diff --git a/apps/worker/video_processing/pip_engine.py b/apps/worker/video_processing/pip_engine.py
new file mode 100755
index 000000000..cb84550e3
--- /dev/null
+++ b/apps/worker/video_processing/pip_engine.py
@@ -0,0 +1,483 @@
+"""画中画（PiP）引擎 — 基于 FFmpeg overlay 滤镜实现多图层叠加.
+
+支持能力：
+- 多图层叠加：主画面 + 多个副画面
+- 位置：9宫格 + 自由坐标（像素或百分比）
+- 大小：宽高缩放（像素或百分比）
+- 圆角裁剪：支持圆角矩形裁剪
+- 透明度：0-100%
+- 入场出场动画：淡入淡出、滑入滑出
+- 时间同步：每个副画面独立开始时间和持续时长
+- 降级策略：素材不存在时跳过，不阻断渲染
+"""
+
+from __future__ import annotations
+
+import logging
+from dataclasses import dataclass, field
+from pathlib import Path
+from typing import Any
+
+logger = logging.getLogger(__name__)
+
+
+# ── 位置常量 ──────────────────────────────────────────────────────────────────
+
+# 9宫格位置枚举
+POSITION_TOP_LEFT = "top_left"
+POSITION_TOP_CENTER = "top_center"
+POSITION_TOP_RIGHT = "top_right"
+POSITION_CENTER_LEFT = "center_left"
+POSITION_CENTER = "center"
+POSITION_CENTER_RIGHT = "center_right"
+POSITION_BOTTOM_LEFT = "bottom_left"
+POSITION_BOTTOM_CENTER = "bottom_center"
+POSITION_BOTTOM_RIGHT = "bottom_right"
+
+_VALID_POSITIONS = {
+    POSITION_TOP_LEFT,
+    POSITION_TOP_CENTER,
+    POSITION_TOP_RIGHT,
+    POSITION_CENTER_LEFT,
+    POSITION_CENTER,
+    POSITION_CENTER_RIGHT,
+    POSITION_BOTTOM_LEFT,
+    POSITION_BOTTOM_CENTER,
+    POSITION_BOTTOM_RIGHT,
+}
+
+# 动画类型
+ANIMATION_FADE = "fade"  # 淡入淡出
+ANIMATION_SLIDE_LEFT = "slide_left"  # 从左滑入
+ANIMATION_SLIDE_RIGHT = "slide_right"  # 从右滑入
+ANIMATION_SLIDE_TOP = "slide_top"  # 从上滑入
+ANIMATION_SLIDE_BOTTOM = "slide_bottom"  # 从下滑入
+
+_VALID_ANIMATIONS = {
+    ANIMATION_FADE,
+    ANIMATION_SLIDE_LEFT,
+    ANIMATION_SLIDE_RIGHT,
+    ANIMATION_SLIDE_TOP,
+    ANIMATION_SLIDE_BOTTOM,
+}
+
+
+# ── 数据模型 ──────────────────────────────────────────────────────────────────
+
+
+@dataclass
+class PiPLayerConfig:
+    """单个画中画图层配置."""
+
+    # 素材来源
+    source: str = ""  # 素材ID或视频URL
+    source_type: str = "asset_id"  # "asset_id" | "url" | "local_path"
+
+    # 位置配置
+    position: str = POSITION_BOTTOM_RIGHT  # 9宫格位置或 "custom"
+    x: int | str = 0  # 自定义x坐标（像素或百分比如 "30%"）
+    y: int | str = 0  # 自定义y坐标
+    margin: int = 20  # 9宫格模式下的边距（像素）
+
+    # 大小配置
+    width: int | str = "25%"  # 宽度（像素或百分比）
+    height: int | str = ""  # 高度（空则按比例自适应）
+
+    # 样式
+    opacity: float = 1.0  # 透明度 0.0-1.0
+    corner_radius: int = 0  # 圆角半径（像素），0表示无圆角
+    border_width: int = 0  # 边框宽度
+    border_color: str = "white"  # 边框颜色
+
+    # 时间控制
+    start_time: float = 0.0  # 开始显示时间（秒）
+    duration: float = 0.0  # 持续时长（秒），0表示全程显示
+
+    # 动画
+    animation_in: str = ""  # 入场动画类型
+    animation_out: str = ""  # 出场动画类型
+    animation_duration: float = 0.5  # 动画时长（秒）
+
+    # 层级
+    z_index: int = 1  # 图层顺序，数字越大越在上层
+
+    def validate(self) -> tuple[bool, str]:
+        """校验配置合法性，返回 (是否合法, 错误信息)."""
+        if not self.source:
+            return False, "source不能为空"
+
+        if self.position != "custom" and self.position not in _VALID_POSITIONS:
+            return False, f"无效的position: {self.position}"
+
+        if self.opacity < 0 or self.opacity > 1:
+            return False, "opacity必须在0-1之间"
+
+        if self.corner_radius < 0:
+            return False, "corner_radius不能为负数"
+
+        if self.start_time < 0:
+            return False, "start_time不能为负数"
+
+        if self.duration < 0:
+            return False, "duration不能为负数"
+
+        if self.animation_in and self.animation_in not in _VALID_ANIMATIONS:
+            return False, f"无效的入场动画: {self.animation_in}"
+
+        if self.animation_out and self.animation_out not in _VALID_ANIMATIONS:
+            return False, f"无效的出场动画: {self.animation_out}"
+
+        if self.animation_duration < 0:
+            return False, "animation_duration不能为负数"
+
+        return True, ""
+
+
+@dataclass
+class PiPConfig:
+    """画中画整体配置."""
+
+    enabled: bool = False
+    layers: list[PiPLayerConfig] = field(default_factory=list)
+
+    @classmethod
+    def from_dict(cls, data: dict[str, Any] | None) -> "PiPConfig":
+        """从字典解析配置."""
+        if not data or not data.get("enabled", False):
+            return cls(enabled=False)
+
+        layers_data = data.get("layers", [])
+        layers = []
+        for layer_data in layers_data:
+            try:
+                layer = PiPLayerConfig(
+                    source=layer_data.get("source", ""),
+                    source_type=layer_data.get("source_type", "asset_id"),
+                    position=layer_data.get("position", POSITION_BOTTOM_RIGHT),
+                    x=layer_data.get("x", 0),
+                    y=layer_data.get("y", 0),
+                    margin=int(layer_data.get("margin", 20)),
+                    width=layer_data.get("width", "25%"),
+                    height=layer_data.get("height", ""),
+                    opacity=float(layer_data.get("opacity", 1.0)),
+                    corner_radius=int(layer_data.get("corner_radius", 0)),
+                    border_width=int(layer_data.get("border_width", 0)),
+                    border_color=layer_data.get("border_color", "white"),
+                    start_time=float(layer_data.get("start_time", 0.0)),
+                    duration=float(layer_data.get("duration", 0.0)),
+                    animation_in=layer_data.get("animation_in", ""),
+                    animation_out=layer_data.get("animation_out", ""),
+                    animation_duration=float(layer_data.get("animation_duration", 0.5)),
+                    z_index=int(layer_data.get("z_index", 1)),
+                )
+                valid, err = layer.validate()
+                if valid:
+                    layers.append(layer)
+                else:
+                    logger.warning("PiP图层配置无效，跳过: %s", err)
+            except (ValueError, TypeError) as e:
+                logger.warning("PiP图层解析失败，跳过: %s", e)
+
+        # 按 z_index 排序
+        layers.sort(key=lambda layer: layer.z_index)
+
+        return cls(enabled=bool(layers), layers=layers)
+
+
+# ── PiP 引擎 ──────────────────────────────────────────────────────────────────
+
+
+class PiPEngine:
+    """画中画引擎 — 生成 FFmpeg 滤镜链实现多图层叠加."""
+
+    def __init__(
+        self,
+        output_width: int,
+        output_height: int,
+        output_fps: int = 30,
+    ):
+        self.output_width = output_width
+        self.output_height = output_height
+        self.output_fps = output_fps
+
+    def _parse_size(self, value: int | str, base: int) -> int:
+        """解析尺寸值（像素或百分比）."""
+        if isinstance(value, int):
+            return max(1, value)
+        if isinstance(value, str) and value.endswith("%"):
+            pct = float(value.rstrip("%")) / 100.0
+            return max(1, int(base * pct))
+        try:
+            return max(1, int(value))
+        except (ValueError, TypeError):
+            return int(base * 0.25)  # 默认25%
+
+    def _parse_position(
+        self,
+        layer: PiPLayerConfig,
+        pip_width: int,
+        pip_height: int,
+    ) -> tuple[int, int]:
+        """计算画中画的实际位置 (x, y)."""
+        W = self.output_width
+        H = self.output_height
+        m = layer.margin
+
+        if layer.position == "custom":
+            x = self._parse_size(layer.x, W)
+            y = self._parse_size(layer.y, H)
+            return (x, y)
+
+        pos_map = {
+            POSITION_TOP_LEFT: (m, m),
+            POSITION_TOP_CENTER: ((W - pip_width) // 2, m),
+            POSITION_TOP_RIGHT: (W - pip_width - m, m),
+            POSITION_CENTER_LEFT: (m, (H - pip_height) // 2),
+            POSITION_CENTER: ((W - pip_width) // 2, (H - pip_height) // 2),
+            POSITION_CENTER_RIGHT: (W - pip_width - m, (H - pip_height) // 2),
+            POSITION_BOTTOM_LEFT: (m, H - pip_height - m),
+            POSITION_BOTTOM_CENTER: ((W - pip_width) // 2, H - pip_height - m),
+            POSITION_BOTTOM_RIGHT: (W - pip_width - m, H - pip_height - m),
+        }
+        return pos_map.get(layer.position, pos_map[POSITION_BOTTOM_RIGHT])
+
+    def _build_pip_pre_filter(
+        self,
+        input_label: str,
+        layer: PiPLayerConfig,
+        pip_width: int,
+        pip_height: int,
+        output_label: str,
+    ) -> str:
+        """构建单个PiP图层的预处理滤镜链.
+
+        处理顺序：scale → 圆角裁剪（可选）→ 边框（可选）→ 透明度 → 动画（可选）
+        """
+        filters: list[str] = []
+
+        # Step 1: scale
+        filters.append(f"scale={pip_width}:{pip_height}")
+        filters.append("setsar=1")
+
+        # Step 2: 圆角裁剪
+        if layer.corner_radius > 0:
+            r = min(layer.corner_radius, pip_width // 2, pip_height // 2)
+            # 使用 geq + 圆形遮罩实现圆角
+            # 更简单的方式：用 rounded 滤镜（FFmpeg 5.0+）或 format + alpha
+            # 这里用更通用的方式：创建圆角遮罩 + overlay 到透明背景
+            filters.append(
+                f"format=yuva420p,"
+                f"geq="
+                f"lum='lum(X,Y)':"
+                f"cb='cb(X,Y)':"
+                f"cr='cr(X,Y)':"
+                f"a='if(lt(X,{r})*lt(Y,{r}),"
+                f"gt(hypot({r}-X,{r}-Y),{r})*0+1,"
+                f"if(gt(X,W-{r})*lt(Y,{r}),"
+                f"gt(hypot(X-(W-{r}),{r}-Y),{r})*0+1,"
+                f"if(lt(X,{r})*gt(Y,H-{r}),"
+                f"gt(hypot({r}-X,Y-(H-{r})),{r})*0+1,"
+                f"if(gt(X,W-{r})*gt(Y,H-{r}),"
+                f"gt(hypot(X-(W-{r}),Y-(H-{r})),{r})*0+1,1))))'"
+            )
+
+        # Step 3: 边框
+        if layer.border_width > 0:
+            bw = layer.border_width
+            color = layer.border_color
+            filters.append(f"pad={pip_width + 2*bw}:{pip_height + 2*bw}:{bw}:{bw}:{color}")
+
+        # Step 4: 透明度
+        if layer.opacity < 1.0:
+            alpha = layer.opacity
+            filters.append(f"format=yuva420p,colorchannelmixer=aa={alpha}")
+
+        # Step 5: 入场出场动画
+        if layer.animation_in or layer.animation_out:
+            filters.extend(self._build_animation_filters(layer, pip_width, pip_height))
+
+        filter_str = f"[{input_label}]{','.join(filters)}[{output_label}]"
+        return filter_str
+
+    def _build_animation_filters(
+        self,
+        layer: PiPLayerConfig,
+        pip_width: int,
+        pip_height: int,
+    ) -> list[str]:
+        """构建入场出场动画滤镜."""
+        filters: list[str] = []
+        anim_dur = layer.animation_duration
+
+        if layer.animation_in == ANIMATION_FADE:
+            # 淡入
+            filters.append(f"fade=t=in:st=0:d={anim_dur}:alpha=1")
+        elif layer.animation_in == ANIMATION_SLIDE_LEFT:
+            # 从左滑入 — 用 overlay 动态x实现，这里先标记位置表达式
+            pass  # slide 动画在 overlay 表达式中处理
+        elif layer.animation_in == ANIMATION_SLIDE_RIGHT:
+            pass
+        elif layer.animation_in == ANIMATION_SLIDE_TOP:
+            pass
+        elif layer.animation_in == ANIMATION_SLIDE_BOTTOM:
+            pass
+
+        if layer.animation_out == ANIMATION_FADE:
+            # 淡出需要知道总时长，这里用表达式
+            if layer.duration > 0:
+                start_fade = layer.duration - anim_dur
+                filters.append(f"fade=t=out:st={max(0, start_fade)}:d={anim_dur}:alpha=1")
+
+        return filters
+
+    def _build_overlay_expr(
+        self,
+        layer: PiPLayerConfig,
+        base_x: int,
+        base_y: int,
+        pip_width: int,
+        pip_height: int,
+    ) -> tuple[str, str]:
+        """构建 overlay 滤镜的 x/y 表达式（支持滑动动画）.
+
+        Returns:
+            (x_expr, y_expr) — FFmpeg表达式字符串
+        """
+        W = self.output_width
+        H = self.output_height
+        anim_dur = layer.animation_duration
+
+        x_expr = str(base_x)
+        y_expr = str(base_y)
+
+        # 入场滑入动画
+        if layer.animation_in == ANIMATION_SLIDE_LEFT:
+            # 从左侧滑入：x 从 -pip_width 变化到 base_x
+            x_expr = f"'{base_x}+(X)*0+if(lt(t,{anim_dur}),{-pip_width}+t/{anim_dur}*({base_x}+{pip_width}),{base_x})'"
+        elif layer.animation_in == ANIMATION_SLIDE_RIGHT:
+            # 从右侧滑入：x 从 W 变化到 base_x
+            x_expr = f"'{base_x}+if(lt(t,{anim_dur}),{W}-t/{anim_dur}*({W}-{base_x}),{base_x})'"
+        elif layer.animation_in == ANIMATION_SLIDE_TOP:
+            y_expr = f"'{base_y}+if(lt(t,{anim_dur}),{-pip_height}+t/{anim_dur}*({base_y}+{pip_height}),{base_y})'"
+        elif layer.animation_in == ANIMATION_SLIDE_BOTTOM:
+            y_expr = f"'{base_y}+if(lt(t,{anim_dur}),{H}-t/{anim_dur}*({H}-{base_y}),{base_y})'"
+
+        # 出场滑出动画（需要总时长）
+        if layer.duration > 0 and anim_dur > 0:
+            out_start = layer.duration - anim_dur
+            if layer.animation_out == ANIMATION_SLIDE_LEFT:
+                x_expr = f"'{base_x}+if(gt(t,{out_start}),{base_x}-(t-{out_start})/{anim_dur}*({base_x}+{pip_width}),{base_x})'"
+            elif layer.animation_out == ANIMATION_SLIDE_RIGHT:
+                x_expr = f"'{base_x}+if(gt(t,{out_start}),{base_x}+(t-{out_start})/{anim_dur}*({W}-{base_x}+{pip_width}),{base_x})'"
+            elif layer.animation_out == ANIMATION_SLIDE_TOP:
+                y_expr = f"'{base_y}+if(gt(t,{out_start}),{base_y}-(t-{out_start})/{anim_dur}*({base_y}+{pip_height}),{base_y})'"
+            elif layer.animation_out == ANIMATION_SLIDE_BOTTOM:
+                y_expr = f"'{base_y}+if(gt(t,{out_start}),{base_y}+(t-{out_start})/{anim_dur}*({H}-{base_y}+{pip_height}),{base_y})'"
+
+        return (x_expr, y_expr)
+
+    def build_pip_filters(
+        self,
+        base_label: str,
+        pip_sources: list[tuple[str, PiPLayerConfig, Path]],
+        *,
+        base_input_idx: int = 0,
+    ) -> tuple[str, list[str], str]:
+        """构建完整的画中画滤镜链和输入参数.
+
+        Args:
+            base_label: 底层视频的滤镜标签（如 "final_video" 或 "v0"，不带方括号）
+            pip_sources: [(input_label, layer_config, source_path), ...]
+            base_input_idx: PiP 素材在整个 FFmpeg 输入中的起始索引
+
+        Returns:
+            (filter_parts, input_args, final_label)
+            - filter_parts: 滤镜字符串列表（用 ; 连接后成为 filter_complex）
+            - input_args: 额外的输入参数列表 ["-i", path, "-i", path, ...]
+            - final_label: 最终合成后的输出标签（不带方括号）
+        """
+        if not pip_sources:
+            return [], [], base_label
+
+        filter_parts: list[str] = []
+        input_args: list[str] = []
+        current_label = base_label
+
+        for i, (input_label, layer, path) in enumerate(pip_sources):
+            # 添加输入
+            input_args.extend(["-i", str(path)])
+
+            # 计算实际大小
+            pip_w = self._parse_size(layer.width, self.output_width)
+            if layer.height:
+                pip_h = self._parse_size(layer.height, self.output_height)
+            else:
+                # 按宽度等比例（假设16:9，实际会scale时保持比例）
+                pip_h = int(pip_w * 9 / 16)
+
+            # 实际输入索引 = 起始索引 + 当前偏移
+            actual_input_idx = base_input_idx + i
+
+            # 预处理标签
+            pre_label = f"pip_pre_{i}"
+
+            # 构建预处理滤镜
+            pre_filter = self._build_pip_pre_filter(
+                input_label=f"{actual_input_idx}:v",
+                layer=layer,
+                pip_width=pip_w,
+                pip_height=pip_h,
+                output_label=pre_label,
+            )
+            filter_parts.append(pre_filter)
+
+            # 计算位置
+            base_x, base_y = self._parse_position(layer, pip_w, pip_h)
+
+            # 构建overlay表达式（支持滑动动画）
+            x_expr, y_expr = self._build_overlay_expr(layer, base_x, base_y, pip_w, pip_h)
+
+            # 时间控制（enable表达式）
+            enable_expr = ""
+            if layer.start_time > 0 or layer.duration > 0:
+                start = layer.start_time
+                if layer.duration > 0:
+                    end = start + layer.duration
+                    enable_expr = f":enable='between(t,{start},{end})'"
+                else:
+                    enable_expr = f":enable='gte(t,{start})'"
+
+            # 合成标签
+            combined_label = f"pip_combined_{i}"
+
+            # overlay 滤镜
+            overlay_filter = (
+                f"[{current_label}][{pre_label}]" f"overlay={x_expr}:{y_expr}{enable_expr}" f"[{combined_label}]"
+            )
+            filter_parts.append(overlay_filter)
+
+            current_label = combined_label
+
+        return filter_parts, input_args, current_label
+
+    def validate_layer_source(
+        self,
+        layer: PiPLayerConfig,
+        asset_path_map: dict[str, Path],
+    ) -> Path | None:
+        """验证图层素材是否可用，返回本地路径或None（降级跳过）."""
+        try:
+            if layer.source_type == "local_path":
+                path = Path(layer.source)
+                if path.exists():
+                    return path
+            elif layer.source_type == "asset_id":
+                if layer.source in asset_path_map:
+                    return asset_path_map[layer.source]
+            elif layer.source_type == "url":
+                # URL类型由调用者负责下载，这里返回标记
+                return None  # 暂时不支持直接URL
+        except Exception as e:
+            logger.warning("PiP素材验证失败: %s", e)
+
+        return None
diff --git a/apps/worker/video_processing/unified_render_service.py b/apps/worker/video_processing/unified_render_service.py
index 2a0575af7..631ba24f5 100755
--- a/apps/worker/video_processing/unified_render_service.py
+++ b/apps/worker/video_processing/unified_render_service.py
@@ -39,6 +39,7 @@ from video_processing.ffmpeg_utils import (
     probe_video_info,
     run_ffmpeg,
 )
+from video_processing.pip_engine import PiPConfig, PiPEngine, PiPLayerConfig
 from video_processing.render_audio import RenderContext, merge_audio_video, mix_audio
 from video_processing.render_subtitles import generate_ass_subtitles
 
@@ -203,15 +204,21 @@ class UnifiedRenderService:
         # 4. 生成 ASS 字幕文件（如果有 title/subtitle 配置）
         ass_path = self._maybe_generate_ass(video_duration)
 
+        # 4.5 解析画中画配置
+        pip_config = PiPConfig.from_dict((self.plan.config or {}).get("pip_config"))
+        pip_sources = self._resolve_pip_sources(pip_config) if pip_config.enabled else []
+        has_pip = len(pip_sources) > 0
+
         # 灰度埋点：开始渲染
         layer_roles = [layer.role for layer in layers]
         clip_counts = {layer.role: len(layer.clips) for layer in layers}
         logger.info(
-            "[unified-render] start render: plan_id=%s clip_count=%d layers=%s clip_counts=%s",
+            "[unified-render] start render: plan_id=%s clip_count=%d layers=%s clip_counts=%s pip_layers=%d",
             self.plan.id,
             len(resolved),
             layer_roles,
             clip_counts,
+            len(pip_sources),
         )
 
         # 5. 视频主渲染
@@ -219,7 +226,8 @@ class UnifiedRenderService:
         video_only_path = self.work_dir / f"rendered_{self.plan.id}_video.mp4"
         output_path = self.work_dir / f"rendered_{self.plan.id}.mp4"
 
-        is_pass_through = self._can_use_pass_through(layers)
+        # 有画中画时不走直通（需要额外图层叠加）
+        is_pass_through = self._can_use_pass_through(layers) and not has_pip
         pass_through_has_audio = False
         used_stream_copy = False
 
@@ -245,6 +253,11 @@ class UnifiedRenderService:
                 )
         else:
             filter_complex, input_args = self._build_filter_complex(layers, ass_path=ass_path)
+
+            # 追加画中画滤镜
+            if has_pip:
+                filter_complex, input_args = self._append_pip_filters(filter_complex, input_args, pip_sources)
+
             self._execute_ffmpeg(filter_complex, input_args, video_only_path)
 
         t_video_end = time.time()
@@ -984,3 +997,96 @@ class UnifiedRenderService:
         if clip.duration > 0:
             return min(clip.duration, clip.actual_duration) if clip.actual_duration > 0 else clip.duration
         return clip.actual_duration if clip.actual_duration > 0 else 0.0
+
+    # ── 画中画（PiP）相关方法 ──────────────────────────────────────────────────
+
+    def _resolve_pip_sources(self, pip_config: PiPConfig) -> list[tuple[str, PiPLayerConfig, Path]]:
+        """解析画中画图层的素材源，返回可用的图层列表.
+
+        降级策略：素材不存在或无效的图层自动跳过，不阻断渲染。
+
+        Returns:
+            [(input_label_placeholder, layer_config, local_path), ...]
+            input_label 在 build_pip_filters 中会用实际的输入索引替换
+        """
+        if not pip_config.enabled:
+            return []
+
+        engine = PiPEngine(
+            output_width=self.output_width,
+            output_height=self.output_height,
+            output_fps=self.output_fps,
+        )
+
+        result = []
+        for i, layer in enumerate(pip_config.layers):
+            path = engine.validate_layer_source(layer, self.asset_path_map)
+            if path is None:
+                logger.warning("PiP图层素材不可用，跳过: layer_index=%d source=%s", i, layer.source)
+                continue
+            # 标签占位，实际输入索引由 build_pip_filters 内部管理
+            result.append((f"pip_src_{i}", layer, path))
+
+        return result
+
+    def _append_pip_filters(
+        self,
+        filter_complex: str,
+        input_args: list[str],
+        pip_sources: list[tuple[str, Any, Path]],
+    ) -> tuple[str, list[str]]:
+        """将画中画滤镜追加到 filter_complex 末尾.
+
+        处理逻辑：
+        1. 将原 final_video 标签重命名为 pip_base（作为PiP的底层视频）
+        2. 追加 PiP 预处理和 overlay 滤镜
+        3. PiP 最终输出命名为 final_video
+
+        Args:
+            filter_complex: 原 filter_complex 字符串
+            input_args: 原输入参数列表
+            pip_sources: PiP 素材列表 [(label, layer_config, path), ...]
+
+        Returns:
+            (new_filter_complex, new_input_args)
+        """
+        if not pip_sources:
+            return filter_complex, input_args
+
+        pip_engine = PiPEngine(
+            output_width=self.output_width,
+            output_height=self.output_height,
+            output_fps=self.output_fps,
+        )
+
+        # 1. 将原 final_video 改为 pip_base
+        new_filter = filter_complex.replace("[final_video]", "[pip_base]")
+
+        # 2. 构建 PiP 滤镜链
+        # 主输入数量 = len(input_args) // 2（每个输入占 "-i path" 两个参数）
+        base_input_idx = len(input_args) // 2
+        pip_filter_parts, pip_input_args, final_label = pip_engine.build_pip_filters(
+            base_label="pip_base",
+            pip_sources=pip_sources,
+            base_input_idx=base_input_idx,
+        )
+
+        if not pip_filter_parts:
+            # 没有有效PiP滤镜，恢复原标签
+            return filter_complex, input_args
+
+        # 3. 追加 PiP 滤镜 + 最终格式转换（输出为 final_video）
+        pip_filter_str = ";".join(pip_filter_parts)
+        final_format = f"[{final_label}]format=yuv420p[final_video]"
+        new_filter = f"{new_filter};{pip_filter_str};{final_format}"
+
+        # 4. 追加输入参数
+        new_input_args = list(input_args) + pip_input_args
+
+        logger.info(
+            "[unified-render] appended PiP filters: layers=%d new_inputs=%d",
+            len(pip_sources),
+            len(pip_input_args) // 2,
+        )
+
+        return new_filter, new_input_args
diff --git a/tests/unit/test_pip_engine.py b/tests/unit/test_pip_engine.py
new file mode 100755
index 000000000..7a5e71cff
--- /dev/null
+++ b/tests/unit/test_pip_engine.py
@@ -0,0 +1,597 @@
+"""画中画（PiP）引擎单元测试."""
+
+from __future__ import annotations
+
+from pathlib import Path
+from unittest.mock import patch
+
+import pytest
+from video_processing.pip_engine import (
+    ANIMATION_FADE,
+    ANIMATION_SLIDE_BOTTOM,
+    ANIMATION_SLIDE_LEFT,
+    ANIMATION_SLIDE_RIGHT,
+    ANIMATION_SLIDE_TOP,
+    POSITION_BOTTOM_LEFT,
+    POSITION_BOTTOM_RIGHT,
+    POSITION_CENTER,
+    POSITION_TOP_LEFT,
+    POSITION_TOP_RIGHT,
+    PiPConfig,
+    PiPEngine,
+    PiPLayerConfig,
+)
+
+# ── PiPLayerConfig.validate 测试 ──────────────────────────────────────────────
+
+
+class TestPiPLayerConfigValidate:
+    """PiP图层配置校验测试."""
+
+    def test_valid_config(self):
+        """正常配置应该通过校验."""
+        layer = PiPLayerConfig(source="asset_001")
+        ok, err = layer.validate()
+        assert ok
+        assert err == ""
+
+    def test_empty_source(self):
+        """空source应该失败."""
+        layer = PiPLayerConfig(source="")
+        ok, err = layer.validate()
+        assert not ok
+        assert "source" in err
+
+    def test_invalid_position(self):
+        """无效位置应该失败."""
+        layer = PiPLayerConfig(source="asset_001", position="invalid_pos")
+        ok, err = layer.validate()
+        assert not ok
+        assert "position" in err
+
+    def test_custom_position_valid(self):
+        """custom位置应该通过."""
+        layer = PiPLayerConfig(source="asset_001", position="custom", x=100, y=50)
+        ok, err = layer.validate()
+        assert ok
+
+    def test_opacity_out_of_range_high(self):
+        """opacity超过1应该失败."""
+        layer = PiPLayerConfig(source="asset_001", opacity=1.5)
+        ok, err = layer.validate()
+        assert not ok
+        assert "opacity" in err
+
+    def test_opacity_out_of_range_low(self):
+        """opacity小于0应该失败."""
+        layer = PiPLayerConfig(source="asset_001", opacity=-0.5)
+        ok, err = layer.validate()
+        assert not ok
+        assert "opacity" in err
+
+    def test_opacity_boundary_values(self):
+        """opacity边界值应该通过."""
+        for val in [0.0, 0.5, 1.0]:
+            layer = PiPLayerConfig(source="asset_001", opacity=val)
+            ok, _ = layer.validate()
+            assert ok
+
+    def test_negative_corner_radius(self):
+        """负圆角应该失败."""
+        layer = PiPLayerConfig(source="asset_001", corner_radius=-5)
+        ok, err = layer.validate()
+        assert not ok
+        assert "corner_radius" in err
+
+    def test_negative_start_time(self):
+        """负开始时间应该失败."""
+        layer = PiPLayerConfig(source="asset_001", start_time=-1.0)
+        ok, err = layer.validate()
+        assert not ok
+        assert "start_time" in err
+
+    def test_negative_duration(self):
+        """负持续时间应该失败."""
+        layer = PiPLayerConfig(source="asset_001", duration=-5.0)
+        ok, err = layer.validate()
+        assert not ok
+        assert "duration" in err
+
+    def test_invalid_animation_in(self):
+        """无效入场动画应该失败."""
+        layer = PiPLayerConfig(source="asset_001", animation_in="spin")
+        ok, err = layer.validate()
+        assert not ok
+        assert "入场动画" in err
+
+    def test_all_valid_animations(self):
+        """所有有效动画类型应该通过."""
+        for anim in [
+            ANIMATION_FADE,
+            ANIMATION_SLIDE_LEFT,
+            ANIMATION_SLIDE_RIGHT,
+            ANIMATION_SLIDE_TOP,
+            ANIMATION_SLIDE_BOTTOM,
+        ]:
+            layer = PiPLayerConfig(source="asset_001", animation_in=anim, animation_out=anim)
+            ok, _ = layer.validate()
+            assert ok
+
+    def test_zero_duration_valid(self):
+        """duration=0（全程显示）应该通过."""
+        layer = PiPLayerConfig(source="asset_001", duration=0.0)
+        ok, _ = layer.validate()
+        assert ok
+
+
+# ── PiPConfig.from_dict 测试 ──────────────────────────────────────────────────
+
+
+class TestPiPConfigFromDict:
+    """PiP配置字典解析测试."""
+
+    def test_none_config(self):
+        """None配置应该返回disabled."""
+        config = PiPConfig.from_dict(None)
+        assert not config.enabled
+        assert len(config.layers) == 0
+
+    def test_empty_config(self):
+        """空字典应该返回disabled."""
+        config = PiPConfig.from_dict({})
+        assert not config.enabled
+
+    def test_enabled_false(self):
+        """enabled=False应该返回disabled."""
+        config = PiPConfig.from_dict({"enabled": False, "layers": [{"source": "a"}]})
+        assert not config.enabled
+
+    def test_single_layer(self):
+        """单图层解析."""
+        data = {
+            "enabled": True,
+            "layers": [
+                {
+                    "source": "asset_001",
+                    "position": POSITION_TOP_RIGHT,
+                    "width": "30%",
+                    "opacity": 0.9,
+                    "corner_radius": 10,
+                    "start_time": 2.0,
+                    "duration": 5.0,
+                    "z_index": 2,
+                }
+            ],
+        }
+        config = PiPConfig.from_dict(data)
+        assert config.enabled
+        assert len(config.layers) == 1
+        layer = config.layers[0]
+        assert layer.source == "asset_001"
+        assert layer.position == POSITION_TOP_RIGHT
+        assert layer.width == "30%"
+        assert layer.opacity == 0.9
+        assert layer.corner_radius == 10
+        assert layer.start_time == 2.0
+        assert layer.duration == 5.0
+        assert layer.z_index == 2
+
+    def test_multiple_layers_sorted_by_z_index(self):
+        """多图层应该按z_index排序."""
+        data = {
+            "enabled": True,
+            "layers": [
+                {"source": "asset_high", "z_index": 5},
+                {"source": "asset_low", "z_index": 1},
+                {"source": "asset_mid", "z_index": 3},
+            ],
+        }
+        config = PiPConfig.from_dict(data)
+        assert len(config.layers) == 3
+        assert config.layers[0].source == "asset_low"
+        assert config.layers[1].source == "asset_mid"
+        assert config.layers[2].source == "asset_high"
+
+    def test_invalid_layer_skipped(self):
+        """无效图层应该被跳过."""
+        data = {
+            "enabled": True,
+            "layers": [
+                {"source": "asset_good"},
+                {"source": "", "position": "invalid"},  # 空source
+                {"source": "asset_good2", "opacity": 2.0},  # opacity超范围
+            ],
+        }
+        config = PiPConfig.from_dict(data)
+        # 第1个有效，第2、3个无效
+        assert len(config.layers) == 1
+        assert config.layers[0].source == "asset_good"
+
+    def test_all_invalid_layers_disabled(self):
+        """所有图层都无效时enabled为False."""
+        data = {
+            "enabled": True,
+            "layers": [
+                {"source": ""},
+                {"source": ""},
+            ],
+        }
+        config = PiPConfig.from_dict(data)
+        assert not config.enabled
+        assert len(config.layers) == 0
+
+    def test_default_values(self):
+        """默认值应该正确."""
+        data = {
+            "enabled": True,
+            "layers": [{"source": "asset_001"}],
+        }
+        config = PiPConfig.from_dict(data)
+        layer = config.layers[0]
+        assert layer.position == POSITION_BOTTOM_RIGHT
+        assert layer.width == "25%"
+        assert layer.opacity == 1.0
+        assert layer.corner_radius == 0
+        assert layer.start_time == 0.0
+        assert layer.duration == 0.0
+        assert layer.z_index == 1
+
+
+# ── PiPEngine 位置计算测试 ────────────────────────────────────────────────────
+
+
+class TestPiPEnginePosition:
+    """PiP引擎位置计算测试."""
+
+    @pytest.fixture
+    def engine(self):
+        return PiPEngine(output_width=1920, output_height=1080, output_fps=30)
+
+    def test_top_left_position(self, engine):
+        """左上角位置."""
+        layer = PiPLayerConfig(source="a", position=POSITION_TOP_LEFT, margin=20)
+        x, y = engine._parse_position(layer, 480, 270)
+        assert x == 20
+        assert y == 20
+
+    def test_top_right_position(self, engine):
+        """右上角位置."""
+        layer = PiPLayerConfig(source="a", position=POSITION_TOP_RIGHT, margin=20)
+        x, y = engine._parse_position(layer, 480, 270)
+        assert x == 1920 - 480 - 20
+        assert y == 20
+
+    def test_bottom_right_position(self, engine):
+        """右下角位置（默认）."""
+        layer = PiPLayerConfig(source="a", position=POSITION_BOTTOM_RIGHT, margin=30)
+        x, y = engine._parse_position(layer, 480, 270)
+        assert x == 1920 - 480 - 30
+        assert y == 1080 - 270 - 30
+
+    def test_bottom_left_position(self, engine):
+        """左下角位置."""
+        layer = PiPLayerConfig(source="a", position=POSITION_BOTTOM_LEFT, margin=15)
+        x, y = engine._parse_position(layer, 480, 270)
+        assert x == 15
+        assert y == 1080 - 270 - 15
+
+    def test_center_position(self, engine):
+        """中心位置."""
+        layer = PiPLayerConfig(source="a", position=POSITION_CENTER, margin=0)
+        x, y = engine._parse_position(layer, 480, 270)
+        assert x == (1920 - 480) // 2
+        assert y == (1080 - 270) // 2
+
+    def test_custom_position_pixel(self, engine):
+        """自定义像素位置."""
+        layer = PiPLayerConfig(source="a", position="custom", x=100, y=200)
+        x, y = engine._parse_position(layer, 480, 270)
+        assert x == 100
+        assert y == 200
+
+    def test_custom_position_percentage(self, engine):
+        """自定义百分比位置."""
+        layer = PiPLayerConfig(source="a", position="custom", x="50%", y="25%")
+        x, y = engine._parse_position(layer, 480, 270)
+        assert x == 1920 // 2
+        assert y == 1080 // 4
+
+    def test_top_center_position(self, engine):
+        """顶部居中位置."""
+        layer = PiPLayerConfig(source="a", position="top_center", margin=10)
+        x, y = engine._parse_position(layer, 480, 270)
+        assert x == (1920 - 480) // 2
+        assert y == 10
+
+    def test_invalid_position_fallback(self, engine):
+        """无效位置应该fallback到右下角."""
+        layer = PiPLayerConfig(source="a", position="unknown_position", margin=20)
+        # 直接测试_parse_position（注意：validate会拦截，但_parse_position自己也有fallback）
+        x, y = engine._parse_position(layer, 480, 270)
+        assert x == 1920 - 480 - 20
+        assert y == 1080 - 270 - 20
+
+
+# ── PiPEngine 尺寸解析测试 ────────────────────────────────────────────────────
+
+
+class TestPiPEngineSize:
+    """PiP引擎尺寸解析测试."""
+
+    @pytest.fixture
+    def engine(self):
+        return PiPEngine(output_width=1920, output_height=1080, output_fps=30)
+
+    def test_pixel_size_int(self, engine):
+        """像素尺寸（整数）."""
+        assert engine._parse_size(500, 1920) == 500
+
+    def test_pixel_size_str(self, engine):
+        """像素尺寸（字符串数字）."""
+        assert engine._parse_size("500", 1920) == 500
+
+    def test_percentage_size(self, engine):
+        """百分比尺寸."""
+        assert engine._parse_size("50%", 1920) == 960
+        assert engine._parse_size("25%", 1920) == 480
+
+    def test_zero_size_default(self, engine):
+        """0或无效值应该有最小值保护."""
+        assert engine._parse_size(0, 1920) == 1
+        assert engine._parse_size("", 1920) == 480  # 默认25%
+
+    def test_negative_size_default(self, engine):
+        """负值应该取绝对值后至少为1."""
+        # _parse_size 用 max(1, value)，负值会走 except 分支
+        result = engine._parse_size("-100", 1920)
+        # 会走ValueError分支，返回默认值
+        assert result > 0
+
+
+# ── PiPEngine 滤镜构建测试 ────────────────────────────────────────────────────
+
+
+class TestPiPEngineBuildFilters:
+    """PiP引擎滤镜构建测试."""
+
+    @pytest.fixture
+    def engine(self):
+        return PiPEngine(output_width=1920, output_height=1080, output_fps=30)
+
+    @pytest.fixture
+    def fake_video(self, tmp_path):
+        """创建一个假的视频文件路径."""
+        path = tmp_path / "test_video.mp4"
+        path.write_bytes(b"fake video data")
+        return path
+
+    def test_empty_sources(self, engine):
+        """空素材列表应该返回空."""
+        filters, inputs, label = engine.build_pip_filters("base_label", [])
+        assert filters == []
+        assert inputs == []
+        assert label == "base_label"
+
+    def test_single_layer_basic(self, engine, fake_video):
+        """单图层基础滤镜构建."""
+        layer = PiPLayerConfig(
+            source="asset_001",
+            position=POSITION_TOP_RIGHT,
+            width="25%",
+        )
+        sources = [("pip_src_0", layer, fake_video)]
+
+        filters, inputs, final_label = engine.build_pip_filters("base_video", sources, base_input_idx=3)
+
+        # 应该有2个滤镜: 预处理 + overlay
+        assert len(filters) == 2
+        # 输入参数应该有2个（-i + path）
+        assert len(inputs) == 2
+        assert inputs[0] == "-i"
+        assert inputs[1] == str(fake_video)
+
+        # 预处理滤镜应该使用正确的输入索引
+        assert "3:v" in filters[0]
+        # 应该包含scale
+        assert "scale=" in filters[0]
+        # 应该有pip_pre_0标签
+        assert "[pip_pre_0]" in filters[0]
+
+        # overlay滤镜
+        assert "overlay=" in filters[1]
+        assert "[base_video][pip_pre_0]" in filters[1]
+
+    def test_single_layer_final_label(self, engine, fake_video):
+        """最终输出标签应该正确."""
+        layer = PiPLayerConfig(source="a", position=POSITION_CENTER)
+        sources = [("s0", layer, fake_video)]
+
+        _, _, final_label = engine.build_pip_filters("main_v", sources)
+        assert final_label == "pip_combined_0"
+
+    def test_multiple_layers(self, engine, fake_video):
+        """多图层叠加."""
+        layer1 = PiPLayerConfig(source="a", position=POSITION_TOP_LEFT, z_index=1)
+        layer2 = PiPLayerConfig(source="b", position=POSITION_BOTTOM_RIGHT, z_index=2)
+        sources = [
+            ("s0", layer1, fake_video),
+            ("s1", layer2, fake_video),
+        ]
+
+        filters, inputs, final_label = engine.build_pip_filters("base", sources, base_input_idx=0)
+
+        # 2层 × 2个滤镜（预处理+overlay）= 4个滤镜
+        assert len(filters) == 4
+        # 2个输入文件
+        assert len(inputs) == 4  # 2 × (-i + path)
+
+        # 输入索引应该连续
+        assert "0:v" in filters[0]
+        assert "1:v" in filters[2]
+
+        # 最终标签应该是第二个overlay的输出
+        assert final_label == "pip_combined_1"
+
+    def test_with_opacity(self, engine, fake_video):
+        """透明度应该在滤镜中体现."""
+        layer = PiPLayerConfig(source="a", position=POSITION_CENTER, opacity=0.5)
+        sources = [("s0", layer, fake_video)]
+
+        filters, _, _ = engine.build_pip_filters("base", sources)
+        pre_filter = filters[0]
+        assert "colorchannelmixer=aa=0.5" in pre_filter
+        assert "yuva420p" in pre_filter
+
+    def test_with_corner_radius(self, engine, fake_video):
+        """圆角裁剪应该在滤镜中体现."""
+        layer = PiPLayerConfig(source="a", position=POSITION_CENTER, corner_radius=20)
+        sources = [("s0", layer, fake_video)]
+
+        filters, _, _ = engine.build_pip_filters("base", sources)
+        pre_filter = filters[0]
+        assert "geq=" in pre_filter
+
+    def test_with_border(self, engine, fake_video):
+        """边框应该在滤镜中体现."""
+        layer = PiPLayerConfig(source="a", position=POSITION_CENTER, border_width=3, border_color="red")
+        sources = [("s0", layer, fake_video)]
+
+        filters, _, _ = engine.build_pip_filters("base", sources)
+        pre_filter = filters[0]
+        assert "pad=" in pre_filter
+        assert "red" in pre_filter
+
+    def test_timing_start_time_and_duration(self, engine, fake_video):
+        """时间控制应该生成enable表达式."""
+        layer = PiPLayerConfig(source="a", position=POSITION_CENTER, start_time=5.0, duration=10.0)
+        sources = [("s0", layer, fake_video)]
+
+        filters, _, _ = engine.build_pip_filters("base", sources)
+        overlay_filter = filters[1]
+        assert "enable=" in overlay_filter
+        assert "between(t,5.0,15.0)" in overlay_filter
+
+    def test_timing_start_time_only(self, engine, fake_video):
+        """只有开始时间（全程显示到结束）."""
+        layer = PiPLayerConfig(source="a", position=POSITION_CENTER, start_time=3.0, duration=0.0)
+        sources = [("s0", layer, fake_video)]
+
+        filters, _, _ = engine.build_pip_filters("base", sources)
+        overlay_filter = filters[1]
+        assert "enable=" in overlay_filter
+        assert "gte(t,3.0)" in overlay_filter
+
+    def test_no_timing_no_enable(self, engine, fake_video):
+        """无时间限制时不应该有enable表达式."""
+        layer = PiPLayerConfig(source="a", position=POSITION_CENTER, start_time=0.0, duration=0.0)
+        sources = [("s0", layer, fake_video)]
+
+        filters, _, _ = engine.build_pip_filters("base", sources)
+        overlay_filter = filters[1]
+        assert "enable=" not in overlay_filter
+
+    def test_fade_animation(self, engine, fake_video):
+        """淡入淡出动画."""
+        layer = PiPLayerConfig(
+            source="a",
+            position=POSITION_CENTER,
+            animation_in=ANIMATION_FADE,
+            animation_out=ANIMATION_FADE,
+            duration=10.0,
+            animation_duration=0.8,
+        )
+        sources = [("s0", layer, fake_video)]
+
+        filters, _, _ = engine.build_pip_filters("base", sources)
+        pre_filter = filters[0]
+        assert "fade=t=in" in pre_filter
+        assert "fade=t=out" in pre_filter
+        assert "alpha=1" in pre_filter
+
+    def test_slide_animation_in(self, engine, fake_video):
+        """滑入动画应该在overlay表达式中."""
+        layer = PiPLayerConfig(
+            source="a",
+            position=POSITION_CENTER,
+            animation_in=ANIMATION_SLIDE_LEFT,
+            animation_duration=0.5,
+        )
+        sources = [("s0", layer, fake_video)]
+
+        filters, _, _ = engine.build_pip_filters("base", sources)
+        overlay_filter = filters[1]
+        # x表达式应该包含动态变化
+        assert "overlay=" in overlay_filter
+
+    def test_full_opacity_no_alpha(self, engine, fake_video):
+        """opacity=1时不应该有colorchannelmixer."""
+        layer = PiPLayerConfig(source="a", position=POSITION_CENTER, opacity=1.0)
+        sources = [("s0", layer, fake_video)]
+
+        filters, _, _ = engine.build_pip_filters("base", sources)
+        pre_filter = filters[0]
+        assert "colorchannelmixer" not in pre_filter
+
+    def test_zero_corner_radius_no_geq(self, engine, fake_video):
+        """corner_radius=0时不应该有geq滤镜."""
+        layer = PiPLayerConfig(source="a", position=POSITION_CENTER, corner_radius=0)
+        sources = [("s0", layer, fake_video)]
+
+        filters, _, _ = engine.build_pip_filters("base", sources)
+        pre_filter = filters[0]
+        assert "geq=" not in pre_filter
+
+
+# ── PiPEngine 素材验证（降级策略）测试 ────────────────────────────────────────
+
+
+class TestPiPEngineValidateSource:
+    """PiP引擎素材验证与降级测试."""
+
+    @pytest.fixture
+    def engine(self):
+        return PiPEngine(output_width=1920, output_height=1080, output_fps=30)
+
+    def test_asset_id_in_map(self, engine, tmp_path):
+        """asset_id在map中应该返回路径."""
+        asset_path = tmp_path / "test.mp4"
+        asset_path.write_bytes(b"data")
+        asset_map = {"asset_001": asset_path}
+
+        layer = PiPLayerConfig(source="asset_001", source_type="asset_id")
+        result = engine.validate_layer_source(layer, asset_map)
+        assert result == asset_path
+
+    def test_asset_id_not_in_map(self, engine):
+        """asset_id不在map中应该返回None（降级）."""
+        layer = PiPLayerConfig(source="nonexistent", source_type="asset_id")
+        result = engine.validate_layer_source(layer, {})
+        assert result is None
+
+    def test_local_path_exists(self, engine, tmp_path):
+        """本地路径存在应该返回."""
+        path = tmp_path / "video.mp4"
+        path.write_bytes(b"data")
+
+        layer = PiPLayerConfig(source=str(path), source_type="local_path")
+        result = engine.validate_layer_source(layer, {})
+        assert result == path
+
+    def test_local_path_not_exists(self, engine):
+        """本地路径不存在应该返回None（降级）."""
+        layer = PiPLayerConfig(source="/nonexistent/path.mp4", source_type="local_path")
+        result = engine.validate_layer_source(layer, {})
+        assert result is None
+
+    def test_url_type_not_supported(self, engine):
+        """URL类型暂时不支持，返回None."""
+        layer = PiPLayerConfig(source="http://example.com/video.mp4", source_type="url")
+        result = engine.validate_layer_source(layer, {})
+        assert result is None
+
+    def test_exception_handling(self, engine):
+        """异常情况应该返回None（不阻断）."""
+        layer = PiPLayerConfig(source=None, source_type="local_path")  # type: ignore
+        # 模拟异常情况
+        result = engine.validate_layer_source(layer, {})
+        assert result is None