xiaoxia-saas/apps/api/app/services/plan_generator_service.py

"""PlanGeneratorService — 基于模板+素材自动生成剪辑计划.

核心职责:
- 根据 EditTemplate 的 editing_mode 和 TemplateClipConfig 列表，
  自动生成 EditPlan + EditPlanClip 列表
- 四种模式素材分配策略:
  - ONE_TAKE: 素材顺序分配给 main 类型 clips
  - PIP: 第1个素材→main(全屏背景)，其余→overlay clips
  - VOICE_OVER: 素材→main clips (B-roll)，标记需要配音叠加
  - VOICE_PIP: 第1个→background, 第2个→corner_voice, 其余→b_roll
"""

from __future__ import annotations

import logging
from typing import Any, List, Optional

from sqlalchemy.orm import Session

from packages.adapters.sqlalchemy_impl import (
    SQLAlchemyEditPlanClipRepository,
    SQLAlchemyEditPlanRepository,
)
from packages.domain.config_schemas import normalize_plan_config
from packages.domain.edit_plan import EditPlan
from packages.domain.edit_plan_clip import EditPlanClip
from packages.domain.edit_template import EditTemplate
from packages.domain.editing_mode import EditingMode
from packages.domain.template_clip_config import ClipType, TemplateClipConfig

logger = logging.getLogger(__name__)


# ── 默认片段时长（秒） ────────────────────────────────────────────────────────
_DEFAULT_CLIP_DURATION = 5.0
_DEFAULT_INTRO_DURATION = 3.0
_DEFAULT_OUTRO_DURATION = 3.0


class PlanGeneratorService:
    """剪辑计划生成器

    基于模板 + 素材，自动生成 EditPlan 及 EditPlanClip 列表。
    """

    def __init__(self, db: Session) -> None:
        self._plan_repo = SQLAlchemyEditPlanRepository(db)
        self._clip_repo = SQLAlchemyEditPlanClipRepository(db)

    # ── 公开接口 ─────────────────────────────────────────────────────────────

    def generate_from_template(
        self,
        template: EditTemplate,
        clip_configs: List[TemplateClipConfig],
        asset_ids: List[str],
        *,
        project_id: str = "",
        created_by_user_id: str = "",
        name: str = "",
    ) -> dict[str, Any]:
        """基于模板+素材生成剪辑计划

        Args:
            template: 剪辑模板实体
            clip_configs: 模板片段配置列表（可为空，自动生成默认结构）
            asset_ids: 素材 ID 列表
            project_id: 所属项目 ID
            created_by_user_id: 创建者用户 ID
            name: 计划名称（为空则自动取模板名）

        Returns:
            dict: {"plan": EditPlan, "clips": List[EditPlanClip]}
        """
        editing_mode = template.editing_mode or EditingMode.ONE_TAKE.value
        plan_name = name.strip() or f"{template.name} - 剪辑计划"

        # 1. 构建 plan config（继承模板的 title/subtitle/bgm，记录 editing_mode）
        plan_config = self._build_plan_config(template, editing_mode)

        # 2. 创建 EditPlan
        plan = EditPlan.create(
            template_id=template.id,
            name=plan_name,
            config=plan_config,
            total_duration=0.0,
            project_id=project_id,
            created_by_user_id=created_by_user_id,
        )
        plan = self._plan_repo.create(plan)
        logger.info(
            "生成剪辑计划: plan_id=%s template=%s mode=%s assets=%d",
            plan.id,
            template.id,
            editing_mode,
            len(asset_ids),
        )

        # 3. 生成片段列表
        if clip_configs:
            clips = self._create_clips_from_configs(plan.id, clip_configs)
        else:
            clips = self._generate_default_clips(plan.id, editing_mode, len(asset_ids))

        # 4. 按 editing_mode 分配素材
        if asset_ids:
            self._distribute_assets(clips, asset_ids, editing_mode)

        # 5. 持久化所有 clips 并计算总时长
        created_clips: List[EditPlanClip] = []
        total_duration = 0.0
        for clip in clips:
            saved = self._clip_repo.create(clip)
            created_clips.append(saved)
            total_duration += saved.duration

        # 6. 更新 plan 的 total_duration
        plan.total_duration = total_duration
        plan = self._plan_repo.update(plan)

        # 7. 流转到 editing 状态
        try:
            plan.start_editing()
            plan = self._plan_repo.update(plan)
        except ValueError as exc:
            logger.warning("计划状态流转失败: plan_id=%s error=%s", plan.id, exc)

        logger.info(
            "剪辑计划生成完成: plan_id=%s clips=%d duration=%.1f",
            plan.id,
            len(created_clips),
            total_duration,
        )

        return {"plan": plan, "clips": created_clips}

    # ── 内部方法 ─────────────────────────────────────────────────────────────

    def _build_plan_config(
        self,
        template: EditTemplate,
        editing_mode: str,
    ) -> dict[str, Any]:
        """从模板配置构建 plan config"""
        template_config = template.config or {}
        plan_config: dict[str, Any] = {
            "editing_mode": editing_mode,
        }
        # 继承模板的 cover/title/subtitle/bgm 配置
        for key in ("cover", "title", "subtitle", "bgm"):
            if key in template_config:
                plan_config[key] = template_config[key]

        return normalize_plan_config(plan_config)

    def _create_clips_from_configs(
        self,
        plan_id: str,
        clip_configs: List[TemplateClipConfig],
    ) -> List[EditPlanClip]:
        """从 TemplateClipConfig 列表创建 EditPlanClip 列表（未持久化）"""
        clips: List[EditPlanClip] = []
        # 按 order 排序
        sorted_configs = sorted(clip_configs, key=lambda c: c.order)

        for cfg in sorted_configs:
            # 计算时长：取 min_duration 和 max_duration 的中间值
            if cfg.min_duration > 0 and cfg.max_duration > 0:
                duration = (cfg.min_duration + cfg.max_duration) / 2
            elif cfg.min_duration > 0:
                duration = cfg.min_duration
            elif cfg.max_duration > 0:
                duration = cfg.max_duration
            else:
                duration = _DEFAULT_CLIP_DURATION

            # clip_type 可能是枚举或字符串
            clip_type = cfg.clip_type.value if hasattr(cfg.clip_type, "value") else cfg.clip_type

            # transition_effect 可能是枚举或字符串
            transition = (
                cfg.transition_effect.value if hasattr(cfg.transition_effect, "value") else cfg.transition_effect
            )

            clip = EditPlanClip.create(
                plan_id=plan_id,
                clip_type=clip_type,
                order=cfg.order,
                template_clip_config_id=cfg.id,
                text_content=getattr(cfg, "text_template", "") or "",
                duration=duration,
                transition_effect=transition or "cut",
            )
            clips.append(clip)

        return clips

    def _generate_default_clips(
        self,
        plan_id: str,
        editing_mode: str,
        asset_count: int,
    ) -> List[EditPlanClip]:
        """无 clip_configs 时，根据 editing_mode 生成默认 clip 结构

        - ONE_TAKE: N 个 main clips（N = asset_count，至少1个）
        - PIP: 1 个 main + (N-1) 个 overlay（N = asset_count）
        - VOICE_OVER: N 个 main clips + 标记需要配音
        - VOICE_PIP: 1 个 background + 1 个 corner_voice + (N-2) 个 b_roll
        """
        n = max(asset_count, 1)
        clips: List[EditPlanClip] = []
        order = 0

        if editing_mode == EditingMode.PIP.value:
            # 1 个 main（全屏背景）
            clips.append(
                EditPlanClip.create(
                    plan_id=plan_id,
                    clip_type=ClipType.MAIN.value,
                    order=order,
                    duration=_DEFAULT_CLIP_DURATION,
                )
            )
            order += 1
            # 剩余为 overlay
            for i in range(1, n):
                clips.append(
                    EditPlanClip.create(
                        plan_id=plan_id,
                        clip_type="overlay",
                        order=order,
                        duration=_DEFAULT_CLIP_DURATION,
                    )
                )
                order += 1

        elif editing_mode == EditingMode.VOICE_OVER.value:
            # N 个 main clips（B-roll）
            for i in range(n):
                clips.append(
                    EditPlanClip.create(
                        plan_id=plan_id,
                        clip_type=ClipType.MAIN.value,
                        order=order,
                        duration=_DEFAULT_CLIP_DURATION,
                        config={"role": "b_roll"},
                    )
                )
                order += 1

        elif editing_mode == EditingMode.VOICE_PIP.value:
            # 1 个 background
            clips.append(
                EditPlanClip.create(
                    plan_id=plan_id,
                    clip_type="background",
                    order=order,
                    duration=_DEFAULT_CLIP_DURATION,
                )
            )
            order += 1
            # 1 个 corner_voice
            clips.append(
                EditPlanClip.create(
                    plan_id=plan_id,
                    clip_type="corner_voice",
                    order=order,
                    duration=_DEFAULT_CLIP_DURATION,
                )
            )
            order += 1
            # 剩余为 b_roll
            for i in range(2, n):
                clips.append(
                    EditPlanClip.create(
                        plan_id=plan_id,
                        clip_type="b_roll",
                        order=order,
                        duration=_DEFAULT_CLIP_DURATION,
                    )
                )
                order += 1

        else:
            # ONE_TAKE: N 个 main clips
            for i in range(n):
                clips.append(
                    EditPlanClip.create(
                        plan_id=plan_id,
                        clip_type=ClipType.MAIN.value,
                        order=order,
                        duration=_DEFAULT_CLIP_DURATION,
                    )
                )
                order += 1

        return clips

    def _distribute_assets(
        self,
        clips: List[EditPlanClip],
        asset_ids: List[str],
        editing_mode: str,
    ) -> None:
        """按 editing_mode 将素材分配到 clips（就地修改，未持久化）

        分配策略:
        - ONE_TAKE: 素材按顺序依次分配给 main 类型 clips
        - PIP: 第1个素材→main(全屏背景)，其余→交替分配给 overlay clips
        - VOICE_OVER: 素材→main clips (B-roll)
        - VOICE_PIP: 第1个→background, 第2个→corner_voice, 其余→b_roll
        """
        if not asset_ids or not clips:
            return

        if editing_mode == EditingMode.ONE_TAKE.value:
            self._distribute_one_take(clips, asset_ids)
        elif editing_mode == EditingMode.PIP.value:
            self._distribute_pip(clips, asset_ids)
        elif editing_mode == EditingMode.VOICE_OVER.value:
            self._distribute_voice_over(clips, asset_ids)
        elif editing_mode == EditingMode.VOICE_PIP.value:
            self._distribute_voice_pip(clips, asset_ids)
        else:
            # 未知模式，退化为 one_take
            self._distribute_one_take(clips, asset_ids)

    def _distribute_one_take(
        self,
        clips: List[EditPlanClip],
        asset_ids: List[str],
    ) -> None:
        """ONE_TAKE: 素材按顺序依次分配给 main 类型 clips"""
        main_clips = [c for c in clips if c.clip_type == ClipType.MAIN.value]
        for i, clip in enumerate(main_clips):
            if i < len(asset_ids):
                clip.assign_asset(asset_ids[i])

    def _distribute_pip(
        self,
        clips: List[EditPlanClip],
        asset_ids: List[str],
    ) -> None:
        """PIP: 第1个素材→main(全屏背景)，其余→overlay clips"""
        # 第1个素材 → main clip
        main_clips = [c for c in clips if c.clip_type == ClipType.MAIN.value]
        if main_clips and asset_ids:
            main_clips[0].assign_asset(asset_ids[0])

        # 其余素材 → overlay clips
        overlay_clips = [c for c in clips if c.clip_type == "overlay"]
        remaining = asset_ids[1:]
        for i, clip in enumerate(overlay_clips):
            if i < len(remaining):
                clip.assign_asset(remaining[i])

    def _distribute_voice_over(
        self,
        clips: List[EditPlanClip],
        asset_ids: List[str],
    ) -> None:
        """VOICE_OVER: 素材→main clips (B-roll)"""
        main_clips = [c for c in clips if c.clip_type == ClipType.MAIN.value]
        for i, clip in enumerate(main_clips):
            if i < len(asset_ids):
                clip.assign_asset(asset_ids[i])

    def _distribute_voice_pip(
        self,
        clips: List[EditPlanClip],
        asset_ids: List[str],
    ) -> None:
        """VOICE_PIP: 第1个→background, 第2个→corner_voice, 其余→b_roll"""
        bg_clips = [c for c in clips if c.clip_type == "background"]
        corner_clips = [c for c in clips if c.clip_type == "corner_voice"]
        broll_clips = [c for c in clips if c.clip_type == "b_roll"]

        # 第1个素材 → background
        if bg_clips and len(asset_ids) > 0:
            bg_clips[0].assign_asset(asset_ids[0])

        # 第2个素材 → corner_voice
        if corner_clips and len(asset_ids) > 1:
            corner_clips[0].assign_asset(asset_ids[1])

        # 其余素材 → b_roll
        remaining = asset_ids[2:]
        for i, clip in enumerate(broll_clips):
            if i < len(remaining):
                clip.assign_asset(remaining[i])