fix(gpu-direct): passthrough template title/subtitle/BGM config (#2093)

Co-authored-by: xiaoxia <dev@xiaoxiajianji.com> Co-committed-by: xiaoxia <dev@xiaoxiajianji.com>
2026-09-29 14:54:32 +08:00
parent f9f3e6bfb9
commit da59a6c9a6
3 changed files with 1012 additions and 39 deletions
@@ -2313,17 +2313,37 @@ class UnifiedRenderService:
            if bgm_path is not None and not bgm_path.exists():
                bgm_path = None

-            # 字幕：标题 + ASR 时间轴
+            # 字幕/标题/BGM 配置整包透传
            title_cfg = cfg.get("title", {}) or cfg.get("title_config", {}) or {}
+            if not isinstance(title_cfg, dict):
+                title_cfg = {}
            title_text = ""
-            if isinstance(title_cfg, dict) and title_cfg.get("enabled", True):
+            if title_cfg.get("enabled", True):
                title_text = title_cfg.get("text", "") or ""

-            subtitle_segments: list[Any] = []
            sub_cfg = cfg.get("subtitle", {}) or {}
-            if isinstance(sub_cfg, dict) and sub_cfg.get("enabled", True):
+            if not isinstance(sub_cfg, dict):
+                sub_cfg = {}
+            subtitle_segments: list[Any] = []
+            static_subtitle_text = ""
+            if sub_cfg.get("enabled", True):
                if sub_cfg.get("auto_generated") and self._asr_timeline_cache is not None:
                    subtitle_segments = list(self._asr_timeline_cache.segments)
+                else:
+                    # 静态字幕文本（用户手输）：pipeline 内部会构造全片长 segment
+                    static_subtitle_text = (sub_cfg.get("text", "") or "").strip()
+
+            bgm_cfg = cfg.get("bgm", {}) or {}
+            if not isinstance(bgm_cfg, dict):
+                bgm_cfg = {}
+            # 若 bgm.enabled 显式关闭，则强制 bgm_path=None（_prepare_bgm 已按 enabled 返回 None，双保险）
+            if not bgm_cfg.get("enabled", True):
+                bgm_path = None
+            # 注入微片段 BGM 偏移（同 CPU 路径）
+            if bgm_path is not None and not bgm_cfg.get("audio_offset"):
+                _micro_off = self._get_micro_bgm_offset()
+                if _micro_off:
+                    bgm_cfg = {**bgm_cfg, "audio_offset": _micro_off}

            # 边缘裁剪：dedup 开启时在 GPU 内做四边随机 2~5% 裁剪（gpu_direct_pipeline 内部随机）
            dedup = self._dedup_enabled()
@@ -2345,6 +2365,31 @@ class UnifiedRenderService:
                _vol = float((c.config or {}).get("volume", 1.0))
                clip_volumes_list.append(_vol if _vol > 0 else 0.0)

+            # extra_audio_tracks 音量：从 audio_tracks_config 读（TTS/配音素材库），
+            # 无法精确匹配 track_id 时保留默认 1.0
+            at_cfg = cfg.get("audio_tracks") or {}
+            tts_volume = 1.0
+            vo_volume = 1.0
+            if isinstance(at_cfg, dict):
+                _tracks = at_cfg.get("tracks", []) or []
+                for _t in _tracks:
+                    if not isinstance(_t, dict):
+                        continue
+                    try:
+                        _vol = float(_t.get("volume", 1.0))
+                    except (TypeError, ValueError):
+                        _vol = 1.0
+                    _tt = str(_t.get("track_type", ""))
+                    if _tt == "voiceover" and _t.get("audio_path"):
+                        vo_volume = max(0.0, min(2.0, _vol))
+                    # TTS 一般没有固定 track_type 标记，保持默认 1.0
+
+            extra_audio_tracks_cfg: list[tuple[Any, float]] = []
+            if tts_merged:
+                extra_audio_tracks_cfg.append((tts_merged, tts_volume))
+            if voiceover_track:
+                extra_audio_tracks_cfg.append((voiceover_track, vo_volume))
+
            plan = gdp.build_direct_render(
                resolved_clips=video_clips,
                output_width=self.output_width,
@@ -2357,7 +2402,11 @@ class UnifiedRenderService:
                total_duration=video_duration,
                clip_has_audio=clip_has_audio_list,
                clip_volumes=clip_volumes_list,
-                extra_audio_tracks=extra_audio_tracks,
+                extra_audio_tracks=extra_audio_tracks_cfg,
+                title_config=title_cfg,
+                subtitle_config=sub_cfg,
+                bgm_config=bgm_cfg,
+                static_subtitle_text=static_subtitle_text,
            )

            client = get_gpu_encoder()