: Overlapping build-script and .gitignore changes; both add project-local
.cmake_extpreference and.gitignorenormalizations.
Suggested labels
Checked
🐰 I hop through code with ears askew,
Chunks of captions, models new,
Fades and blurs in gentle flight,
Timeline hums into the night,
A carrot-toast to features true!
🚥 Pre-merge checks | ✅ 2 | ❌ 1
❌ Failed checks (1 warning)
| Check name | Status | Explanation | Resolution |
|---|---|---|---|
| Docstring Coverage | Docstring coverage is 1.54% which is insufficient. The required threshold is 80.00%. | Write docstrings for the functions missing them to satisfy the coverage threshold. |
✅ Passed checks (2 passed)
| Check name | Status | Explanation |
|---|---|---|
| Title check | ✅ Passed | The title accurately summarizes the PR's main contributions: advanced video editor implementation, native WGC integration, and AI auto-captions. |
| Description check | ✅ Passed | The PR description covers all required template sections with comprehensive details about the changes, motivation, type of change, and related information. |
✏️ Tip: You can configure your own custom pre-merge checks in the settings.
✨ Finishing Touches
🧪 Generate unit tests (beta)
- Create PR with unit tests
Comment @coderabbitai help to get the list of available commands and usage tips.
Contributor
|
|
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 20
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (3)
src/components/video-editor/timeline/ItemGlass.module.css (1)
168-180:⚠️ Potential issue | 🟡 MinorAdd the missing hover selector for solid caption items.
glassCyan.selectedis wired up, butglassCyan:hoveris still missing. Non-selected caption blocks therefore won't reveal their resize caps on hover like the other resizable variants.♻️ Proposed fix
.glassPurple:hover .zoomEndCap, .glassPurple.selected .zoomEndCap, +.glassCyan:hover .zoomEndCap, .glassCyan.selected .zoomEndCap, .glassCyanDashed:hover .zoomEndCap, .glassCyanDashed.selected .zoomEndCap {🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@src/components/video-editor/timeline/ItemGlass.module.css` around lines 168 - 180, The hover selector for the solid cyan caption variant is missing; add a rule matching .glassCyan:hover .zoomEndCap (in the same selector group where .glassCyan.selected .zoomEndCap and other .glassX:hover .zoomEndCap rules live) so that non-selected cyan caption blocks reveal their resize caps on hover just like the other solid color variants.src/components/video-editor/projectPersistence.ts (2)
406-415:⚠️ Potential issue | 🟠 MajorPreserve >100% region gain when loading projects.
src/components/video-editor/AudioSettingsPanel.tsx, Lines 166-171 exposes 0–200% gain, but this normalizer snaps any persistedregion.volumeabove1back to1. Reloading a project will quietly change the mix.💡 Suggested fix
- volume: isFiniteNumber(region.volume) ? clamp(region.volume, 0, 1) : 1, + volume: isFiniteNumber(region.volume) ? clamp(region.volume, 0, 2) : 1,🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@src/components/video-editor/projectPersistence.ts` around lines 406 - 415, The save/restore logic in projectPersistence.ts currently clamps persisted region.volume to a max of 1, which strips gains >100% (the UI supports 0–200%). Update the volume normalization in the region mapping (the line using isFiniteNumber(region.volume) ? clamp(region.volume, 0, 1) : 1) to allow up to 2 (e.g. clamp(region.volume, 0, 2)) so persisted volumes up to 200% are preserved while keeping the same default fallback of 1.
329-393:⚠️ Potential issue | 🔴 CriticalRound-trip
blurannotations here too.
normalizeProjectEditorstill collapses every non-image/figureannotation to"text"and never preservesblurIntensity. Saving and reopening a project will silently strip blur redactions even thoughsrc/components/video-editor/types.ts, Lines 130-201 andsrc/components/video-editor/AnnotationSettingsPanel.tsx, Lines 148-170 now expose that state.💡 Suggested fix
- type: region.type === "image" || region.type === "figure" ? region.type : "text", + type: + region.type === "image" || + region.type === "figure" || + region.type === "blur" + ? region.type + : "text", content: typeof region.content === "string" ? region.content : "", textContent: typeof region.textContent === "string" ? region.textContent : undefined, imageContent: typeof region.imageContent === "string" ? region.imageContent : undefined, @@ figureData: region.figureData ? { ...DEFAULT_FIGURE_DATA, ...region.figureData, } : undefined, + blurIntensity: + region.type === "blur" + ? clamp( + isFiniteNumber(region.blurIntensity) + ? region.blurIntensity + : DEFAULT_BLUR_INTENSITY, + 2, + 50, + ) + : undefined, };Also add
DEFAULT_BLUR_INTENSITYto the./typesimport list.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@src/components/video-editor/projectPersistence.ts` around lines 329 - 393, normalizeProjectEditor currently drops blur redaction state: update the normalizedAnnotationRegions mapping in projectPersistence.ts to preserve blurIntensity on every region by adding a blurIntensity property (e.g. blurIntensity: isFiniteNumber(region.blurIntensity) ? region.blurIntensity : DEFAULT_BLUR_INTENSITY) to the returned AnnotationRegion object, and add DEFAULT_BLUR_INTENSITY to the import list from ./types so the default is available.
🧹 Nitpick comments (11)
src/index.css (1)
125-140: Consider adding Firefox scrollbar support.The
.subtle-scrollbarclass currently only styles WebKit-based browsers (Chrome, Safari, Edge). Firefox users won't see the custom scrollbar styling. Consider adding Firefox-specific properties for broader browser support.🦊 Proposed addition for Firefox support
+ .subtle-scrollbar { + scrollbar-width: thin; + scrollbar-color: rgba(255, 255, 255, 0.1) transparent; + } + .subtle-scrollbar::-webkit-scrollbar { width: 4px; }🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@src/index.css` around lines 125 - 140, The .subtle-scrollbar rules only target WebKit; add Firefox support by adding Firefox-specific properties for the .subtle-scrollbar class: set scrollbar-width: thin and use scrollbar-color with the same thumb and track colors (e.g., scrollbar-color: rgba(255,255,255,0.1) transparent) and add a :hover variant (e.g., scrollbar-color: rgba(255,255,255,0.2) transparent) so the existing ::-webkit-scrollbar styles and the new Firefox properties produce consistent appearance across browsers.scripts/build-whisper-runtime.mjs (2)
140-144: Inconsistent path quoting compared to sibling scripts.This returns the unquoted path (
return localCmake), whilebuild-windows-capture.mjsandbuild-cursor-monitor.mjsreturn quoted paths (return \"${localCmake}"`). If the.cmake_ext` directory or any parent path contains spaces, this script will fail while the others succeed.Since this script uses
execFileSync(which handles arguments as an array) rather thanexecSync(which uses shell string parsing), the unquoted path is actually correct here. However, the inconsistency is worth noting for future maintenance.📝 Clarification
The difference is intentional based on how each script invokes CMake:
build-whisper-runtime.mjsusesexecFileSync(cmake, [...args])— array-based, no shell parsing, no quotes neededbuild-windows-capture.mjsusesexecSync(\${cmake} ...`)` — string-based shell command, quotes needed for paths with spacesBoth are correct for their respective usage patterns.
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@scripts/build-whisper-runtime.mjs` around lines 140 - 144, The localCmake return in build-whisper-runtime.mjs currently returns an unquoted path (localCmake) which differs from sibling scripts; because this file uses execFileSync (array args) the unquoted form is correct — update the code by leaving the unquoted return but add a short clarifying comment near the localCmake declaration/return explaining that execFileSync is used (and therefore no quoting is necessary), referencing localCmake and execFileSync so future maintainers understand the intentional difference.
245-245: Relative path usage could be fragile if working directory changes.Using
path.relative(projectRoot, archivePath)assumes thetarcommand runs fromprojectRoot. WhileexecFileSyncinheritsprocess.cwd()(which equalsprojectRoothere), this implicit dependency could break if:
- The script's working directory changes earlier in execution
- The code is refactored to run from a different directory
Consider using absolute paths for robustness, or explicitly setting
cwd: projectRootin theexecFileSyncoptions.♻️ Proposed fix using absolute paths
- execFileSync("tar", ["-xzf", path.relative(projectRoot, archivePath), "-C", path.relative(projectRoot, extractRoot)], { stdio: "inherit" }); + execFileSync("tar", ["-xzf", archivePath, "-C", extractRoot], { stdio: "inherit", cwd: projectRoot });Alternatively, keep relative paths but make the
cwdexplicit:- execFileSync("tar", ["-xzf", path.relative(projectRoot, archivePath), "-C", path.relative(projectRoot, extractRoot)], { stdio: "inherit" }); + execFileSync("tar", ["-xzf", path.relative(projectRoot, archivePath), "-C", path.relative(projectRoot, extractRoot)], { stdio: "inherit", cwd: projectRoot });🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@scripts/build-whisper-runtime.mjs` at line 245, The execFileSync call that runs "tar" uses path.relative(projectRoot, archivePath) and path.relative(projectRoot, extractRoot), which is fragile if cwd changes; update the execFileSync invocation in scripts/build-whisper-runtime.mjs (the execFileSync call invoking "tar") to use absolute paths (e.g., path.resolve(projectRoot, archivePath) and path.resolve(projectRoot, extractRoot)) or, alternatively, keep the relative paths but pass an explicit cwd option ({ cwd: projectRoot, stdio: "inherit" }) so the tar command does not depend on process.cwd(); ensure you update the references to projectRoot, archivePath, and extractRoot accordingly.scripts/build-windows-capture.mjs (1)
28-32: Consider extracting the hardcoded CMake version to a shared constant.The path
cmake-4.3.0-windows-x86_64is duplicated acrossbuild-windows-capture.mjs,build-cursor-monitor.mjs(line 29), andbuild-whisper-runtime.mjs(line 141). If the bundled CMake version changes, all three files must be updated in lockstep.Consider extracting this to a shared configuration or constant file imported by all build scripts.
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@scripts/build-windows-capture.mjs` around lines 28 - 32, The hardcoded bundled CMake directory string ("cmake-4.3.0-windows-x86_64") is duplicated (see localCmake variable) across multiple build scripts; extract that segment into a single exported constant (e.g., BUNDLED_CMAKE_DIR or BUNDLED_CMAKE_VERSION) in a shared module (like buildConstants.mjs) and import it into each script (used where localCmake is constructed in build-windows-capture.mjs, build-cursor-monitor.mjs, build-whisper-runtime.mjs), then replace the inline string with the imported constant so updates require changing only the shared constant.scripts/build-cursor-monitor.mjs (1)
19-76: Optional: Consider extracting shared CMake discovery logic.The
findCmake()function is nearly identical acrossbuild-windows-capture.mjs,build-cursor-monitor.mjs, and partially inbuild-whisper-runtime.mjs. Extracting to a shared module would reduce maintenance burden.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@scripts/build-cursor-monitor.mjs` around lines 19 - 76, The findCmake() implementation is duplicated across build-cursor-monitor.mjs, build-windows-capture.mjs and partially in build-whisper-runtime.mjs; extract this logic into a single shared module (e.g., export function findCmake from a new scripts/find-cmake.mjs), replace the local findCmake definitions with imports (import { findCmake } from './find-cmake.mjs') in each of those files, and ensure the shared function preserves existing behavior (returns "cmake", quoted local/standalone/VS paths, or null) and any callers (e.g., in build-cursor-monitor.mjs) continue to work with the same function name and return contract.src/components/video-editor/AnnotationOverlay.tsx (1)
114-123: Minor: Redundantbackdrop-blur-mdTailwind class.The
backdrop-blur-mdTailwind class is immediately overridden by the inlinebackdropFilterstyle. Consider removing the Tailwind class for clarity since the dynamic blur intensity is always applied via inline styles.♻️ Suggested cleanup
case 'blur': return ( <div - className="w-full h-full rounded-lg backdrop-blur-md" + className="w-full h-full rounded-lg" style={{ backdropFilter: `blur(${annotation.blurIntensity ?? 12}px)`, WebkitBackdropFilter: `blur(${annotation.blurIntensity ?? 12}px)` }} /> );🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@src/components/video-editor/AnnotationOverlay.tsx` around lines 114 - 123, In the AnnotationOverlay component inside the switch branch for case 'blur', remove the redundant Tailwind class "backdrop-blur-md" from the returned div since the blur is applied dynamically via the inline styles using annotation.blurIntensity; keep the inline style keys backdropFilter and WebkitBackdropFilter as-is so the dynamic blur value is retained and the visual output remains unchanged.src/lib/exporter/annotationRenderer.ts (1)
345-345: Replaceanytype assertion with a proper type.The static analysis correctly flags the
anycast. While this works at runtime, you can use a more specific type:♻️ Suggested fix
// Apply the blur filter and draw the captured region back ctx.filter = `blur(${intensity}px)`; - ctx.drawImage(offscreen as any, srcX, srcY); + ctx.drawImage(offscreen as CanvasImageSource, srcX, srcY);Alternatively, if TypeScript still complains about
OffscreenCanvas, you can use:ctx.drawImage(offscreen as unknown as CanvasImageSource, srcX, srcY);🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@src/lib/exporter/annotationRenderer.ts` at line 345, The drawImage call uses an unsafe any cast on the offscreen variable; replace the `offscreen as any` in the ctx.drawImage call with a proper type such as `CanvasImageSource` (e.g., cast offscreen to CanvasImageSource) or, if the compiler still complains about OffscreenCanvas, use a double-cast like `offscreen as unknown as CanvasImageSource`; update the call in the code that contains ctx.drawImage(offscreen as any, srcX, srcY) so the type is explicit and not any.src/components/video-editor/VideoPlayback.tsx (1)
670-674: Consider validating seek time parameter.The
seekmethod passes the time directly tovideo.currentTimewithout validation. While browsers typically clamp out-of-range values, explicit validation would make the behavior more predictable:🛡️ Optional validation
seek: (time: number) => { const video = videoRef.current; if (!video) return; + const clampedTime = Math.max(0, Math.min(time, video.duration || 0)); - video.currentTime = time; + video.currentTime = clampedTime; },🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@src/components/video-editor/VideoPlayback.tsx` around lines 670 - 674, The seek method on the VideoPlayback component sets video.currentTime directly from the incoming time; validate and clamp the value before assignment by checking that time is a finite number and not NaN, then clamp it to the valid range (>= 0 and <= video.duration when available) and only assign to video.currentTime if valid; reference the seek function, videoRef.current, and video.currentTime when making this change.src/components/video-editor/SettingsPanel.tsx (2)
1473-1476: Replaceanywith proper type assertion.The
anycast bypasses type safety. SinceselectedModelshould be one of the model values, use a proper type assertion.♻️ Proposed fix
value={autoCaptionSettings.selectedModel || "small"} - onValueChange={(value) => updateAutoCaptionSettings({ selectedModel: value as any })} + onValueChange={(value) => updateAutoCaptionSettings({ selectedModel: value as WhisperModelInfo["value"] })}🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@src/components/video-editor/SettingsPanel.tsx` around lines 1473 - 1476, The Select is casting the incoming value to any which loses type safety; change the onValueChange handler to assert the value to the proper model type used by autoCaptionSettings (e.g. the union/enum type for selectedModel) instead of any. Locate the Select in SettingsPanel.tsx and update the onValueChange call to pass value as the correct type (the same type used for autoCaptionSettings.selectedModel) when calling updateAutoCaptionSettings so TypeScript can validate the model values.
1556-1564: Replaceanywith proper type for generation range.Use a specific type assertion instead of
anyfor better type safety.♻️ Proposed fix
onValueChange={(val) => val && onAutoCaptionSettingsChange?.({ ...autoCaptionSettings!, - generationRange: val as any, + generationRange: val as "full" | "selected", }) }🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@src/components/video-editor/SettingsPanel.tsx` around lines 1556 - 1564, The code is using a broad "any" cast for generationRange; replace it with the proper type used by autoCaptionSettings (e.g., the GenerationRange enum or union type) so TypeScript can validate values. Locate the value setter in SettingsPanel (the props/variables autoCaptionSettings and onAutoCaptionSettingsChange) and change the assertion from "as any" to "as GenerationRange" (or the exact union type like "'full' | 'partial'") or use a typed parser/mapper that converts the incoming string to the correct GenerationRange value before calling onAutoCaptionSettingsChange.src/components/video-editor/timeline/TimelineEditor.tsx (1)
445-452: Consider typed interface for internal handler access.The
anycasts bypass TypeScript's type safety when accessing internal handlers. While functional, this pattern is fragile.♻️ Suggested improvement using a typed interface
+interface TimelineContainerElement extends HTMLDivElement { + __handleMouseDown?: (e: React.MouseEvent<HTMLDivElement>) => void; + __handleTimelineClick?: (e: React.MouseEvent<HTMLDivElement>) => void; +} // In TimelineAxis: onMouseDown={(e) => { - (e.currentTarget.parentElement as any)?.__handleMouseDown?.(e); + (e.currentTarget.parentElement as TimelineContainerElement)?.__handleMouseDown?.(e); }} onClick={(e) => { - (e.currentTarget.parentElement as any)?.__handleTimelineClick?.(e); + (e.currentTarget.parentElement as TimelineContainerElement)?.__handleTimelineClick?.(e); }}Also update lines 720-721:
- (localTimelineRef.current as any).__handleMouseDown = handleMouseDown; - (localTimelineRef.current as any).__handleTimelineClick = handleTimelineClick; + (localTimelineRef.current as TimelineContainerElement).__handleMouseDown = handleMouseDown; + (localTimelineRef.current as TimelineContainerElement).__handleTimelineClick = handleTimelineClick;🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@src/components/video-editor/timeline/TimelineEditor.tsx` around lines 445 - 452, Replace the unsafe any casts when calling parent internal handlers by defining and using a small typed interface (e.g., an interface with optional methods __handleMouseDown(e: MouseEvent) and __handleTimelineClick(e: MouseEvent)) and cast e.currentTarget.parentElement to that interface when invoking __handleMouseDown and __handleTimelineClick; update both the onMouseDown and onClick sites (the lines invoking __handleMouseDown and __handleTimelineClick) and the other occurrences mentioned (around lines 720-721) to use this typed interface so TypeScript enforces the handler signatures instead of using any.
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In `@electron/ipc/handlers.ts`:
- Around line 1635-1653: Chunked caption IDs currently restart (e.g.,
"caption-1") per chunk causing the renderer's dedupe to drop later chunks;
update parsing and emission to namespace IDs per chunk: change
parseWhisperJsonCues() and parseSrtCues() to accept a chunkPrefix (or
chunkIndex/offsetMs) and produce IDs like `${chunkPrefix}-caption-${n}`, then
ensure before pushing/sending you apply that prefix to each cue in adjustedCues
(where allCues.push(...adjustedCues) and safeSend(webContents,
'auto-caption-chunk', { cues: adjustedCues }) are called) so every chunk's cues
have globally unique IDs and won’t be de-duplicated by the renderer.
- Around line 1608-1664: The loop treats offsetMs as if it were range-relative
which causes early termination when startTimeMs > 0; fix by converting to
range-relative values before doing progress/last-chunk checks: compute
rangeRelativeOffset = offsetMs - startTimeMs and rangeDurationMs =
totalDurationMs - startTimeMs (when totalDurationMs>0), then use
rangeRelativeOffset in updateChunkProgress math, in the isLastChunk check inside
the .filter (replace offsetMs with rangeRelativeOffset), and in the final break
condition (compare rangeRelativeOffset + CHUNK_SIZE_MS >= rangeDurationMs).
Update references to offsetMs in updateChunkProgress, the isLastChunk
calculation, and the final totalDurationMs break logic accordingly.
In `@electron/native/wgc-capture/src/monitor_utils.cpp`:
- Around line 49-65: The current early returns for a 32-bit partial ID match
(using m.handle and displayId) and the single-monitor fallback prevent the
bounds-based correction in findMonitorByBounds (called from main.cpp) from ever
running; fix by deferring these weak-ID fallbacks: either call the
bounds-matching helper (findMonitorByBounds or equivalent) here before returning
the 32-bit match/single-monitor handle, or change these branches to return
nullptr (instead of returning monitors[0].handle or m.handle) so the caller can
invoke findMonitorByBounds(displayId) and correct the monitor selection. Ensure
you update the code paths that reference m.handle, monitors, and displayId
accordingly.
In `@src/components/video-editor/AnnotationSettingsPanel.tsx`:
- Around line 148-170: The new "Blur" tab and its label "Blur Intensity" are
hard-coded; update AnnotationSettingsPanel (the TabsTrigger with value "blur"
and the label showing Blur Intensity) to use the localization function t(...)
like other labels in the panel—replace the literal "Blur" in the TabsTrigger and
the literal "Blur Intensity" in the label (which displays
annotation.blurIntensity ?? DEFAULT_BLUR_INTENSITY) with localized strings
(e.g., t('...')) and ensure keys are consistent with the panel's i18n keys so
onBlurIntensityChange and Slider behavior remain unchanged.
In `@src/components/video-editor/AudioSettingsPanel.tsx`:
- Around line 69-85: The new AudioSettingsPanel component is shipping hard-coded
English labels (e.g., "Original Audio", "Audio Region", "Adjust the volume of
the video's audio", and other strings between the noted ranges) — update
AudioSettingsPanel to use the project's scoped translation hook/util used by
neighboring editor panels (e.g., the same useTranslation/useTranslations hook)
and replace all literal strings (including "Original Audio", "Audio Region",
"Mute", "Volume", "Fade In", "Remove Audio Region", the helper text for
isMaster, and filenames rendered from audio.audioPath) with translation keys;
import and call the same translation function at the top of the
AudioSettingsPanel, replace the hard-coded literals with t('...') (or the
project's equivalent) using descriptive keys, and ensure keys are added to the
appropriate locale files for the ranges indicated (69-85, 130-175, 183-213).
- Around line 36-40: The waveform loading in useEffect should be guarded against
race conditions and rejections: when audio.audioPath changes, create a
cancellable context (e.g., an AbortController or a local "stale" flag) before
calling generateWaveform(audio.audioPath, 120), call .then(result => { if
(!aborted) setWaveform(result) }) and add .catch(err => { if (!aborted)
handle/log error or setWaveform(null) }); also perform cleanup in the effect
return to mark the request aborted/stale so a previous async load cannot
overwrite the waveform for a newly selected clip; update references to
generateWaveform, setWaveform, and the useEffect dependency on audio.audioPath.
In `@src/components/video-editor/editorPreferences.ts`:
- Around line 36-38: The preferences code is persisting
masterAudioMuted/masterAudioSoloed/masterAudioVolume but omits audioTrackVolume,
so add "audioTrackVolume" wherever the master audio controls are enumerated and
processed: include the "audioTrackVolume" key alongside
"masterAudioMuted"/"masterAudioSoloed"/"masterAudioVolume" in the preference
keys/union, in the defaults/initial state, and in the
serialization/deserialization/save/load logic (the same spots that currently
reference those three master audio keys) so that audioTrackVolume is read from
and written to storage and restored on reload.
In `@src/components/video-editor/SettingsPanel.tsx`:
- Around line 1850-1860: The master audio block in SettingsPanel.tsx renders
AudioSettingsPanel with no-op callbacks for onFadeInMsChange, onFadeOutMsChange
and onDelete which leads to clickable but non-functional controls; update either
AudioSettingsPanel or its usage: modify AudioSettingsPanel to accept these
handlers as optional props and conditionally render fade/delete controls when
handlers are provided, or add a boolean prop (e.g. isMaster or variant="master")
to AudioSettingsPanel and hide fade/delete UI when isMaster is true, then update
the SettingsPanel.tsx instantiation (the AudioSettingsPanel call with
masterAudioMock) to pass the new prop instead of no-op functions so master audio
controls are not shown as non-functional.
In `@src/components/video-editor/timeline/Item.tsx`:
- Around line 99-105: The custom pointer handler handlePointerDown captures
mouseDownPosRef but currently calls only listeners.onPointerDown, dropping other
dnd-kit handlers from useItem (breaking keyboard drag/accessibility); instead
stop invoking listeners.onPointerDown directly and apply the entire listeners
object to the interactive element (e.g., spread {...listeners} on the element
that uses handlePointerDown) so all keyboard and accessibility handlers are
preserved while still setting mouseDownPosRef in handlePointerDown; make the
same change for the other similar handler usage referenced (the second
occurrence that currently calls listeners.onPointerDown).
- Around line 89-95: The useEffect in the Item component currently can
setWaveform from a stale generateWaveform promise after audioPath changes or on
unmount; update the effect to support cancellation by creating a local cancelled
flag or AbortController at the top of the effect, call generateWaveform with an
abort signal if it accepts one (or otherwise capture the current token), and in
the promise resolution check the flag/signal before calling setWaveform; also
return a cleanup function that flips the cancelled flag or aborts the controller
to prevent state updates on unmounted/stale items (referencing Item, useEffect,
generateWaveform, and setWaveform).
In `@src/components/video-editor/timeline/Row.tsx`:
- Around line 21-35: The overlay wrapper div that contains the label and
controls is blocking pointer events on the track; update the wrapper (the div
with className "absolute left-1.5 top-1/2 -translate-y-1/2 z-20 flex
items-center gap-2") to be pointer-events-none, ensure the label element (the
div rendering {label}) remains pointer-events-none, and keep the control nodes
as pointer-events-auto so they can receive input (controls already opt into it)
— this makes the overlay transparent to clicks/drags except for the controls.
In `@src/components/video-editor/timeline/TimelineEditor.tsx`:
- Around line 313-323: The handleMouseUp callback reads the timeSelection state
but the useEffect dependency array (the array ending with keyframes) does not
include timeSelection, causing a stale closure; update the dependency array for
the effect that registers handleMouseUp to include timeSelection (or
alternatively refactor handleMouseUp to read a mutable ref that you update when
timeSelection changes) so handleMouseUp always sees the latest timeSelection
value—refer to the handleMouseUp function and the useEffect whose dependencies
currently list isDragging, onSeek, timelineRef, sidebarWidth, range.start,
range.end, videoDurationMs, pixelsToValue, keyframes.
In `@src/components/video-editor/VideoEditor.tsx`:
- Around line 2996-3000: The export path is not honoring per-region or master
solo state—when calling the exporter (where audioRegions, masterAudioVolume,
audioTrackVolume, masterAudioMuted, previewWidth are passed) also forward
masterAudioSoloed and each region's soloed flag, and then update the audio
mixing code in src/lib/exporter/audioEncoder.ts to consume these new flags
(e.g., apply masterAudioSoloed to mute nonsoloed tracks and respect
region.soloed when building the mix). Ensure the exporter interface and any call
sites accept the new solo-related params and that audioEncoder's mixing logic
gives priority to soloed tracks consistent with preview playback.
- Around line 1165-1169: The project snapshot creation omits the new audio mix
fields restored by applyLoadedProject(): masterAudioMuted, masterAudioSoloed,
masterAudioVolume, and audioTrackVolume, so edits to them aren’t persisted or
tracked by hasUnsavedChanges; update the snapshot object (the place where the
project snapshot is constructed prior to saving/serializing) to include these
four fields with the same keys used in applyLoadedProject(), ensure any
equality/dirty-check logic that computes hasUnsavedChanges compares these fields
as well, and update any serialization/deserialization mappings so saved projects
round-trip these audio mix properties.
- Around line 3880-3881: The SettingsPanel is calling the state setter directly
(onSelectCaption={setSelectedCaptionId}) which bypasses the clearing/side-effect
logic in handleSelectCaption; change the prop to use the shared handler instead
(onSelectCaption={handleSelectCaption}) so all caption selections go through
handleSelectCaption and its clearing logic (refer to handleSelectCaption and
selectedCaptionId/onSelectCaption props to locate the usage).
- Around line 1496-1512: The effect using getWhisperModelStatus must clear
whisperModelPath whenever autoCaptionSettings.selectedModel changes so stale
paths aren't reused; in the useEffect that calls
window.electronAPI.getWhisperModelStatus(autoCaptionSettings.selectedModel)
ensure you call setWhisperModelPath(null) (or set to result.path when present)
before any early returns on failure and when result.exists is false, rather than
preserving the previous path; update the logic around
setWhisperModelDownloadStatus/setWhisperModelDownloadProgress and remove the
early-return that prevents clearing whisperModelPath so switching to another
built-in or the "custom" option cannot reuse an old file.
In `@src/lib/exporter/annotationRenderer.ts`:
- Around line 307-316: The clip path and draw operations in
annotationRenderer.ts are still using the original width/height while the actual
captured region is clamped to srcW/srcH, causing misalignment when annotations
extend past canvas edges; update the clip path creation and any subsequent
drawing/putImageData calls to use srcW and srcH (and adjust x/y offsets
accordingly) instead of width/height so the clipped region matches the captured
ImageData, and ensure any scaling or translation applied to the blur rendering
accounts for the difference between (width,height) and (srcW,srcH) to preserve
visual position (use the existing srcX/srcY, srcW/srcH, x, y, width, height
variables to compute offsets).
In `@src/lib/exporter/audioEncoder.ts`:
- Around line 28-30: The export mix currently ignores solo/mute state (it only
checks masterAudioMuted); update the volume calculation used by totalVolume in
src/lib/exporter/audioEncoder.ts (and the equivalent calculations around the
other occurrences) to mirror the preview logic in VideoEditor.tsx: consider
masterAudioSoloed, per-region audioRegions[].soloed, and audioRegions[].muted
when deciding which regions contribute to the mix and when computing each
region's effective volume (combine masterAudioVolume, audioTrackVolume, region
volume, and mute/solo rules). Specifically, detect if any region or the master
is soloed and apply solo/mute precedence so exported audio matches playback.
In `@src/utils/audioWaveform.ts`:
- Around line 16-22: The AudioContext created in the decoding path (variable
audioContext) uses an unsafe (window as any).webkitAudioContext cast and is
never closed; update the fallback constructor to a proper typed access (e.g.,
read webkitAudioContext off window with a narrow type) and ensure
audioContext.close() is called after use: close it after successful decode and
also in the catch/finally path so the transient AudioContext is always released;
locate the creation of audioContext and the decodeAudioData usage to implement
the typed fallback and add the close() calls.
---
Outside diff comments:
In `@src/components/video-editor/projectPersistence.ts`:
- Around line 406-415: The save/restore logic in projectPersistence.ts currently
clamps persisted region.volume to a max of 1, which strips gains >100% (the UI
supports 0–200%). Update the volume normalization in the region mapping (the
line using isFiniteNumber(region.volume) ? clamp(region.volume, 0, 1) : 1) to
allow up to 2 (e.g. clamp(region.volume, 0, 2)) so persisted volumes up to 200%
are preserved while keeping the same default fallback of 1.
- Around line 329-393: normalizeProjectEditor currently drops blur redaction
state: update the normalizedAnnotationRegions mapping in projectPersistence.ts
to preserve blurIntensity on every region by adding a blurIntensity property
(e.g. blurIntensity: isFiniteNumber(region.blurIntensity) ? region.blurIntensity
: DEFAULT_BLUR_INTENSITY) to the returned AnnotationRegion object, and add
DEFAULT_BLUR_INTENSITY to the import list from ./types so the default is
available.
In `@src/components/video-editor/timeline/ItemGlass.module.css`:
- Around line 168-180: The hover selector for the solid cyan caption variant is
missing; add a rule matching .glassCyan:hover .zoomEndCap (in the same selector
group where .glassCyan.selected .zoomEndCap and other .glassX:hover .zoomEndCap
rules live) so that non-selected cyan caption blocks reveal their resize caps on
hover just like the other solid color variants.
---
Nitpick comments:
In `@scripts/build-cursor-monitor.mjs`:
- Around line 19-76: The findCmake() implementation is duplicated across
build-cursor-monitor.mjs, build-windows-capture.mjs and partially in
build-whisper-runtime.mjs; extract this logic into a single shared module (e.g.,
export function findCmake from a new scripts/find-cmake.mjs), replace the local
findCmake definitions with imports (import { findCmake } from
'./find-cmake.mjs') in each of those files, and ensure the shared function
preserves existing behavior (returns "cmake", quoted local/standalone/VS paths,
or null) and any callers (e.g., in build-cursor-monitor.mjs) continue to work
with the same function name and return contract.
In `@scripts/build-whisper-runtime.mjs`:
- Around line 140-144: The localCmake return in build-whisper-runtime.mjs
currently returns an unquoted path (localCmake) which differs from sibling
scripts; because this file uses execFileSync (array args) the unquoted form is
correct — update the code by leaving the unquoted return but add a short
clarifying comment near the localCmake declaration/return explaining that
execFileSync is used (and therefore no quoting is necessary), referencing
localCmake and execFileSync so future maintainers understand the intentional
difference.
- Line 245: The execFileSync call that runs "tar" uses
path.relative(projectRoot, archivePath) and path.relative(projectRoot,
extractRoot), which is fragile if cwd changes; update the execFileSync
invocation in scripts/build-whisper-runtime.mjs (the execFileSync call invoking
"tar") to use absolute paths (e.g., path.resolve(projectRoot, archivePath) and
path.resolve(projectRoot, extractRoot)) or, alternatively, keep the relative
paths but pass an explicit cwd option ({ cwd: projectRoot, stdio: "inherit" })
so the tar command does not depend on process.cwd(); ensure you update the
references to projectRoot, archivePath, and extractRoot accordingly.
In `@scripts/build-windows-capture.mjs`:
- Around line 28-32: The hardcoded bundled CMake directory string
("cmake-4.3.0-windows-x86_64") is duplicated (see localCmake variable) across
multiple build scripts; extract that segment into a single exported constant
(e.g., BUNDLED_CMAKE_DIR or BUNDLED_CMAKE_VERSION) in a shared module (like
buildConstants.mjs) and import it into each script (used where localCmake is
constructed in build-windows-capture.mjs, build-cursor-monitor.mjs,
build-whisper-runtime.mjs), then replace the inline string with the imported
constant so updates require changing only the shared constant.
In `@src/components/video-editor/AnnotationOverlay.tsx`:
- Around line 114-123: In the AnnotationOverlay component inside the switch
branch for case 'blur', remove the redundant Tailwind class "backdrop-blur-md"
from the returned div since the blur is applied dynamically via the inline
styles using annotation.blurIntensity; keep the inline style keys backdropFilter
and WebkitBackdropFilter as-is so the dynamic blur value is retained and the
visual output remains unchanged.
In `@src/components/video-editor/SettingsPanel.tsx`:
- Around line 1473-1476: The Select is casting the incoming value to any which
loses type safety; change the onValueChange handler to assert the value to the
proper model type used by autoCaptionSettings (e.g. the union/enum type for
selectedModel) instead of any. Locate the Select in SettingsPanel.tsx and update
the onValueChange call to pass value as the correct type (the same type used for
autoCaptionSettings.selectedModel) when calling updateAutoCaptionSettings so
TypeScript can validate the model values.
- Around line 1556-1564: The code is using a broad "any" cast for
generationRange; replace it with the proper type used by autoCaptionSettings
(e.g., the GenerationRange enum or union type) so TypeScript can validate
values. Locate the value setter in SettingsPanel (the props/variables
autoCaptionSettings and onAutoCaptionSettingsChange) and change the assertion
from "as any" to "as GenerationRange" (or the exact union type like "'full' |
'partial'") or use a typed parser/mapper that converts the incoming string to
the correct GenerationRange value before calling onAutoCaptionSettingsChange.
In `@src/components/video-editor/timeline/TimelineEditor.tsx`:
- Around line 445-452: Replace the unsafe any casts when calling parent internal
handlers by defining and using a small typed interface (e.g., an interface with
optional methods __handleMouseDown(e: MouseEvent) and __handleTimelineClick(e:
MouseEvent)) and cast e.currentTarget.parentElement to that interface when
invoking __handleMouseDown and __handleTimelineClick; update both the
onMouseDown and onClick sites (the lines invoking __handleMouseDown and
__handleTimelineClick) and the other occurrences mentioned (around lines
720-721) to use this typed interface so TypeScript enforces the handler
signatures instead of using any.
In `@src/components/video-editor/VideoPlayback.tsx`:
- Around line 670-674: The seek method on the VideoPlayback component sets
video.currentTime directly from the incoming time; validate and clamp the value
before assignment by checking that time is a finite number and not NaN, then
clamp it to the valid range (>= 0 and <= video.duration when available) and only
assign to video.currentTime if valid; reference the seek function,
videoRef.current, and video.currentTime when making this change.
In `@src/index.css`:
- Around line 125-140: The .subtle-scrollbar rules only target WebKit; add
Firefox support by adding Firefox-specific properties for the .subtle-scrollbar
class: set scrollbar-width: thin and use scrollbar-color with the same thumb and
track colors (e.g., scrollbar-color: rgba(255,255,255,0.1) transparent) and add
a :hover variant (e.g., scrollbar-color: rgba(255,255,255,0.2) transparent) so
the existing ::-webkit-scrollbar styles and the new Firefox properties produce
consistent appearance across browsers.
In `@src/lib/exporter/annotationRenderer.ts`:
- Line 345: The drawImage call uses an unsafe any cast on the offscreen
variable; replace the `offscreen as any` in the ctx.drawImage call with a proper
type such as `CanvasImageSource` (e.g., cast offscreen to CanvasImageSource) or,
if the compiler still complains about OffscreenCanvas, use a double-cast like
`offscreen as unknown as CanvasImageSource`; update the call in the code that
contains ctx.drawImage(offscreen as any, srcX, srcY) so the type is explicit and
not any.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Repository UI
Review profile: CHILL
Plan: Pro
Run ID: 9242e161-8b04-4058-beee-f40fb88ce0dc
⛔ Files ignored due to path filters (5)
electron/native/bin/win32-x64/whisper-bench.exeis excluded by!**/*.exeelectron/native/bin/win32-x64/whisper-cli.exeis excluded by!**/*.exeelectron/native/bin/win32-x64/whisper-quantize.exeis excluded by!**/*.exeelectron/native/bin/win32-x64/whisper-server.exeis excluded by!**/*.exeelectron/native/bin/win32-x64/whisper-vad-speech-segments.exeis excluded by!**/*.exe
📒 Files selected for processing (36)
.gitignoreelectron-builder.json5electron/electron-env.d.tselectron/ipc/handlers.tselectron/native/bin/win32-x64/whisper-runtime.jsonelectron/native/wgc-capture/src/main.cppelectron/native/wgc-capture/src/monitor_utils.cppelectron/native/wgc-capture/src/monitor_utils.helectron/preload.tselectron/windows.tspackage.jsonscripts/build-cursor-monitor.mjsscripts/build-whisper-runtime.mjsscripts/build-windows-capture.mjssrc/components/video-editor/AnnotationOverlay.tsxsrc/components/video-editor/AnnotationSettingsPanel.tsxsrc/components/video-editor/AudioSettingsPanel.tsxsrc/components/video-editor/SettingsPanel.tsxsrc/components/video-editor/VideoEditor.tsxsrc/components/video-editor/VideoPlayback.tsxsrc/components/video-editor/editorPreferences.tssrc/components/video-editor/projectPersistence.tssrc/components/video-editor/timeline/Item.tsxsrc/components/video-editor/timeline/ItemGlass.module.csssrc/components/video-editor/timeline/KeyframeMarkers.tsxsrc/components/video-editor/timeline/Row.tsxsrc/components/video-editor/timeline/TimelineEditor.tsxsrc/components/video-editor/timeline/TimelineWrapper.tsxsrc/components/video-editor/types.tssrc/components/video-editor/videoPlayback/videoEventHandlers.tssrc/index.csssrc/lib/exporter/annotationRenderer.tssrc/lib/exporter/audioEncoder.tssrc/lib/exporter/types.tssrc/lib/exporter/videoExporter.tssrc/utils/audioWaveform.ts
Comment on lines
+148
to
+170
| <TabsTrigger value="blur" className="data-[state=active]:bg-[#2563EB] data-[state=active]:text-white text-slate-400 py-2 rounded-lg transition-all gap-2"> | ||
| <Droplets className="w-4 h-4" /> | ||
| Blur | ||
| </TabsTrigger> | ||
| </TabsList> | ||
|
|
||
| <TabsContent value="blur" className="mt-0 space-y-4"> | ||
| <div> | ||
| <label className="text-xs font-medium text-slate-200 mb-2 block"> | ||
| Blur Intensity: {annotation.blurIntensity ?? DEFAULT_BLUR_INTENSITY}px | ||
| </label> | ||
| <Slider | ||
| value={[annotation.blurIntensity ?? DEFAULT_BLUR_INTENSITY]} | ||
| onValueChange={([value]) => { | ||
| if (onBlurIntensityChange) { | ||
| onBlurIntensityChange(value); | ||
| } | ||
| }} | ||
| min={2} | ||
| max={50} | ||
| step={1} | ||
| className="w-full" | ||
| /> |
There was a problem hiding this comment.
Localize the new blur labels.
"Blur" and "Blur Intensity" are hard-coded while the rest of this panel already goes through t(...). In non-English locales this new tab will be the odd untranslated section.
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In `@src/components/video-editor/AnnotationSettingsPanel.tsx` around lines 148 -
170, The new "Blur" tab and its label "Blur Intensity" are hard-coded; update
AnnotationSettingsPanel (the TabsTrigger with value "blur" and the label showing
Blur Intensity) to use the localization function t(...) like other labels in the
panel—replace the literal "Blur" in the TabsTrigger and the literal "Blur
Intensity" in the label (which displays annotation.blurIntensity ??
DEFAULT_BLUR_INTENSITY) with localized strings (e.g., t('...')) and ensure keys
are consistent with the panel's i18n keys so onBlurIntensityChange and Slider
behavior remain unchanged.
Comment on lines
+36
to
+40
| useEffect(() => { | ||
| if (audio.audioPath) { | ||
| generateWaveform(audio.audioPath, 120).then(setWaveform); | ||
| } | ||
| }, [audio.audioPath]); |
There was a problem hiding this comment.
Guard the async waveform load.
generateWaveform(...).then(setWaveform) has no cleanup or rejection handling. Switching selections quickly can paint the previous clip's waveform into the next panel, and decode failures become unhandled rejections.
💡 Suggested fix
useEffect(() => {
- if (audio.audioPath) {
- generateWaveform(audio.audioPath, 120).then(setWaveform);
- }
+ let cancelled = false;
+ setWaveform(null);
+
+ if (!audio.audioPath) {
+ return () => {
+ cancelled = true;
+ };
+ }
+
+ generateWaveform(audio.audioPath, 120)
+ .then((nextWaveform) => {
+ if (!cancelled) setWaveform(nextWaveform);
+ })
+ .catch(() => {
+ if (!cancelled) setWaveform(null);
+ });
+
+ return () => {
+ cancelled = true;
+ };
}, [audio.audioPath]);🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In `@src/components/video-editor/AudioSettingsPanel.tsx` around lines 36 - 40, The
waveform loading in useEffect should be guarded against race conditions and
rejections: when audio.audioPath changes, create a cancellable context (e.g., an
AbortController or a local "stale" flag) before calling
generateWaveform(audio.audioPath, 120), call .then(result => { if (!aborted)
setWaveform(result) }) and add .catch(err => { if (!aborted) handle/log error or
setWaveform(null) }); also perform cleanup in the effect return to mark the
request aborted/stale so a previous async load cannot overwrite the waveform for
a newly selected clip; update references to generateWaveform, setWaveform, and
the useEffect dependency on audio.audioPath.
… management bugs in PR webadderall#122 - Implement comprehensive solo/mute export pipeline in audioEncoder.ts - Fix duplicate caption IDs by adding chunk-specific indices in handlers.ts - Prioritize native monitor selection by bounds in monitor_utils.cpp - Fix alignment for blur annotations extending beyond canvas boundaries - Resolve timeline selection desync and sidebar visibility for blur annotations - Fix memory leaks in audio waveform generation with AudioContext cleanup
Contributor
|
|
Contributor
|
|
1 similar comment
Contributor
|
|
There was a problem hiding this comment.
Actionable comments posted: 4
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (2)
src/lib/exporter/audioEncoder.ts (1)
493-515:⚠️ Potential issue | 🟡 MinorDisconnect masterGainNode during cleanup.
The cleanup disconnects
sourceNodeanddestinationNodebut notmasterGainNode, which was connected between them. This may leave a dangling reference.🛠️ Proposed fix
destinationNode.stream.getTracks().forEach((track) => track.stop()) sourceNode.disconnect() + masterGainNode.disconnect() destinationNode.disconnect() await audioContext.close()🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@src/lib/exporter/audioEncoder.ts` around lines 493 - 515, In the finally cleanup block, also disconnect the masterGainNode to avoid dangling connections: locate the cleanup that currently disconnects sourceNode and destinationNode and add a safe check to call masterGainNode.disconnect() (e.g., if (masterGainNode) masterGainNode.disconnect()), optionally followed by nulling it; ensure this runs before closing audioContext or stopping tracks so the graph is fully torn down (refer to masterGainNode, sourceNode, destinationNode, and audioContext).src/components/video-editor/VideoEditor.tsx (1)
896-979:⚠️ Potential issue | 🟡 MinorMaster audio fields missing from unsaved changes detection, but will persist when saving.
currentPersistedEditorStateexcludesmasterAudioMuted,masterAudioSoloed,masterAudioVolume,audioTrackVolume, andisMasterSelected, which meanshasUnsavedChangeswon't detect edits to master audio controls.However, these fields are not lost when saving. The project snapshot saved at line 1210 calls
buildPersistedEditorState(normalizedEditor), wherenormalizedEditoris the result ofnormalizeProjectEditor(line 1102), which includes all master audio fields. So changes to these fields will persist correctly—they just won't be detected as unsaved until a full project reload.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@src/components/video-editor/VideoEditor.tsx` around lines 896 - 979, currentPersistedEditorState omits master audio fields so hasUnsavedChanges won't detect changes; update the buildPersistedEditorState invocation inside currentPersistedEditorState to include masterAudioMuted, masterAudioSoloed, masterAudioVolume, audioTrackVolume, and isMasterSelected, and add those same symbols to the useMemo dependency array so React recomputes when those master-audio values change; reference buildPersistedEditorState and currentPersistedEditorState to locate the call and dependency list to modify.
♻️ Duplicate comments (4)
src/components/video-editor/AudioSettingsPanel.tsx (1)
36-44:⚠️ Potential issue | 🟡 MinorAdd rejection handler to prevent unhandled promise rejections.
The effect correctly guards against stale updates with the
activeflag, addressing the race condition. However, rejected promises fromgenerateWaveformstill become unhandled rejections.🛠️ Proposed fix
useEffect(() => { let active = true; if (audio.audioPath) { - generateWaveform(audio.audioPath, 120).then(result => { - if (active) setWaveform(result); - }); + generateWaveform(audio.audioPath, 120) + .then(result => { + if (active) setWaveform(result); + }) + .catch(() => { + if (active) setWaveform(null); + }); } return () => { active = false; }; }, [audio.audioPath]);🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@src/components/video-editor/AudioSettingsPanel.tsx` around lines 36 - 44, The effect in AudioSettingsPanel uses generateWaveform(audio.audioPath, 120).then(...) but doesn't handle rejections, causing unhandled promise rejections; update the useEffect that references generateWaveform, audio.audioPath, active, and setWaveform to handle errors by attaching a .catch(...) to the promise (or switch to an async IIFE with try/catch) and ensure errors are either logged (e.g., via console.error or a logger) or silenced, while preserving the existing active guard to avoid stale state updates.src/components/video-editor/VideoEditor.tsx (1)
1496-1514:⚠️ Potential issue | 🟠 MajorClear
whisperModelPathon model switch to prevent stale paths.The effect queries model status but only updates
whisperModelPathwhenresult.exists && result.path. When switching from a downloaded model to another model (downloaded or not), the old path persists if the new query returns early or has no path. This can cause transcription to run against the wrong model file.🐛 Proposed fix
useEffect(() => { void (async () => { + // Clear path immediately on model change to prevent stale references + setWhisperModelPath(null); + const result = await window.electronAPI.getWhisperModelStatus( autoCaptionSettings.selectedModel, ); if (!result.success) { return; } if (result.exists && result.path) { - setWhisperModelPath((currentPath) => currentPath ?? result.path ?? null); + setWhisperModelPath(result.path); setWhisperModelDownloadStatus("downloaded"); setWhisperModelDownloadProgress(100); } else { setWhisperModelDownloadStatus("idle"); setWhisperModelDownloadProgress(0); } })(); }, [autoCaptionSettings.selectedModel]);🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@src/components/video-editor/VideoEditor.tsx` around lines 1496 - 1514, When switching models the effect in VideoEditor that calls window.electronAPI.getWhisperModelStatus (triggered by autoCaptionSettings.selectedModel) can leave a stale whisperModelPath; update the useEffect so it proactively clears whisperModelPath at the start of the async check (or explicitly sets it to null when result.exists is false or result.path is missing) by calling setWhisperModelPath(null) before/when handling the response so the previous model path cannot persist and cause transcription to use the wrong file.src/lib/exporter/annotationRenderer.ts (1)
333-345:⚠️ Potential issue | 🟡 MinorClip path still uses original dimensions instead of clamped source bounds.
The past review correctly identified that when the blur annotation extends beyond canvas bounds, the clip path on lines 337-339 uses the original
x, y, width, heightwhile the captured region usessrcX, srcY, srcWidth, srcHeight. This creates visual misalignment because a smaller captured region gets drawn into a larger clipped area.The issue persists despite being marked as addressed. Additionally, the
as anycast on line 345 can be replaced with a proper type.🐛 Proposed fix
// Create rounded rect clipping path (matches UI's rounded-lg approx 8px) const radius = 8 * scaleFactor; ctx.beginPath(); if (ctx.roundRect) { - ctx.roundRect(x, y, width, height, radius); + ctx.roundRect(srcX, srcY, srcWidth, srcHeight, radius); } else { - ctx.rect(x, y, width, height); + ctx.rect(srcX, srcY, srcWidth, srcHeight); } ctx.clip(); // Apply the blur filter and draw the captured region back at its source position ctx.filter = `blur(${intensity}px)`; - ctx.drawImage(offscreen as any, srcX, srcY); + ctx.drawImage(offscreen as CanvasImageSource, srcX, srcY);🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@src/lib/exporter/annotationRenderer.ts` around lines 333 - 345, The clip path currently uses the original annotation bounds (x, y, width, height) which can be larger than the actually captured region; update the clipping rectangle creation to use the clamped capture bounds (srcX, srcY, srcWidth, srcHeight) instead of x/y/width/height so the clip matches what will be drawn, and ensure to convert those source coords to canvas/destination coordinates if needed using scaleFactor; also replace the loose cast on offscreen (the drawImage call) by using a proper type (e.g., cast offscreen to HTMLCanvasElement or OffscreenCanvas) rather than "as any". Ensure you update the calls around ctx.roundRect/ctx.rect/ctx.clip and the ctx.drawImage(offscreen ...) invocation in annotationRenderer.ts to use these variables (srcX, srcY, srcWidth, srcHeight, scaleFactor, offscreen).src/utils/audioWaveform.ts (1)
16-16:⚠️ Potential issue | 🟡 MinorType the webkit fallback and handle AudioContext construction failure.
The
anycast forwebkitAudioContextbypasses type checking. If AudioContext construction fails (e.g., on a server or unsupported environment), the error is uncaught before the try block.🛠️ Proposed fix
- const audioContext = new (window.AudioContext || (window as any).webkitAudioContext)(); - try { + const AudioContextCtor = + window.AudioContext ?? + (window as Window & { webkitAudioContext?: typeof AudioContext }).webkitAudioContext; + if (!AudioContextCtor) { + return new Array(samples).fill(0); + } + + const audioContext = new AudioContextCtor(); + try {🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@src/utils/audioWaveform.ts` at line 16, Replace the untyped immediate construction of the AudioContext so creation can be safely feature-detected and failures caught: detect a constructor reference (e.g. const AudioContextCtor = (window as any).AudioContext || (window as any).webkitAudioContext), give it a proper typed fallback instead of using an unchecked any where possible, check if AudioContextCtor is truthy before attempting to instantiate, then wrap new AudioContextCtor() in a try/catch and handle the failure (e.g. return null or a safe fallback) so the symbol audioContext is only created when supported and construction errors are caught; update any callers that expect audioContext accordingly.
🧹 Nitpick comments (6)
src/components/video-editor/VideoEditor.tsx (2)
2722-2729: Remove unused catch variable.The variable
eis caught but never used. Use an empty catch binding.♻️ Proposed fix
const nodeEntry = audioRegionNodesRef.current.get(id); if (nodeEntry) { try { nodeEntry.source.disconnect(); nodeEntry.gain.disconnect(); - } catch (e) { /* ignore */ } + } catch { /* ignore */ } audioRegionNodesRef.current.delete(id); }🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@src/components/video-editor/VideoEditor.tsx` around lines 2722 - 2729, In the VideoEditor.tsx block where nodeEntry is disconnected (the try/catch inside the logic that handles audioRegionNodesRef.current.delete(id)), remove the unused catch variable by changing the catch clause from catching a named parameter to using an empty catch binding (i.e., replace "catch (e)" with "catch { }") so the thrown error is ignored without declaring an unused identifier; update the catch for the try that calls nodeEntry.source.disconnect() and nodeEntry.gain.disconnect().
1864-1868: Ref value in useEffect dependency array won't trigger re-renders.
videoPlaybackRef.current?.videois a ref accessor in the dependency array. Changes to ref values don't trigger React re-renders, so this effect may not run when expected. ThepreviewVersionandisAudioEngineReadydependencies compensate for this, but the dependency is misleading.♻️ Suggested approach
- }, [videoPlaybackRef.current?.video, previewVersion, isAudioEngineReady]); + // Note: videoPlaybackRef.current?.video is accessed inside the effect but not listed as a dependency + // since refs don't trigger re-renders. previewVersion triggers re-routing when video changes. + }, [previewVersion, isAudioEngineReady]);🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@src/components/video-editor/VideoEditor.tsx` around lines 1864 - 1868, The effect's dependency array incorrectly includes the ref accessor videoPlaybackRef.current?.video (in the useEffect block surrounding the Web Audio routing), which won't trigger re-runs when the underlying element changes; remove that ref accessor and instead depend on stable values or a proper signal: either use the ref object videoPlaybackRef (the ref itself) or add/consume a piece of state that updates when the video element is attached (e.g., setVideoElement on ref callback), and keep previewVersion and isAudioEngineReady as dependencies so the effect reruns reliably when the video element or playback readiness changes.src/utils/audioWaveform.ts (1)
44-48: Remove unused catch variable.The static analysis correctly flags that
eis unused. Use an empty catch or omit the binding.♻️ Proposed fix
} finally { try { await audioContext.close(); - } catch (e) { + } catch { // ignore close errors } }🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@src/utils/audioWaveform.ts` around lines 44 - 48, The catch block after calling audioContext.close() uses an unused binding `e`; change the catch to an empty binding by replacing `catch (e) { /* ignore close errors */ }` with `catch { /* ignore close errors */ }` (or remove the binding entirely) so the unused variable is eliminated while still suppressing close errors around the audioContext.close() call.src/lib/exporter/audioEncoder.ts (1)
466-472: Empty catch block and sync threshold comment.Line 468's empty catch block is flagged by static analysis. While fire-and-forget
play()is common, the empty block violates linting rules. The inline comment on line 470 is helpful.♻️ Proposed fix for linter compliance
if (audioEl.paused) { audioEl.currentTime = audioOffset - audioEl.play().catch(() => {}) + audioEl.play().catch(() => { /* fire-and-forget for export sync */ }) } else if (Math.abs(audioEl.currentTime - audioOffset) > 0.1) { // Tightened sync for export audioEl.currentTime = audioOffset }🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@src/lib/exporter/audioEncoder.ts` around lines 466 - 472, The empty catch after attempting audioEl.play() should be replaced with explicit handling to satisfy linting: in the block where you check audioEl.paused and call audioEl.play() (referencing audioEl and audioOffset), catch the promise rejection and handle it (e.g., log a debug/warn via existing logger or call console.debug/console.warn with the error and context) instead of swallowing it; keep the existing sync logic that sets audioEl.currentTime when Math.abs(audioEl.currentTime - audioOffset) > 0.1 so the tightened sync comment remains valid.electron/ipc/handlers.ts (1)
1541-1683: LGTM! Chunked caption generation correctly addresses past issues.The refactored implementation properly handles:
- Range-relative progress calculation (lines 1610-1612): Correctly computes progress using
rangeOffsetMs = offsetMs - startTimeMs.- Unique caption IDs (line 1641): Uses
caption-${offsetMs}-${idx}to prevent collisions across chunks.- Overlap deduplication (lines 1647-1651): Filters cues by start time to prevent duplicates while preserving all cues in the final chunk.
The chunking strategy (5-minute chunks with 10-second overlap) balances processing efficiency with word-boundary accuracy.
♻️ Optional: Add type annotation for better type safety
- const allCues: any[] = []; + const allCues: CaptionCuePayload[] = [];🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@electron/ipc/handlers.ts` around lines 1541 - 1683, The code in generateAutoCaptionsFromVideo is good, but add explicit TypeScript types for clarity: annotate the function signature (generateAutoCaptionsFromVideo) return type and the options parameter, type constants CHUNK_SIZE_MS and OVERLAP_MS as numbers, type allCues as Array<YourCueType> (define an interface for cue with id,startMs,endMs,text), and type updateChunkProgress callback (e.g., (progress: number) => void) so callers and editors get proper type checking and IDE help.electron/native/wgc-capture/src/monitor_utils.cpp (1)
30-36: Remove the orphaned UTF-8 helper until it has a call site.
wstringToStringis currently unused in this translation unit, so it just adds dead code. Either drop it for now or wire it into the logging path when you start printingdeviceName.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@electron/native/wgc-capture/src/monitor_utils.cpp` around lines 30 - 36, The function wstringToString is unused dead code; remove this orphaned helper from monitor_utils.cpp (remove the static std::string wstringToString(const std::wstring& wstr) definition) or, if you intend to keep it, wire it into the logging path by calling wstringToString when printing deviceName (e.g., convert deviceName before passing to the logger in the code that logs monitor/device names) so it is no longer unreferenced; ensure no other translation unit depends on this static helper before deleting.
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In `@electron/native/wgc-capture/src/monitor_utils.cpp`:
- Around line 53-79: findMonitorByBounds is comparing incoming Display.bounds
(DIPs from Electron) directly to MONITORINFOEX.rcMonitor (physical pixels),
causing mismatches on mixed-DPI setups; fix by converting the incoming
x,y,width,height from DIPs to physical pixels before comparisons (or convert
monitor rcMonitor to DIPs). In practice, update findMonitorByBounds (and/or
enumerateMonitors) to call GetDpiForMonitor (or GetDpiForWindow / DPI_AWARENESS
APIs) for each monitor handle, compute scale = dpi / 96.0, multiply the input
bounds (x,y,width,height) by that scale (or divide rcMonitor by scale) and then
perform the exact bounds and top-left comparisons; keep the MonitorFromRect
fallback as-is. Include the conversion logic adjacent to the loops that
reference enumerateMonitors, m.handle, and MONITORINFOEX.rcMonitor so
comparisons are done in the same coordinate space.
- Around line 72-74: Validate the incoming int parameters x, y, width, and
height before building the RECT used by MonitorFromRect: ensure width and height
are positive (reject or early-return/fallback if <= 0), compute right = x +
width and bottom = y + height using a wider type (e.g., 64-bit) or checked
arithmetic and clamp to INT_MAX/INT_MIN to avoid signed int overflow, then
verify right > left and bottom > top before constructing RECT and calling
MonitorFromRect; if the rectangle is invalid after these checks, handle it by
returning a safe default (e.g., null HMONITOR) or using an alternate
monitor-selection path.
In `@src/components/video-editor/VideoEditor.tsx`:
- Around line 391-394: The state initializer for autoCaptionSettings uses an
unsafe cast (initialEditorPreferences.whisperSelectedModel as any); change this
to use the proper typed model from AutoCaptionSettings (or a union of supported
model strings) and fall back to "small" if invalid. Specifically, update the
selectedModel assignment in the autoCaptionSettings initializer to cast/validate
against AutoCaptionSettings['selectedModel'] (or a defined WhisperModel union)
instead of using any, and ensure DEFAULT_AUTO_CAPTION_SETTINGS and
initialEditorPreferences.whisperSelectedModel are reconciled so selectedModel is
always a valid AutoCaptionSettings value.
- Around line 2045-2057: The AudioContext fallback uses an untyped any cast;
declare a typed Window extension (e.g., add a global interface Window {
webkitAudioContext?: typeof AudioContext } or replicate the same declaration
used in audioWaveform.ts) and remove the (window as any) cast. Then instantiate
with new (window.AudioContext || window.webkitAudioContext)() inside the
AudioContext init block (the try block that creates ctx in VideoEditor), so the
webkit fallback is typed and the any cast is eliminated.
---
Outside diff comments:
In `@src/components/video-editor/VideoEditor.tsx`:
- Around line 896-979: currentPersistedEditorState omits master audio fields so
hasUnsavedChanges won't detect changes; update the buildPersistedEditorState
invocation inside currentPersistedEditorState to include masterAudioMuted,
masterAudioSoloed, masterAudioVolume, audioTrackVolume, and isMasterSelected,
and add those same symbols to the useMemo dependency array so React recomputes
when those master-audio values change; reference buildPersistedEditorState and
currentPersistedEditorState to locate the call and dependency list to modify.
In `@src/lib/exporter/audioEncoder.ts`:
- Around line 493-515: In the finally cleanup block, also disconnect the
masterGainNode to avoid dangling connections: locate the cleanup that currently
disconnects sourceNode and destinationNode and add a safe check to call
masterGainNode.disconnect() (e.g., if (masterGainNode)
masterGainNode.disconnect()), optionally followed by nulling it; ensure this
runs before closing audioContext or stopping tracks so the graph is fully torn
down (refer to masterGainNode, sourceNode, destinationNode, and audioContext).
---
Duplicate comments:
In `@src/components/video-editor/AudioSettingsPanel.tsx`:
- Around line 36-44: The effect in AudioSettingsPanel uses
generateWaveform(audio.audioPath, 120).then(...) but doesn't handle rejections,
causing unhandled promise rejections; update the useEffect that references
generateWaveform, audio.audioPath, active, and setWaveform to handle errors by
attaching a .catch(...) to the promise (or switch to an async IIFE with
try/catch) and ensure errors are either logged (e.g., via console.error or a
logger) or silenced, while preserving the existing active guard to avoid stale
state updates.
In `@src/components/video-editor/VideoEditor.tsx`:
- Around line 1496-1514: When switching models the effect in VideoEditor that
calls window.electronAPI.getWhisperModelStatus (triggered by
autoCaptionSettings.selectedModel) can leave a stale whisperModelPath; update
the useEffect so it proactively clears whisperModelPath at the start of the
async check (or explicitly sets it to null when result.exists is false or
result.path is missing) by calling setWhisperModelPath(null) before/when
handling the response so the previous model path cannot persist and cause
transcription to use the wrong file.
In `@src/lib/exporter/annotationRenderer.ts`:
- Around line 333-345: The clip path currently uses the original annotation
bounds (x, y, width, height) which can be larger than the actually captured
region; update the clipping rectangle creation to use the clamped capture bounds
(srcX, srcY, srcWidth, srcHeight) instead of x/y/width/height so the clip
matches what will be drawn, and ensure to convert those source coords to
canvas/destination coordinates if needed using scaleFactor; also replace the
loose cast on offscreen (the drawImage call) by using a proper type (e.g., cast
offscreen to HTMLCanvasElement or OffscreenCanvas) rather than "as any". Ensure
you update the calls around ctx.roundRect/ctx.rect/ctx.clip and the
ctx.drawImage(offscreen ...) invocation in annotationRenderer.ts to use these
variables (srcX, srcY, srcWidth, srcHeight, scaleFactor, offscreen).
In `@src/utils/audioWaveform.ts`:
- Line 16: Replace the untyped immediate construction of the AudioContext so
creation can be safely feature-detected and failures caught: detect a
constructor reference (e.g. const AudioContextCtor = (window as
any).AudioContext || (window as any).webkitAudioContext), give it a proper typed
fallback instead of using an unchecked any where possible, check if
AudioContextCtor is truthy before attempting to instantiate, then wrap new
AudioContextCtor() in a try/catch and handle the failure (e.g. return null or a
safe fallback) so the symbol audioContext is only created when supported and
construction errors are caught; update any callers that expect audioContext
accordingly.
---
Nitpick comments:
In `@electron/ipc/handlers.ts`:
- Around line 1541-1683: The code in generateAutoCaptionsFromVideo is good, but
add explicit TypeScript types for clarity: annotate the function signature
(generateAutoCaptionsFromVideo) return type and the options parameter, type
constants CHUNK_SIZE_MS and OVERLAP_MS as numbers, type allCues as
Array<YourCueType> (define an interface for cue with id,startMs,endMs,text), and
type updateChunkProgress callback (e.g., (progress: number) => void) so callers
and editors get proper type checking and IDE help.
In `@electron/native/wgc-capture/src/monitor_utils.cpp`:
- Around line 30-36: The function wstringToString is unused dead code; remove
this orphaned helper from monitor_utils.cpp (remove the static std::string
wstringToString(const std::wstring& wstr) definition) or, if you intend to keep
it, wire it into the logging path by calling wstringToString when printing
deviceName (e.g., convert deviceName before passing to the logger in the code
that logs monitor/device names) so it is no longer unreferenced; ensure no other
translation unit depends on this static helper before deleting.
In `@src/components/video-editor/VideoEditor.tsx`:
- Around line 2722-2729: In the VideoEditor.tsx block where nodeEntry is
disconnected (the try/catch inside the logic that handles
audioRegionNodesRef.current.delete(id)), remove the unused catch variable by
changing the catch clause from catching a named parameter to using an empty
catch binding (i.e., replace "catch (e)" with "catch { }") so the thrown error
is ignored without declaring an unused identifier; update the catch for the try
that calls nodeEntry.source.disconnect() and nodeEntry.gain.disconnect().
- Around line 1864-1868: The effect's dependency array incorrectly includes the
ref accessor videoPlaybackRef.current?.video (in the useEffect block surrounding
the Web Audio routing), which won't trigger re-runs when the underlying element
changes; remove that ref accessor and instead depend on stable values or a
proper signal: either use the ref object videoPlaybackRef (the ref itself) or
add/consume a piece of state that updates when the video element is attached
(e.g., setVideoElement on ref callback), and keep previewVersion and
isAudioEngineReady as dependencies so the effect reruns reliably when the video
element or playback readiness changes.
In `@src/lib/exporter/audioEncoder.ts`:
- Around line 466-472: The empty catch after attempting audioEl.play() should be
replaced with explicit handling to satisfy linting: in the block where you check
audioEl.paused and call audioEl.play() (referencing audioEl and audioOffset),
catch the promise rejection and handle it (e.g., log a debug/warn via existing
logger or call console.debug/console.warn with the error and context) instead of
swallowing it; keep the existing sync logic that sets audioEl.currentTime when
Math.abs(audioEl.currentTime - audioOffset) > 0.1 so the tightened sync comment
remains valid.
In `@src/utils/audioWaveform.ts`:
- Around line 44-48: The catch block after calling audioContext.close() uses an
unused binding `e`; change the catch to an empty binding by replacing `catch (e)
{ /* ignore close errors */ }` with `catch { /* ignore close errors */ }` (or
remove the binding entirely) so the unused variable is eliminated while still
suppressing close errors around the audioContext.close() call.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Repository UI
Review profile: CHILL
Plan: Pro
Run ID: b280690e-d755-4e0e-af85-cca02819f1b1
📒 Files selected for processing (9)
electron/ipc/handlers.tselectron/native/wgc-capture/src/monitor_utils.cppsrc/components/video-editor/AudioSettingsPanel.tsxsrc/components/video-editor/VideoEditor.tsxsrc/components/video-editor/projectPersistence.tssrc/lib/exporter/annotationRenderer.tssrc/lib/exporter/audioEncoder.tssrc/lib/exporter/videoExporter.tssrc/utils/audioWaveform.ts
🚧 Files skipped from review as they are similar to previous changes (2)
- src/lib/exporter/videoExporter.ts
- src/components/video-editor/projectPersistence.ts
Comment on lines
+53
to
+79
| HMONITOR findMonitorByBounds(int x, int y, int width, int height) { | ||
| auto monitors = enumerateMonitors(); | ||
|
|
||
| // 1. Try exact bounds match | ||
| for (const auto& m : monitors) { | ||
| if (m.x == x && m.y == y && m.width == width && m.height == height) { | ||
| std::cerr << "Found monitor by exact bounds: " << x << "," << y << " " << width << "x" << height << std::endl; | ||
| return m.handle; | ||
| } | ||
| } | ||
|
|
||
| // 2. Try matching top-left point (sometimes size differs due to DPI scaling, but top-left is usually more stable in screen-space) | ||
| for (const auto& m : monitors) { | ||
| if (m.x == x && m.y == y) { | ||
| std::cerr << "Found monitor by top-left point match: " << x << "," << y << std::endl; | ||
| return m.handle; | ||
| } | ||
| } | ||
|
|
||
| // 3. Last resort: MonitorFromRect/Point (OS choice for the best matching monitor for these coordinates) | ||
| RECT rect = { x, y, x + width, y + height }; | ||
| HMONITOR hMonitor = MonitorFromRect(&rect, MONITOR_DEFAULTTONULL); | ||
| if (hMonitor) { | ||
| std::cerr << "Found monitor via Windows OS MonitorFromRect fallback" << std::endl; | ||
| return hMonitor; | ||
| } | ||
|
|
There was a problem hiding this comment.
🧩 Analysis chain
🏁 Script executed:
#!/bin/bash
# Trace where WGC display bounds are populated and whether a DIP->screen conversion is applied.
rg -nP --type=cpp --type=ts --type=tsx --type=js --type=jsx \
'\b(hasDisplayBounds|displayX|displayY|displayW|displayH)\b' -C3
rg -nP --type=ts --type=tsx --type=js --type=jsx \
'\b(dipToScreenRect|dipToScreenPoint|screen\.get(AllDisplays|PrimaryDisplay))\b' -C3Repository: webadderall/Recordly
Length of output: 119
🌐 Web query:
Electron screen API on Windows: are Display.boundscoordinates DIP-based or physical screen pixels, and what conversion should be used before comparing them to Win32MONITORINFOEX.rcMonitor values?
💡 Result:
Electron's Display.bounds coordinates are in DIP (device-independent pixels), not physical screen pixels. Win32 MONITORINFOEX.rcMonitor returns coordinates in physical pixels (virtual screen coordinates). To compare them, convert Electron DIP bounds to physical pixels using screen.dipToScreenRect(null, display.bounds) (pass null for window since no specific BrowserWindow is associated with the display query). This scales by the display's scaleFactor relative to the display containing the point/rect. Evidence from Electron docs: Display.bounds explicitly "the bounds of the display in DIP points" [4,5]; screen module distinguishes DIP vs physical, provides dipToScreenPoint/Rect for Windows [2,25]. GitHub PR documenting this [1]. Multi-monitor DPI examples show DIP bounds with fractional scaleFactors like 1.25 needing conversion [3,24]. Win32 docs: rcMonitor "display monitor rectangle, expressed in virtual-screen coordinates" which are physical pixels, unaffected by app DPI awareness for GetMonitorInfo [14]. Electron-Chromium per-monitor DPI aware, receives physical from Win32 APIs, exposes DIP to JS [6,22]. Direct comparison without conversion fails in scaled setups, as seen in GitHub issue #14703 where Win32 and Electron bounds differ by scale factor [10].
Citations:
- 1: docs: coordinate system of Display.bounds and Display.workArea electron/electron#27157
- 2: https://electronjs.org/docs/latest/api/screen
- 3: https://www.electronjs.org/docs/api/structures/display/
- 4: https://electronjs.org/docs/latest/api/structures/display
- 5: [Question] How to association Win32Api bounds between Electron screen display bounds? electron/electron#14703
- 6: https://learn.microsoft.com/en-us/windows/win32/api/winuser/ns-winuser-monitorinfo
- 7: https://www.electronjs.org/docs/api/screen
- 8: [Feature Request]: Add scale factor or original bounds in pixels to display object of screen.getAllDisplays() API. electron/electron#38950
- 9: Fix: Improve Per-Monitor DPI Scaling on Windows to Address Blurriness electron/electron#47322
- 10: https://stackoverflow.com/questions/8060280/getting-an-dpi-aware-correct-rect-from-getwindowrect-from-a-external-window/37829235
- 11: fix: window bounds not scaled from screen coordinates electron/electron#29162
- 12: incorrect coordinate calculation Screen method electron/electron#48492
- 13: https://stackoverflow.com/questions/14639332/windows-api-monitorinfo-structure
🏁 Script executed:
# Search for where displayX, displayY, displayW, displayH are defined/used in C++ files
fd -e cpp -e h . electron/native | xargs rg -n 'displayX|displayY|displayW|displayH' -B2 -A2Repository: webadderall/Recordly
Length of output: 2249
🏁 Script executed:
# Look at main.cpp to see how findMonitorByBounds is called
find . -name "main.cpp" -path "*/wgc-capture/*" | head -5 | xargs cat -n | head -150Repository: webadderall/Recordly
Length of output: 6004
🏁 Script executed:
# Search for any DIP conversion functions in the codebase
rg -n 'dipToScreen|DIPToScreen|dip.*screen|screen.*dip' electron/native --type cpp --type h -iRepository: webadderall/Recordly
Length of output: 46
Bounds comparison misses DIP-to-physical pixel conversion.
The display bounds (displayX, displayY, displayW, displayH) are extracted from JSON parameters (lines 135–145 in main.cpp) and passed directly to findMonitorByBounds() without DIP conversion. Since these originate from Electron's JavaScript layer—where Display.bounds are in DIPs—but are compared against MONITORINFOEX.rcMonitor (which uses physical/virtual screen coordinates), this function will fail in mixed-DPI setups.
The C++ code has no DIP conversion. Either apply screen.dipToScreenRect(null, bounds) on the Electron side before passing to the native module, or convert coordinates in C++ after extraction using the display's scale factor.
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In `@electron/native/wgc-capture/src/monitor_utils.cpp` around lines 53 - 79,
findMonitorByBounds is comparing incoming Display.bounds (DIPs from Electron)
directly to MONITORINFOEX.rcMonitor (physical pixels), causing mismatches on
mixed-DPI setups; fix by converting the incoming x,y,width,height from DIPs to
physical pixels before comparisons (or convert monitor rcMonitor to DIPs). In
practice, update findMonitorByBounds (and/or enumerateMonitors) to call
GetDpiForMonitor (or GetDpiForWindow / DPI_AWARENESS APIs) for each monitor
handle, compute scale = dpi / 96.0, multiply the input bounds (x,y,width,height)
by that scale (or divide rcMonitor by scale) and then perform the exact bounds
and top-left comparisons; keep the MonitorFromRect fallback as-is. Include the
conversion logic adjacent to the loops that reference enumerateMonitors,
m.handle, and MONITORINFOEX.rcMonitor so comparisons are done in the same
coordinate space.
Comment on lines
+72
to
+74
| // 3. Last resort: MonitorFromRect/Point (OS choice for the best matching monitor for these coordinates) | ||
| RECT rect = { x, y, x + width, y + height }; | ||
| HMONITOR hMonitor = MonitorFromRect(&rect, MONITOR_DEFAULTTONULL); |
There was a problem hiding this comment.
🧩 Analysis chain
🏁 Script executed:
# First, let's locate and examine the monitor_utils.cpp file
find . -name "monitor_utils.cpp" -type fRepository: webadderall/Recordly
Length of output: 171
🏁 Script executed:
# Check the full function definition around lines 72-74
cat -n electron/native/wgc-capture/src/monitor_utils.cpp | head -100Repository: webadderall/Recordly
Length of output: 4138
🏁 Script executed:
# Look for the function signature and understand parameter types
rg -A 20 "findMonitorByBounds" electron/native/wgc-capture/src/monitor_utils.cppRepository: webadderall/Recordly
Length of output: 1000
🏁 Script executed:
# Check what RECT is and what types it uses
rg -B 5 -A 5 "RECT rect" electron/native/wgc-capture/src/monitor_utils.cppRepository: webadderall/Recordly
Length of output: 480
🏁 Script executed:
# Check includes at the top of the file
head -20 electron/native/wgc-capture/src/monitor_utils.cppRepository: webadderall/Recordly
Length of output: 736
🏁 Script executed:
# Check the monitor_utils.h header file
cat -n electron/native/wgc-capture/src/monitor_utils.hRepository: webadderall/Recordly
Length of output: 633
🏁 Script executed:
# Get information about RECT and LONG types from Windows
web_search
Windows API RECT structure and LONG type definitionRepository: webadderall/Recordly
Length of output: 160
🌐 Web query:
Windows RECT structure LONG type definition overflow risk
💡 Result:
The Windows RECT structure defines a rectangle using four LONG fields: left, top, right, bottom. LONG is a signed 32-bit integer (-2^31 to 2^31-1, or -2147483648 to 2147483647). There is overflow risk in two main scenarios: 1. When computing width/height: width = right - left, height = bottom - top. If right - left exceeds 2^31-1 (2147483647), it overflows to a negative value, causing incorrect results like empty rectangles or failed area checks. Similar for height. This is a common issue in libraries like OpenCV where rect.area uses int(width * height), which can silently overflow. 2. When passing extreme coordinates to GDI functions (e.g., FillRect, StretchBlt), values near LONG limits (~±1e9) may cause undefined behavior, exceptions, or incorrect rendering due to internal calculations or device limits, even if within LONG range. Win32 GDI supports up to ~2^30 pixels practically, but exact hard limits are implementation-dependent and not officially documented beyond signed 32-bit range. Mitigations: Normalize rectangles (ensure left <= right, top <= bottom), validate width/height before use (e.g., check right - left > 0 && <= INT_MAX), use 64-bit intermediates for calculations, or clip to safe bounds like screen resolution multiples. No official Microsoft documentation warns explicitly of RECT-specific overflow beyond general signed integer limits, but rectangle functions assume valid non-empty rects (right > left, bottom > top).
Citations:
- 1: https://learn.microsoft.com/en-us/windows/win32/api/windef/ns-windef-rect
- 2: https://learn.microsoft.com/en-us/windows/win32/winprog/windows-data-types
- 3: https://learn.microsoft.com/en-us/windows/win32/gdi/rectangle-coordinates
- 4: Rect::area() and Size::area() may overflow too easily opencv/opencv#8093
- 5: https://stackoverflow.com/questions/3468495/what-are-the-hard-bounds-for-drawing-coordinates-in-gdi
- 6: https://stackoverflow.com/questions/79646622/why-does-win32-struct-size-have-signed-members
- 7: cv::rectangle with very large cv::Rect triggers undefined behavior sanitizer opencv/opencv#21037
Validate width and height before constructing the RECT.
The function receives int parameters directly from external input without validation. The arithmetic x + width and y + height can overflow as signed int operations, creating invalid rectangle coordinates. Additionally, zero or negative width/height values produce invalid rectangles (where right <= left or bottom <= top), which violate the assumptions of MonitorFromRect and lead to undefined behavior.
🛡️ Proposed fix
`#include` <ShellScalingApi.h>
+#include <limits>
`#include` <iostream>
@@
HMONITOR findMonitorByBounds(int x, int y, int width, int height) {
+ if (width <= 0 || height <= 0) {
+ return nullptr;
+ }
+
+ const auto right = static_cast<long long>(x) + static_cast<long long>(width);
+ const auto bottom = static_cast<long long>(y) + static_cast<long long>(height);
+ if (right > std::numeric_limits<LONG>::max() ||
+ bottom > std::numeric_limits<LONG>::max()) {
+ return nullptr;
+ }
+
auto monitors = enumerateMonitors();
@@
- RECT rect = { x, y, x + width, y + height };
+ RECT rect = { x, y, static_cast<LONG>(right), static_cast<LONG>(bottom) };🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In `@electron/native/wgc-capture/src/monitor_utils.cpp` around lines 72 - 74,
Validate the incoming int parameters x, y, width, and height before building the
RECT used by MonitorFromRect: ensure width and height are positive (reject or
early-return/fallback if <= 0), compute right = x + width and bottom = y +
height using a wider type (e.g., 64-bit) or checked arithmetic and clamp to
INT_MAX/INT_MIN to avoid signed int overflow, then verify right > left and
bottom > top before constructing RECT and calling MonitorFromRect; if the
rectangle is invalid after these checks, handle it by returning a safe default
(e.g., null HMONITOR) or using an alternate monitor-selection path.
Comment on lines
+391
to
+394
| const [autoCaptionSettings, setAutoCaptionSettings] = useState<AutoCaptionSettings>(() => ({ | ||
| ...DEFAULT_AUTO_CAPTION_SETTINGS, | ||
| selectedModel: (initialEditorPreferences.whisperSelectedModel as any) || "small", | ||
| })); |
There was a problem hiding this comment.
Type the whisper model selection properly.
The as any cast bypasses type checking for whisperSelectedModel. Use the appropriate type from AutoCaptionSettings or define a union type for supported models.
🛠️ Proposed fix
const [autoCaptionSettings, setAutoCaptionSettings] = useState<AutoCaptionSettings>(() => ({
...DEFAULT_AUTO_CAPTION_SETTINGS,
- selectedModel: (initialEditorPreferences.whisperSelectedModel as any) || "small",
+ selectedModel: (initialEditorPreferences.whisperSelectedModel as AutoCaptionSettings['selectedModel']) ?? "small",
}));🧰 Tools
🪛 Biome (2.4.9)
[error] 393-393: Unexpected any. Specify a different type.
(lint/suspicious/noExplicitAny)
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In `@src/components/video-editor/VideoEditor.tsx` around lines 391 - 394, The
state initializer for autoCaptionSettings uses an unsafe cast
(initialEditorPreferences.whisperSelectedModel as any); change this to use the
proper typed model from AutoCaptionSettings (or a union of supported model
strings) and fall back to "small" if invalid. Specifically, update the
selectedModel assignment in the autoCaptionSettings initializer to cast/validate
against AutoCaptionSettings['selectedModel'] (or a defined WhisperModel union)
instead of using any, and ensure DEFAULT_AUTO_CAPTION_SETTINGS and
initialEditorPreferences.whisperSelectedModel are reconciled so selectedModel is
always a valid AutoCaptionSettings value.
Comment on lines
+2045
to
+2057
|
|
||
| try { | ||
| const ctx = new (window.AudioContext || (window as any).webkitAudioContext)(); | ||
| const masterGain = ctx.createGain(); | ||
| masterGain.connect(ctx.destination); | ||
| audioContextRef.current = ctx; | ||
| masterGainRef.current = masterGain; | ||
| setIsAudioEngineReady(true); | ||
| console.log("[VideoEditor] Web Audio API context initialized"); | ||
| } catch (e) { | ||
| console.error("[VideoEditor] Failed to initialize AudioContext", e); | ||
| } | ||
| }, []); |
There was a problem hiding this comment.
Type the webkit AudioContext fallback.
Same pattern as audioWaveform.ts - the any cast can be replaced with a typed window extension.
🛠️ Proposed fix
try {
- const ctx = new (window.AudioContext || (window as any).webkitAudioContext)();
+ const AudioContextCtor =
+ window.AudioContext ??
+ (window as Window & { webkitAudioContext?: typeof AudioContext }).webkitAudioContext;
+ if (!AudioContextCtor) {
+ console.error("[VideoEditor] AudioContext not supported");
+ return;
+ }
+ const ctx = new AudioContextCtor();
const masterGain = ctx.createGain();📝 Committable suggestion
‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.
Suggested change
| try { | |
| const ctx = new (window.AudioContext || (window as any).webkitAudioContext)(); | |
| const masterGain = ctx.createGain(); | |
| masterGain.connect(ctx.destination); | |
| audioContextRef.current = ctx; | |
| masterGainRef.current = masterGain; | |
| setIsAudioEngineReady(true); | |
| console.log("[VideoEditor] Web Audio API context initialized"); | |
| } catch (e) { | |
| console.error("[VideoEditor] Failed to initialize AudioContext", e); | |
| } | |
| }, []); | |
| try { | |
| const AudioContextCtor = | |
| window.AudioContext ?? | |
| (window as Window & { webkitAudioContext?: typeof AudioContext }).webkitAudioContext; | |
| if (!AudioContextCtor) { | |
| console.error("[VideoEditor] AudioContext not supported"); | |
| return; | |
| } | |
| const ctx = new AudioContextCtor(); | |
| const masterGain = ctx.createGain(); | |
| masterGain.connect(ctx.destination); | |
| audioContextRef.current = ctx; | |
| masterGainRef.current = masterGain; | |
| setIsAudioEngineReady(true); | |
| console.log("[VideoEditor] Web Audio API context initialized"); | |
| } catch (e) { | |
| console.error("[VideoEditor] Failed to initialize AudioContext", e); | |
| } |
🧰 Tools
🪛 Biome (2.4.9)
[error] 2047-2047: Unexpected any. Specify a different type.
(lint/suspicious/noExplicitAny)
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In `@src/components/video-editor/VideoEditor.tsx` around lines 2045 - 2057, The
AudioContext fallback uses an untyped any cast; declare a typed Window extension
(e.g., add a global interface Window { webkitAudioContext?: typeof AudioContext
} or replicate the same declaration used in audioWaveform.ts) and remove the
(window as any) cast. Then instantiate with new (window.AudioContext ||
window.webkitAudioContext)() inside the AudioContext init block (the try block
that creates ctx in VideoEditor), so the webkit fallback is typed and the any
cast is eliminated.
Owner
|
Have you tested this on your Windows computer? Does it capture WGC reliably? |
Owner
|
Please resubmit without this change: 🛠️ 4. Advanced Interaction & Selection System Move vs. Select Modes: A dual-interaction system (V for move, E for select) that resolves UI conflicts between track manipulation and range selection." |
Contributor
Author
|
ya i will fix that
ya, have been build and install on my pc windows |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
🎥 1. New Video Editor Architecture
Implemented a robust video editing foundation designed for performance and extensibility:
Component-Driven Design: Refactored the core editing experience into isolated, high-performance components (VideoPlayback, TimelineEditor, Item, Row).
High-Performance Playback Engine: Built a dedicated event handling system (videoEventHandlers.ts) using requestAnimationFrame loops for frame-accurate timing without React state bottlenecks.
Multitrack Timeline: Support for multiple overlapping media types including Video, Audio, Zoom effects, Trims, Speed regions, and Annotations.
🪟 2. Native Windows Graphics Capture (WGC)
Implemented low-level monitor and capture utilities for Windows:
Low-Latency Monitor Capture: Leveraged native Windows Graphics Capture APIs for high-performance screencasting with minimal CPU overhead.
Monitor Discovery: Utilities to detect and select specific monitors for professional multi-screen recording setups.
🤖 3. AI-Powered Auto-Captioning (Whisper)
Integrated speech-to-text capabilities directly into the timeline:
Whisper Integration: Support for Open AI Whisper models (Tiny, Base, etc.) for local transcription.
Timeline Integration: Transcription cues are now displayed on a dedicated timeline track, allowing for frame-accurate caption editing and synchronization.
🛠️ 4. Advanced Interaction & Selection System
(Finalized in the recent commits)
Move vs. Select Modes: A dual-interaction system (V for move, E for select) that resolves UI conflicts between track manipulation and range selection.
Selection-Aware Playback: The "main bar" (playhead) now respects selection boundaries, automatically jumping to selection starts, looping at ends, and snapping cursors for focused editing.
Interactive Annotation Suite: Added "Blur" annotations for sensitive data protection and customizable text/image overlays.
🧩 5. Performance, Stability & Privacy
Stability Improvement: Finalized recording session persistence and privacy handling (PR #116 cleanup).
Export Optimization: Improved compositor resampling for higher quality final exports.
Type Safety: Full package type-checking integration to ensure codebase reliability.
📦 Commits Included:
1bb8eee: Architecture: Dedicated components for playback and timeline.
63b7205: Interactive media item components and timeline orchestration.
74dfa0d: Initial effects implementation (Audio waveforms, Zoom, Export).
6e910f0: Core VideoEditor component state management.
281a73b: Stability and privacy improvements for recording sessions.
e10270f: CI/CD: Full package type-check.
01350ad: UX: Refined Shift+drag selection and closure handling.
7e8a40f: UI: Dedicated Auto-Caption timeline track.
ced02c9: Native: Windows Graphics Capture monitor utilities.
7d7b61b: AI/Blur: Whisper model selection and Blur annotation effects.
Summary by CodeRabbit
New Features
Improvements