pipeline-ktv/pipeline_ktv/adapter.py

"""KTV产线适配器 — 注册step_types和handlers到pipeline-service。

本模块是pipeline-service的外部适配器，负责：
1. 注册KTV特有的step_types（含元数据）
2. 注册18个step handler函数（含asset_uploading）
3. 提供load_ktv_adapter()一键加载

宿主应用先调用 load_pipeline_service()，再调用 load_ktv_adapter()。
"""

import logging
from pipeline_service import register_handler, register_step_type

from .handlers import (
    handle_audio_preparing,
    handle_video_preparing,
    handle_demucs_separating,
    handle_lyric_calibrating,
    handle_subtitle_rendering,
    handle_subtitle_exporting,
    handle_lyric_generating,
    handle_lyric_evaluating,
    handle_music_generating,
    handle_music_polling,
    handle_character_designing,
    handle_character_image_generating,
    handle_asset_uploading,
    handle_storyboard_generating,
    handle_scene_video_generating,
    handle_scene_video_evaluating,
    handle_scene_video_concatenating,
    handle_ktv_synthesizing,
)

logger = logging.getLogger("pipeline_ktv.adapter")

MODULE_NAME = "pipeline_ktv"
MODULE_VERSION = "1.0.0"


# ── Step type metadata ──

KTV_STEP_TYPES = {
    "audio_preparing": {
        "display_name": "音频准备",
        "category": "media",
        "is_interactive": False,
        "description": "下载/复制音频文件，提取时长",
    },
    "video_preparing": {
        "display_name": "视频准备",
        "category": "media",
        "is_interactive": False,
        "description": "下载/复制视频，提取音频轨道",
    },
    "demucs_separating": {
        "display_name": "Demucs人声分离",
        "category": "media",
        "is_interactive": False,
        "description": "GPU运行Demucs分离人声和伴奏",
    },
    "lyric_calibrating": {
        "display_name": "歌词时间轴校准",
        "category": "llm",
        "is_interactive": False,
        "description": "ASR识别+LLM校准歌词时间戳",
    },
    "subtitle_rendering": {
        "display_name": "字幕渲染",
        "category": "media",
        "is_interactive": False,
        "description": "生成ASS卡拉OK字幕文件",
    },
    "subtitle_exporting": {
        "display_name": "字幕导出",
        "category": "media",
        "is_interactive": False,
        "description": "导出字幕为独立文件",
    },
    "lyric_generating": {
        "display_name": "歌词生成",
        "category": "llm",
        "is_interactive": False,
        "description": "LLM根据主题创作歌词",
    },
    "lyric_evaluating": {
        "display_name": "歌词评估",
        "category": "llm",
        "is_interactive": False,
        "description": "LLM评估歌词质量，低于阈值则拒绝",
    },
    "music_generating": {
        "display_name": "音乐生成",
        "category": "media",
        "is_interactive": False,
        "description": "提交Suno/MiniMax音乐生成任务",
    },
    "music_polling": {
        "display_name": "音乐轮询",
        "category": "media",
        "is_interactive": False,
        "description": "轮询音乐生成API直到完成",
    },
    "character_designing": {
        "display_name": "视觉素材设计",
        "category": "llm",
        "is_interactive": False,
        "description": "LLM设计角色(正面/左侧/右侧)、道具、服饰、场景图片prompt",
    },
    "character_image_generating": {
        "display_name": "素材图片生成",
        "category": "media",
        "is_interactive": False,
        "description": "通过llmage T2I API (wan2.7) 生成角色多视角、道具、服饰、场景图片",
    },
    "asset_uploading": {
        "display_name": "素材上传虚拟库",
        "category": "media",
        "is_interactive": False,
        "description": "将生成的素材图片上传到虚拟素材库，返回asset://素材号引用",
    },
    "storyboard_generating": {
        "display_name": "分镜脚本生成",
        "category": "llm",
        "is_interactive": False,
        "description": "LLM生成MV分镜脚本，引用asset://素材号，标记r2v/t2v模式",
    },
    "scene_video_generating": {
        "display_name": "场景视频生成",
        "category": "media",
        "is_interactive": False,
        "description": "通过Seedance 2.0生成场景视频，含角色场景用r2v模式+素材引用",
    },
    "scene_video_evaluating": {
        "display_name": "场景视频评估",
        "category": "media",
        "is_interactive": False,
        "description": "评估场景视频质量",
    },
    "scene_video_concatenating": {
        "display_name": "场景视频拼接",
        "category": "media",
        "is_interactive": False,
        "description": "ffmpeg拼接场景视频，循环匹配音频时长",
    },
    "ktv_synthesizing": {
        "display_name": "KTV最终合成",
        "category": "media",
        "is_interactive": False,
        "description": "合成KTV双轨+MTV单轨最终视频",
    },
}


# ── Handler mapping ──

KTV_HANDLERS = {
    "audio_preparing": handle_audio_preparing,
    "video_preparing": handle_video_preparing,
    "demucs_separating": handle_demucs_separating,
    "lyric_calibrating": handle_lyric_calibrating,
    "subtitle_rendering": handle_subtitle_rendering,
    "subtitle_exporting": handle_subtitle_exporting,
    "lyric_generating": handle_lyric_generating,
    "lyric_evaluating": handle_lyric_evaluating,
    "music_generating": handle_music_generating,
    "music_polling": handle_music_polling,
    "character_designing": handle_character_designing,
    "character_image_generating": handle_character_image_generating,
    "asset_uploading": handle_asset_uploading,
    "storyboard_generating": handle_storyboard_generating,
    "scene_video_generating": handle_scene_video_generating,
    "scene_video_evaluating": handle_scene_video_evaluating,
    "scene_video_concatenating": handle_scene_video_concatenating,
    "ktv_synthesizing": handle_ktv_synthesizing,
}


def register_ktv_step_types():
    """注册KTV步骤类型元数据到pipeline-service。"""
    for step_type, meta in KTV_STEP_TYPES.items():
        register_step_type(step_type, meta)
    logger.info(f"Registered {len(KTV_STEP_TYPES)} KTV step types")


def register_ktv_handlers():
    """注册KTV步骤处理器到pipeline-service。"""
    for step_type, fn in KTV_HANDLERS.items():
        register_handler(step_type, fn)
    logger.info(f"Registered {len(KTV_HANDLERS)} KTV handlers")


def load_ktv_adapter():
    """一键加载KTV适配器：注册step_types + handlers。

    调用前先确保 load_pipeline_service() 已执行。
    """
    register_ktv_step_types()
    register_ktv_handlers()
    logger.info(f"[{MODULE_NAME}] v{MODULE_VERSION} loaded — KTV pipeline adapter")
    return True