apps/api/app/services/video_compose_service.py

@@ -416,14 +416,14 @@ class VideoComposeService:
 
         filters: list[str] = []
 
-        # 1. scale: 等比缩放，保证覆盖目标区域（scale to larger, then crop）
-        filters.append(f"scale={output_width}:{output_height}" f":force_original_aspect_ratio=increase")
+        # 1. scale: 等比缩放（保持比例，不裁剪）
+        filters.append(f"scale={output_width}:{output_height}" f":force_original_aspect_ratio=decrease")
 
-        # 2. crop: 居中裁剪
-        filters.append(f"crop={output_width}:{output_height}")
+        # 2. pad: 居中+留黑边到目标分辨率（保持原始比例，不裁剪内容）
+        filters.append(f"pad={output_width}:{output_height}:(ow-iw)/2:(oh-ih)/2:black")
 
-        # 2.5 fps: 统一帧率（concat 要求所有输入帧率一致）
-        # 放在 crop 之后、setpts 之前，确保分辨率和帧率都已统一
+        # 3. fps: 统一帧率（concat 要求所有输入帧率一致）
+        # 放在 pad 之后、setpts 之前，确保分辨率和帧率都已统一
         if fps and fps > 0:
             filters.append(f"fps={fps}")
 
@@ -538,11 +538,17 @@ def _build_concat_filter(
     concat_filter = f"{concat_inputs}concat=n={n}:v=1:a=0[outv]"
     parts.append(concat_filter)
 
-    # 音频 concat（如果有）
+    # 音频 concat（如果有）— 先统一音频格式再拼接，否则不同采样率/声道会导致concat失败
     audio_parts: list[str] = []
     for idx, chain in enumerate(clip_chains):
         if chain.audio_label:
-            audio_parts.append(f"[{idx}:a]atrim=0:{chain.duration},asetpts=PTS-STARTPTS[{chain.audio_label}]")
+            # aformat: 统一采样率48000Hz + 双声道stereo + fltp采样格式（AAC标准格式）
+            audio_filters = [
+                "aformat=sample_rates=48000:channel_layouts=stereo:sample_fmts=fltp",
+                f"atrim=0:{chain.duration}",
+                "asetpts=PTS-STARTPTS",
+            ]
+            audio_parts.append(f"[{idx}:a]{','.join(audio_filters)}[{chain.audio_label}]")
 
     if audio_parts:
         parts.extend(audio_parts)

tests/unit/test_video_compose_service.py

@@ -344,8 +344,8 @@ class TestBuildComposeCommand(TestCase):
         crf_idx = cmd.command.index("-crf")
         self.assertEqual(cmd.command[crf_idx + 1], "28")
 
-    def test_filter_chain_contains_scale_and_crop(self):
-        """滤镜链包含 scale 和 crop。"""
+    def test_filter_chain_contains_scale_and_pad(self):
+        """滤镜链包含 scale 和 pad（保持比例留黑边，不裁剪）。"""
         plan = _StubPlan()
         clips = [_make_ready_clip(plan_id=plan.id)]
         svc = _make_service(plan, clips)
@@ -354,7 +354,9 @@ class TestBuildComposeCommand(TestCase):
         chain = cmd.clip_chains[0]
         filter_text = ",".join(chain.filters)
         self.assertIn("scale=", filter_text)
-        self.assertIn("crop=", filter_text)
+        self.assertIn("force_original_aspect_ratio=decrease", filter_text)
+        self.assertIn("pad=", filter_text)
+        self.assertIn("black", filter_text)
         self.assertIn("trim=", filter_text)
 
     def test_filter_chain_contains_fps(self):
@@ -366,11 +368,11 @@ class TestBuildComposeCommand(TestCase):
 
         chain = cmd.clip_chains[0]
         filter_text = ",".join(chain.filters)
-        # fps 必须在 crop 之后、setpts 之前
-        crop_idx = filter_text.index("crop=")
+        # fps 必须在 pad 之后、setpts 之前
+        pad_idx = filter_text.index("pad=")
         fps_idx = filter_text.index("fps=25")
         setpts_idx = filter_text.index("setpts=")
-        self.assertGreater(fps_idx, crop_idx, "fps 应该在 crop 之后")
+        self.assertGreater(fps_idx, pad_idx, "fps 应该在 pad 之后")
         self.assertLess(fps_idx, setpts_idx, "fps 应该在 setpts 之前")
 
     def test_fps_zero_or_none_skips_fps_filter(self):
@@ -533,6 +535,38 @@ class TestBuildConcatFilter(TestCase):
         self.assertIn("concat=n=2:v=1:a=0[outv]", filter_str)
         self.assertEqual(duration, 13.0)
 
+    def test_multi_clip_with_audio_aformat(self):
+        """多片段+音频时，每个音频流都经过 aformat 归一化再concat。"""
+        from app.services.video_compose_service import ClipFilterChain
+
+        chains = [
+            ClipFilterChain(
+                clip_id="c1",
+                input_index=0,
+                video_label="v0",
+                audio_label="a0",
+                filters=["scale=1280:720"],
+                duration=5.0,
+            ),
+            ClipFilterChain(
+                clip_id="c2",
+                input_index=1,
+                video_label="v1",
+                audio_label="a1",
+                filters=["scale=1280:720"],
+                duration=8.0,
+            ),
+        ]
+        filter_str, _ = _build_concat_filter(chains)
+        # 两个音频流都必须有 aformat 归一化（48000Hz + stereo + fltp）
+        self.assertIn("aformat=sample_rates=48000:channel_layouts=stereo:sample_fmts=fltp", filter_str)
+        # 音频 concat
+        self.assertIn("concat=n=2:v=0:a=1[outa]", filter_str)
+        # aformat 在 atrim 之前（先统一格式再裁剪）
+        aformat_idx = filter_str.index("aformat=")
+        atrim_idx = filter_str.index("atrim=")
+        self.assertLess(aformat_idx, atrim_idx, "aformat 应该在 atrim 之前")
+
 
 class TestBuildXfadeFilter(TestCase):
     """_build_xfade_filter 测试。"""