feat(backend): take video thumbnails from the middle frame

Video thumbnails and previews were extracted ~1s in, which lands on
shared intros, title cards or black lead-in frames. Take the frame from
the middle (duration/2) instead — the same frame used for the perceptual
hash — so the thumbnail/preview reflects what dedup compared. Fold the
midpoint logic into a shared extractVideoFrameMiddle helper reused by
both the thumbnail/preview path and VideoFrameMiddle.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

backend/internal/storage/disk.go

@@ -173,12 +173,7 @@ func (s *DiskStorage) VideoFrameMiddle(ctx context.Context, id uuid.UUID) (image
 		}
 		return nil, fmt.Errorf("storage: stat %q: %w", srcPath, err)
 	}
-	// Fall back to a 1s offset if duration can't be probed — better a frame than none.
+	return extractVideoFrameMiddle(ctx, srcPath)
-	at := 1.0
-	if d, err := videoDurationSeconds(ctx, srcPath); err == nil && d > 0 {
-		at = d / 2
-	}
-	return extractVideoFrameAt(ctx, srcPath, at)
 }
 
 // ---------------------------------------------------------------------------
@@ -242,7 +237,7 @@ func (s *DiskStorage) serveGenerated(ctx context.Context, id uuid.UUID, cachePat
 	var img image.Image
 	if decoded, err := decodeImageLimited(srcPath, s.maxPixels); err == nil {
 		img = imaging.Fit(decoded, maxW, maxH, imaging.Lanczos)
-	} else if frame, err := extractVideoFrame(ctx, srcPath); err == nil {
+	} else if frame, err := extractVideoFrameMiddle(ctx, srcPath); err == nil {
 		img = imaging.Fit(frame, maxW, maxH, imaging.Lanczos)
 	} else {
 		img = placeholder(maxW, maxH)
@@ -368,10 +363,18 @@ func (s *DiskStorage) vipsThumbnail(ctx context.Context, srcPath, cachePath stri
 	return f, nil
 }
 
-// extractVideoFrame extracts a single frame ~1 second into the video — a safe
+// extractVideoFrameMiddle extracts a single frame from the middle of the video
-// default for thumbnails. See extractVideoFrameAt for the mechanics.
+// (duration/2), falling back to a 1s offset when the duration can't be probed.
-func extractVideoFrame(ctx context.Context, srcPath string) (image.Image, error) {
+// The midpoint dodges shared intros, title cards and black lead-in frames, and
-	return extractVideoFrameAt(ctx, srcPath, 1)
+// matches the frame used for the perceptual hash so a video's thumbnail/preview
+// shows the same representative frame dedup compared. See extractVideoFrameAt for
+// the mechanics.
+func extractVideoFrameMiddle(ctx context.Context, srcPath string) (image.Image, error) {
+	at := 1.0
+	if d, err := videoDurationSeconds(ctx, srcPath); err == nil && d > 0 {
+		at = d / 2
+	}
+	return extractVideoFrameAt(ctx, srcPath, at)
 }
 
 // extractVideoFrameAt uses ffmpeg to extract a single frame at atSec seconds into
@@ -403,7 +406,7 @@ func extractVideoFrameAt(ctx context.Context, srcPath string, atSec float64) (im
 }
 
 // videoDurationSeconds returns the container duration in seconds via ffprobe.
-// Used to seek to the middle of a clip for perceptual hashing.
+// Used to seek to the middle of a clip for perceptual hashing and thumbnails.
 func videoDurationSeconds(ctx context.Context, srcPath string) (float64, error) {
 	ctx, cancel := context.WithTimeout(ctx, 30*time.Second)
 	defer cancel()

frontend/src/routes/files/+page.svelte

@@ -94,15 +94,22 @@
 	// Select via the keyboard: a plain press toggles the focused card and drops the
 	// range anchor there; a Shift press selects everything from the anchor to the
 	// focused card — the same model as Shift+click on the grid.
+	// Select an inclusive index range in gesture direction (anchor → target) so the
+	// selection's insertion order follows how the user swept, not grid order. The
+	// Set preserves insertion order, so this is what later carries through to e.g.
+	// the order files land in a pool.
+	function selectRange(anchorIdx: number, targetIdx: number) {
+		const step = targetIdx >= anchorIdx ? 1 : -1;
+		for (let i = anchorIdx; i !== targetIdx + step; i += step) {
+			if (files[i]?.id) selectionStore.select(files[i].id!);
+		}
+	}
+
 	function selectFocused(range: boolean) {
 		const idx = focusedId ? files.findIndex((f) => f.id === focusedId) : -1;
 		if (idx < 0) return;
 		if (range && lastSelectedIdx !== null) {
-			const from = Math.min(lastSelectedIdx, idx);
+			selectRange(lastSelectedIdx, idx);
-			const to = Math.max(lastSelectedIdx, idx);
-			for (let i = from; i <= to; i++) {
-				if (files[i]?.id) selectionStore.select(files[i].id!);
-			}
 		} else if (files[idx]?.id) {
 			selectionStore.toggle(files[idx].id!);
 		}
@@ -634,12 +641,8 @@
 			return;
 		}
 		if (e.shiftKey && lastSelectedIdx !== null) {
-			// Range-select between lastSelectedIdx and idx (desktop)
+			// Range-select from the anchor toward idx (desktop), in gesture order.
-			const from = Math.min(lastSelectedIdx, idx);
+			selectRange(lastSelectedIdx, idx);
-			const to = Math.max(lastSelectedIdx, idx);
-			for (let i = from; i <= to; i++) {
-				if (files[i]?.id) selectionStore.select(files[i].id!);
-			}
 			lastSelectedIdx = idx;
 		} else {
 			if (file.id) selectionStore.toggle(file.id);