From 0280c2e192dbf2755a614862291dcdf70166c594 Mon Sep 17 00:00:00 2001
From: zyc <1439655764@qq.com>
Date: Wed, 17 Jun 2026 10:24:28 +0800
Subject: [PATCH] =?UTF-8?q?fix(core):=20=E8=84=9A=E6=9C=AC=20agent=20?=
 =?UTF-8?q?=E5=86=8D=E5=85=BC=E5=AE=B9=20lines/caption=20=E5=8F=98?=
 =?UTF-8?q?=E4=BD=93,=E8=A1=A5=E5=85=A8=E6=97=81=E7=99=BD=E6=BC=8F?=
 =?UTF-8?q?=E7=A9=BA?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

豆包不同型号吐不同 schema:Lite 用 dialogue(字符串)+subtitle;Pro 用
lines:[{role,content}]+caption。上一版只认 dialogue/subtitle,导致 Pro 变体
画面能填、旁白仍空。normalize_draft 再扩容:
- 口播来源并入 lines(每项 content/line/text)、voiceover;subtitle/caption 兜底
- 画面来源并入 visual_description
新增 lines/caption 变体回归用例,全套 22 测试通过。

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
---
 core/backend/apps/ai/script_agent.py | 33 ++++++++++++++--------------
 core/backend/apps/ai/tests.py        | 20 +++++++++++++++++
 2 files changed, 37 insertions(+), 16 deletions(-)

diff --git a/core/backend/apps/ai/script_agent.py b/core/backend/apps/ai/script_agent.py
index 7f4161e..21b5b7d 100644
--- a/core/backend/apps/ai/script_agent.py
+++ b/core/backend/apps/ai/script_agent.py
@@ -288,28 +288,29 @@ def normalize_draft(raw_text: str, *, aspect_ratio: str, total_duration: int) ->
         speaker = speaker if (speaker in valid_ids) else None
         refs = [r for r in (seg.get("entity_refs") or []) if r in valid_ids]
         # 对白(剧情向):[{speaker(合法 entity id 或 null=旁白), line}];默认空 = 纯口播。
-        # 模型变体里 dialogue 可能是字符串(整句口播)而非数组——只在数组时按结构解析。
+        # 模型变体的口播字段五花八门:dialogue(字符串/数组)/ lines(数组)/ 每项 line|text|content。
+        # 这里把任意数组形态归一成结构化对白,字符串形态留给下面当整句旁白。
         raw_dialogue = seg.get("dialogue")
         dialogue = []
-        if isinstance(raw_dialogue, list):
-            for d in raw_dialogue:
-                if not isinstance(d, dict):
-                    continue
-                line = (d.get("line") or d.get("text") or "").strip()
-                if not line:
-                    continue
-                sp = d.get("speaker")
-                dialogue.append({"speaker": sp if sp in valid_ids else None, "line": line})
-        # 旁白:narration > 字符串 dialogue(整句口播)> subtitle(字幕)> 结构化对白拼接
-        narration = (seg.get("narration") or "").strip()
+        dialogue_array = raw_dialogue if isinstance(raw_dialogue, list) else (seg.get("lines") if isinstance(seg.get("lines"), list) else [])
+        for d in dialogue_array:
+            if not isinstance(d, dict):
+                continue
+            line = (d.get("line") or d.get("text") or d.get("content") or "").strip()
+            if not line:
+                continue
+            sp = d.get("speaker")
+            dialogue.append({"speaker": sp if sp in valid_ids else None, "line": line})
+        # 旁白:narration > 字符串 dialogue(整句口播)> 结构化对白/lines 拼接 > subtitle/caption(字幕兜底)
+        narration = (seg.get("narration") or seg.get("voiceover") or "").strip()
         if not narration and isinstance(raw_dialogue, str):
             narration = raw_dialogue.strip()
-        if not narration:
-            narration = (seg.get("subtitle") or "").strip()
         if not narration and dialogue:
             narration = " ".join(d["line"] for d in dialogue)  # 扁平拼接,兼容下游字幕/配音
-        # 画面:visual > visual_prompt > scene(模型变体常用 scene 描述画面)
-        visual = (seg.get("visual") or seg.get("visual_prompt") or seg.get("scene") or "").strip()
+        if not narration:
+            narration = (seg.get("subtitle") or seg.get("caption") or "").strip()
+        # 画面:visual > visual_prompt > scene > shot/visual_description(模型变体常用 scene 描述画面)
+        visual = (seg.get("visual") or seg.get("visual_prompt") or seg.get("scene") or seg.get("visual_description") or "").strip()
         norm_segments.append(
             {
                 "index": i,
diff --git a/core/backend/apps/ai/tests.py b/core/backend/apps/ai/tests.py
index 186f24b..c3ab3b9 100644
--- a/core/backend/apps/ai/tests.py
+++ b/core/backend/apps/ai/tests.py
@@ -31,6 +31,26 @@ class NormalizeDraftTests(SimpleTestCase):
         self.assertEqual(draft["segments"][0]["narration"], "卡裆太社死！")
         self.assertEqual(draft["segments"][0]["visual"], "更衣室扯卡裆旧裤")
 
+    def test_lines_and_caption_variant_fills_narration(self):
+        """另一种变体(Doubao-Seed-2.0-P):shots 用 lines:[{role,content}] 放口播、caption 放字幕。
+        必须把 lines 的 content 当对白/旁白,caption 兜底,而不是只填画面留旁白空。"""
+        raw = json.dumps({
+            "ScriptDraft": {
+                "basicInfo": {"totalDuration": 30, "aspectRatio": "9:16"},
+                "shots": [
+                    {"shotNo": 1, "scene": "化妆台两闺蜜", "lines": [{"role": "女主", "content": "防晒泛白太尴尬！"}, {"role": "闺蜜", "content": "试试这个！"}], "caption": "防晒踩雷？"},
+                    {"shotNo": 2, "scene": "手背挤膏体", "lines": [{"role": "闺蜜", "content": "质地清透不泛白"}], "caption": "清透质地"},
+                ],
+            }
+        }, ensure_ascii=False)
+        draft = normalize_draft(raw, aspect_ratio="9:16", total_duration=30)
+        self.assertEqual(len(draft["segments"]), 2)
+        for seg in draft["segments"]:
+            self.assertTrue(seg["narration"], "旁白不应为空")
+            self.assertTrue(seg["visual"], "画面不应为空")
+        self.assertIn("防晒泛白太尴尬", draft["segments"][0]["narration"])
+        self.assertEqual(len(draft["segments"][0]["dialogue"]), 2)  # lines→结构化对白
+
     def test_string_dialogue_not_iterated_as_chars(self):
         """dialogue 为整句字符串时,要当作旁白而不是逐字符遍历。"""
         raw = json.dumps({