From 1d27c509238ff5d81403d9b3df6a8007f886739c Mon Sep 17 00:00:00 2001
From: Yi-111-a <>
Date: Wed, 30 Sep 2026 00:43:53 +0800
Subject: [PATCH] fix(stories): keep a WebVTT cue whose identifier is a
 metadata word

`captionBlocks` dropped every WebVTT block whose first line was `STYLE`,
`REGION` or `NOTE`, so a cue whose *identifier* happened to be one of
those words lost its dialogue. The remaining cues still imported, which
made the omission easy to miss.

WebVTT's block parser gives a timing line in position two precedence over
the identifier, and `backend/services/srt_parser.py` already applies that
rule, so the same file parsed correctly for dubbing. Apply it here too.

Verified with the issue's reproduction: `importToText` on a file whose
first cue identifier is `intro`, `STYLE` or `REGION` now returns
`"Hello world.\nStill here."` for all three. Regression test fails before
and passes after; 28/28 pass in `importStory.test.js`.

The one frontend test that asserted the old divergent behaviour for
`NOTE` is replaced by the same parametrization the backend suite uses in
`test_webvtt_metadata_words_can_identify_a_cue`, so the two parsers can no
longer disagree on this case.

Fixes #2434
---
 CHANGELOG.md                                  |  1 +
 electron/src/shared/utils/importStory.js      |  6 +++++-
 electron/src/shared/utils/importStory.test.js | 19 ++++++++++++++-----
 3 files changed, 20 insertions(+), 6 deletions(-)

diff --git a/CHANGELOG.md b/CHANGELOG.md
index 1861e91c3..9c4f1d1dc 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -86,6 +86,7 @@ metadata and the backend fallback mirror it.
 - The Twilio guide and integration directory describe the guided setup and in-app integration pages (#2304)
 
 ### Fixed
+- Stories caption import keeps a WebVTT cue whose identifier is STYLE or REGION, matching the backend subtitle parser (#2434) — thanks @17lijunyi!
 - Dub assembly, cached segments and audio tools can read generated WAV files when TorchCodec is missing or cannot load (#2379) — thanks @tokutei58301-boop!
 - Remote API-key and share-PIN clients can record and read export history while native filesystem operations stay local (#2383, #2384) — thanks @sedatdagg!
 - Concurrent job events receive unique sequence numbers (#2384) — thanks @sedatdagg!
diff --git a/electron/src/shared/utils/importStory.js b/electron/src/shared/utils/importStory.js
index 2186d892a..e15a3f695 100644
--- a/electron/src/shared/utils/importStory.js
+++ b/electron/src/shared/utils/importStory.js
@@ -83,7 +83,11 @@ function captionBlocks(text, webvtt) {
     if (!lines.length) return false;
     const first = lines[0];
     if (/^WEBVTT(?:[ \t]|$)/.test(first)) return false;
-    return !isWebVttMetadata(first);
+    if (!isWebVttMetadata(first)) return true;
+    // WebVTT's block parser gives a timing line in position two precedence
+    // over the identifier, so `STYLE`/`REGION`/`NOTE` can name a real cue.
+    // https://www.w3.org/TR/webvtt1/#file-parsing — mirrors parse_srt.
+    return lines.length > 1 && isTimingLine(lines[1]);
   });
 }
 
diff --git a/electron/src/shared/utils/importStory.test.js b/electron/src/shared/utils/importStory.test.js
index b5d8c89e7..18a08bebe 100644
--- a/electron/src/shared/utils/importStory.test.js
+++ b/electron/src/shared/utils/importStory.test.js
@@ -109,11 +109,20 @@ describe('parseSrt', () => {
       '00:00:01.000 --> 00:00:02.000\nSpoken text\n';
     expect(parseSrt(vtt)).toBe('Spoken text');
   });
-  it('never treats a NOTE block as a cue even when it contains a valid timing line', () => {
-    const vtt =
-      'WEBVTT\n\nNOTE\n00:00:01.000 --> 00:00:02.000\nprivate note\n\n' +
-      '00:00:03.000 --> 00:00:04.000\nSpoken text\n';
-    expect(parseSrt(vtt)).toBe('Spoken text');
+  it('lets a timing line in position two outrank a metadata identifier', () => {
+    for (const identifier of ['STYLE', 'REGION', 'NOTE', 'NOTE identifier']) {
+      expect(parseSrt(`WEBVTT\n\n${identifier}\n00:01.000 --> 00:02.000\nSpoken text\n`)).toBe(
+        'Spoken text',
+      );
+    }
+  });
+  it('keeps a cue whose identifier is STYLE or REGION in a mixed file', () => {
+    for (const identifier of ['intro', 'STYLE', 'REGION']) {
+      const vtt =
+        `WEBVTT\n\n${identifier}\n00:00:01.000 --> 00:00:02.000\nHello world.\n\n` +
+        'next\n00:00:03.000 --> 00:00:04.000\nStill here.\n';
+      expect(parseSrt(vtt)).toBe('Hello world.\nStill here.');
+    }
   });
   it('keeps NOTE when it is the spoken dialogue', () => {
     expect(