version: '1.0'
updated: '2026-10-04'
categories:
  image:
    name: 图像
    icon: 🖼️
    order: 2
  visual:
    name: 视觉
    icon: 🎨
    order: 3
  video:
    name: 视频
    icon: 🎞️
    order: 4
  audio:
    name: 音频
    icon: 🔊
    order: 5
  analysis:
    name: 分析
    icon: 🧠
    order: 6
  external:
    name: 外部
    icon: 🧩
    order: 7
capabilities:
- id: tts_generate
  name: 语音合成
  category: audio
  description: 文本 → 语音音频（支持情绪/语速）
  input:
    text: string
    emotion: string?
  output:
    audio_path: string
    duration: number
  params_schema:
    emotion:
      type: enum
      options:
      - normal
      - happy
      - sad
      - angry
      - warm
      default: normal
      label: 情绪
    speed:
      type: number
      min: 0.5
      max: 2.0
      default: 1
      label: 语速
  providers:
  - id: volcano
    name: 火山语音
    api_key_ref: volcano.voice_app_key
    model: cosyvoice
    default: true
  - id: aliyun
    name: 阿里语音
    api_key_ref: aliyun.api_key
    model: cosyvoice-v1
    default: false
  defaults:
    provider: volcano
    emotion: normal
    speed: 1
  execution: local
  impl: voice_synthesizer/aliyun.py + voice_synthesizer/volcano.py
  status: partial
  note: 阿里百炼账号欠费（Arrearage）+ 火山缺 access_token → 充值/配置后可用
  source: 现有资产（voice_synthesizer/，391 行）
- id: bgm_generate
  name: BGM 生成
  category: audio
  description: 按风格/情绪生成背景音乐
  input:
    genre: string
    mood: string?
    duration: number
  output:
    audio_path: string
  params_schema:
    genre:
      type: enum
      options:
      - cinematic
      - lofi
      - upbeat
      - calm
      - inspirational
      default: cinematic
      label: 风格
    mood:
      type: string
      default: ''
      label: 情绪描述
  providers:
  - id: suno
    name: Suno
    api_key_ref: suno.api_key
    default: true
  - id: ace_step
    name: ACE-Step
    api_key_ref: ''
    model: ace-step
  defaults:
    provider: suno
    genre: cinematic
  execution: remote
  impl: bgm_generator/suno.py + bgm_generator/chord_pad.py
  status: ready
  source: 现有资产（bgm_generator/，654 行）
- id: music_select
  name: 配乐选择
  category: audio
  description: 从曲库按场景/情绪挑选 BGM（不生成）
  input:
    scene_type: string
    mood: string
  output:
    audio_path: string
  params_schema:
    max_duration:
      type: number
      default: 60
      min: 5
      max: 300
      label: 最长时长(秒)
  providers:
  - id: local_lib
    name: 本地曲库
    api_key_ref: ''
    default: true
  defaults:
    provider: local_lib
  execution: local
  impl: music_selector/
  status: ready
  source: 现有资产（music_selector/，371 行）
- id: beat_analyze
  name: 节拍分析
  category: audio
  description: 分析音频节拍/BPM → 拍点时间轴（卡点视频用）
  input:
    audio_path: string
  output:
    bpm: number
    beats: number[]
  params_schema:
    sensitivity:
      type: number
      min: 0.1
      max: 1.0
      default: 0.5
      label: 灵敏度
  providers:
  - id: local
    name: 本地分析
    api_key_ref: ''
    default: true
  defaults:
    provider: local
    sensitivity: 0.5
  execution: local
  impl: composer/beat_sync.py
  status: ready
  source: 现有资产（composer/beat_sync）
- id: emotion_analyze
  name: 情绪分析
  category: audio
  description: 分析音频情绪曲线（驱动剪辑节奏）
  input:
    audio_path: string
  output:
    emotion_curve: object[]
  params_schema: {}
  providers:
  - id: local
    name: 本地分析
    api_key_ref: ''
    default: true
  defaults:
    provider: local
  execution: local
  impl: story_director/（本地规则）
  status: ready
  source: 现有资产（audio_line/，3530 行）
- id: audio_mix
  name: 混音
  category: audio
  description: 人声 + BGM + 音效混音（音量/淡入淡出）
  input:
    voice: file
    bgm: file?
    sfx: file[]?
  output:
    audio_path: string
  params_schema:
    voice_volume:
      type: number
      min: 0
      max: 2
      default: 1.0
      label: 人声音量
    bgm_volume:
      type: number
      min: 0
      max: 1
      default: 0.15
      label: BGM 音量
    fade_in:
      type: number
      min: 0
      max: 5
      default: 1.0
      label: 淡入(秒)
    fade_out:
      type: number
      min: 0
      max: 5
      default: 1.0
      label: 淡出(秒)
  providers:
  - id: ffmpeg
    name: 本地 FFmpeg
    api_key_ref: ''
    default: true
  defaults:
    provider: ffmpeg
    voice_volume: 1.0
    bgm_volume: 0.15
    fade_in: 1.0
    fade_out: 1.0
  execution: local
  impl: 'composer/pipeline.py: mux_final'
  status: ready
  source: 现有资产（audio_line/）
- id: image_to_video
  name: 图生视频
  category: visual
  description: 参考图 + 提示词 → 视频片段（故事/短剧主力）
  input:
    image: file
    prompt: string
    duration: number
  output:
    video_path: string
  params_schema:
    duration:
      type: number
      min: 3
      max: 10
      default: 5
      label: 时长(秒)
    seed:
      type: number
      default: 42
      label: 随机种子
    camera_move:
      type: enum
      options:
      - none
      - pan
      - tilt
      - zoom
      - roll
      - orbit
      default: none
      label: 运镜
  providers:
  - id: kling
    name: 可灵
    api_key_ref: kling.api_key
    model: kling-v3
    default: true
  - id: jimeng
    name: 即梦
    api_key_ref: volcano.ark_api_key
    model: seedance
  defaults:
    provider: kling
    duration: 5
    seed: 42
    camera_move: none
  execution: remote
  impl: capabilities/kling_video.py
  status: ready
  source: 现有资产（batch_generator，664 行）；⚠️ 可灵余额不足，可切即梦
- id: person_swap
  name: 人物置换
  category: image
  description: 已有视频 → 替换人物（保持动作/场景）
  input:
    video: file
    character_image: file
  output:
    video_path: string
  params_schema:
    mode:
      type: enum
      options:
      - mix
      - animate
      default: mix
      label: 模式
  providers:
  - id: bailian
    name: 阿里百炼
    api_key_ref: aliyun.api_key
    model: wan2.2-animate-mix
    default: true
  defaults:
    provider: bailian
    mode: mix
  execution: remote
  impl: person_swap/api.py
  status: ready
  source: 2026-10-02 实测校准（真名 wan2.2-animate-mix；std ¥0.6/秒 · pro ¥0.9/秒）
- id: lipsync
  name: 对口型
  category: visual
  description: 人物视频 + 人声 → 口型匹配的新视频（仅"有台词"的镜头做）
  input:
    image: file
    audio: file
  output:
    video_path: string
  params_schema:
    resolution:
      type: enum
      options:
      - 720p
      - 1080p
      default: 1080p
      label: 分辨率
  providers:
  - id: bailian
    name: 阿里百炼 VideoRetalk
    api_key_ref: aliyun.api_key
    model: videoretalk
    default: true
    note: ¥0.08/秒 · 1800 秒免费额度 · 支持 ref_image_url
  - id: fal
    name: fal.ai（Kling Lipsync）
    api_key_ref: fal.api_key
    model: fal-ai/kling-video/lipsync/audio-to-video
    note: ⚠️ 缺 fal key，未启用
  - id: volcano
    name: 火山 OmniHuman
    api_key_ref: volcano.voice_app_key
  defaults:
    provider: bailian
    resolution: 1080p
  limits:
  - 视频需正面近景 · 2~120s · mp4/mov · 15~60fps · 边长 640~2048
  - 音频需干净人声（TTS 产物天然满足）
  execution: remote
  impl: composer/lipsync.py（lipsync_videoretalk）
  status: ready
  source: 2026-10-03 实测跑通（百炼 VideoRetalk · ¥0.08/秒 · 1800 秒免费额度）
- id: character_lock
  name: 角色锁定
  category: visual
  description: 跨镜头角色一致性（固定描述块 + seed）
  input:
    character_id: string
    shots: object[]
  output:
    locked_prompts: string[]
  params_schema:
    consistency_seed:
      type: number
      default: 42
      label: 一致性种子
  providers:
  - id: local
    name: 本地处理
    api_key_ref: ''
    default: true
  defaults:
    provider: local
    consistency_seed: 42
  execution: local
  impl: assets_manager/ + character_generator/
  status: ready
  source: 现有资产（character_locker，68 行）
- id: video_concat
  name: 视频拼接
  category: video
  description: 多片段拼接 + 转场
  input:
    clips: file[]
  output:
    video_path: string
  params_schema:
    transition:
      type: enum
      options:
      - cut
      - dissolve
      - fade
      - wipe
      default: cut
      label: 转场
    transition_duration:
      type: number
      min: 0
      max: 2
      default: 0.5
      label: 转场时长(秒)
    resolution:
      type: string
      default: 1080x1920
      label: 分辨率
    fps:
      type: number
      default: 30
      label: 帧率
  providers:
  - id: ffmpeg
    name: 本地 FFmpeg
    api_key_ref: ''
    default: true
  defaults:
    provider: ffmpeg
    transition: cut
    resolution: 1080x1920
    fps: 30
  execution: local
  impl: 'composer/pipeline.py: render_from_clips'
  status: ready
  source: 现有资产（composer/ffmpeg，702 行）
- id: speed_ramp
  name: 变速
  category: video
  description: 视频变速 / 速度斜坡（卡点用）
  input:
    video: file
  output:
    video_path: string
  params_schema:
    factor:
      type: number
      min: 0.25
      max: 4.0
      default: 1.0
      label: 速度倍率
    ramp:
      type: string
      default: ''
      label: 斜坡描述
  providers:
  - id: ffmpeg
    name: 本地 FFmpeg
    api_key_ref: ''
    default: true
  defaults:
    provider: ffmpeg
    factor: 1.0
  execution: local
  impl: composer/pipeline.py（_build_atempo_filter）
  status: ready
  source: 现有资产（speed_ramp，370 行）
- id: de_ai
  name: 去 AI 化
  category: video
  description: 后处理链：模糊 → 颗粒 → 调色 → S曲线 → 色温（消除 AI 感）
  input:
    video: file
  output:
    video_path: string
  params_schema:
    blur_sigma:
      type: number
      min: 0
      max: 2.0
      default: 0.4
      label: 模糊 sigma
    grain:
      type: number
      min: 0
      max: 50
      default: 12
      label: 颗粒强度
    saturation:
      type: number
      min: -1
      max: 1
      default: -0.12
      label: 饱和偏移
    temperature:
      type: number
      min: 2000
      max: 10000
      default: 5300
      label: 色温(K)
  providers:
  - id: ffmpeg
    name: 本地 FFmpeg
    api_key_ref: ''
    default: true
  defaults:
    provider: ffmpeg
    blur_sigma: 0.4
    grain: 12
    saturation: -0.12
    temperature: 5300
  execution: local
  impl: 'composer/pipeline.py: de_ai 处理'
  status: ready
  consumer: 合成节点（video_concat）自动调用
  source: 🆕 待建（S2）；参数来自 DCS 第八节
- id: subtitle
  name: 字幕生成
  category: video
  description: 台词文本 → 时间轴字幕（SRT/ASS）+ 烧录
  input:
    text: string
    audio_duration: number
  output:
    subtitle_path: string
    video_path: string?
  params_schema:
    style:
      type: enum
      options:
      - 手机系统默认
      - 简洁
      - 强调
      default: 手机系统默认
      label: 样式
    position:
      type: enum
      options:
      - bottom_center
      - bottom_left
      - top_center
      default: bottom_center
      label: 位置
  providers:
  - id: local
    name: 本地生成
    api_key_ref: ''
    default: true
  defaults:
    provider: local
    style: 手机系统默认
    position: bottom_center
  execution: local
  impl: composer/ffmpeg.py（ASS 字幕）
  status: ready
  consumer: 工作流 subtitle 节点
  source: 现有资产（部分实现）
- id: script_generate
  name: 脚本生成
  category: analysis
  description: 主题/素材 → 完整脚本（口播稿 / 短剧剧本）
  input:
    topic: string
    style: string
    duration: number
  output:
    script: string
    script_path: string
  params_schema:
    style:
      type: enum
      options:
      - 口播
      - 短剧
      - 科普
      - 剧情
      default: 口播
      label: 脚本类型
    duration:
      type: number
      min: 15
      max: 300
      default: 60
      label: 目标时长(秒)
    detail:
      type: enum
      options:
      - simple
      - detailed
      - stepwise
      default: detailed
      label: 详细度
  providers:
  - id: deepseek
    name: DeepSeek
    api_key_ref: llm.api_key
    model: deepseek-v4-flash
    default: true
  defaults:
    provider: deepseek
    style: 口播
    duration: 60
    detail: detailed
  execution: local
  impl: directors/llm.py（+ 各导演）
  status: ready
  source: 现有资产（script_generator/，377 行）
- id: script_parse
  name: 文案解析
  category: analysis
  description: 文案/故事 → 结构化导演脚本
  input:
    script: string
    style: string
  output:
    director_script: object
  params_schema:
    style:
      type: enum
      options:
      - knowledge_lecture
      - bedtime_story
      - funny_talk
      - personal_vlog
      default: knowledge_lecture
      label: 内容类型
  providers:
  - id: deepseek
    name: DeepSeek
    api_key_ref: llm.api_key
    model: deepseek-v4-flash
    default: true
  defaults:
    provider: deepseek
    style: knowledge_lecture
  execution: local
  impl: story_director/
  status: ready
  source: 现有资产（director_parser，323 行）
- id: storyboard_plan
  name: 分镜规划
  category: analysis
  description: 导演脚本/故事 → 分镜（景别/机位/运镜/时长/画面/台词）
  input:
    story: string
    character_block: string?
    outline: string?
  output:
    shots: object[]
  params_schema:
    shot_count:
      type: number
      min: 3
      max: 20
      default: 6
      label: 镜头数
    shot_duration:
      type: number
      min: 3
      max: 15
      default: 5
      label: 单镜头时长(秒)
    style:
      type: string
      default: documentary_realism
      label: 拍摄风格
  providers:
  - id: storyboard_v52
    name: AI Storyboard Director v5.2
    api_key_ref: llm.api_key
    model: deepseek-v4-flash
    default: true
  - id: clip_storyboard
    name: clip-storyboard-director
    api_key_ref: llm.api_key
    model: deepseek-v4-flash
  - id: legacy_scene_planner
    name: 本地场景规划（legacy）
    api_key_ref: llm.api_key
    model: deepseek-v4-flash
  defaults:
    provider: legacy_scene_planner
    shot_count: 6
    shot_duration: 5
    style: documentary_realism
  execution: local
  impl: directors/（storyboard_v52 / seedance_director / legacy）
  status: ready
  source: 现有资产（scene_planner 444 行）；v5.2 待接（S2）
- id: prompt_gen
  name: 提示词生成
  category: analysis
  description: 分镜 + 角色 → 每镜头视频生成提示词
  input:
    shots: object[]
    character_block: string
  output:
    prompts: string[]
  params_schema:
    detail:
      type: enum
      options:
      - simple
      - detailed
      - stepwise
      default: detailed
      label: 详细度
  providers:
  - id: deepseek
    name: DeepSeek
    api_key_ref: llm.api_key
    model: deepseek-v4-flash
    default: true
  defaults:
    provider: deepseek
    detail: detailed
  execution: local
  impl: 'composer/pipeline.py: build_shot_prompt'
  status: ready
  source: 现有资产
- id: anchor_extract
  name: 锚点提取
  category: analysis
  description: 脚本 → 一致性锚点（角色/场景/道具/风格）
  input:
    script: object
  output:
    anchors: object
  params_schema: {}
  providers:
  - id: local
    name: 本地处理
    api_key_ref: ''
    default: true
  defaults:
    provider: local
  execution: local
  impl: story_director/
  status: ready
  source: 现有资产（anchor_extractor，155 行）
- id: material_search
  name: 素材搜索
  category: analysis
  description: 关键词 → 素材片段（背景/空镜）
  input:
    search: string
    count: number
    orientation: string
  output:
    clips: file[]
  params_schema:
    count:
      type: number
      min: 1
      max: 20
      default: 3
      label: 数量
    orientation:
      type: enum
      options:
      - portrait
      - landscape
      default: portrait
      label: 画幅方向
  providers:
  - id: pexels
    name: Pexels
    api_key_ref: pexels.api_key
    default: true
  - id: wan2_2
    name: 通义万相
    api_key_ref: aliyun.api_key
    model: wan2.2
  - id: fallback
    name: 本地兜底素材
    api_key_ref: ''
  defaults:
    provider: pexels
    count: 3
    orientation: portrait
  execution: remote
  impl: material_producer/{pexels,kling,wan2_2,fallback}/
  status: ready
  source: 现有资产（material_producer，902 行）
- id: comfyui_workflow
  name: ComfyUI 工作流
  category: external
  description: 调用 RunningHub 上的 ComfyUI 工作流（**一个工作流 = 一个新能力**，可长出面部件表情/风格渲染等）
  input:
    workflow_id: string
    images: file[]?
    text: string?
  output:
    output_files: file[]
  params_schema:
    workflow_id:
      type: string
      default: ''
      label: 工作流 ID
    node_overrides:
      type: object
      default: {}
      label: 节点覆盖参数
  providers:
  - id: runninghub
    name: RunningHub
    api_key_ref: runninghub.api_key
    default: true
  - id: comfydeploy
    name: ComfyDeploy
    api_key_ref: comfydeploy.api_key
  defaults:
    provider: runninghub
  execution: remote
  impl: capabilities/external/comfyui.py
  status: planned
  source: 🆕 待接（S5）
- id: text_to_video
  name: 文生视频
  category: external
  description: 纯文本 → 视频（无参考图，成本较高）
  input:
    prompt: string
    duration: number
  output:
    video_path: string
  params_schema:
    duration:
      type: number
      min: 3
      max: 10
      default: 5
      label: 时长(秒)
  providers:
  - id: kling
    name: 可灵
    api_key_ref: kling.api_key
    model: kling-v3
    default: true
  - id: wanxiang
    name: 通义万相
    api_key_ref: aliyun.api_key
    model: wan2.2-t2v
  defaults:
    provider: kling
    duration: 5
  execution: remote
  impl: 'capabilities/kling_video.py: submit_text2video'
  status: ready
  source: ✅ 实测可用（kling-v3，2026-10-01 生成 5 秒视频成功）
- id: vfx_apply
  name: 特效
  category: external
  description: 子弹时间 / 环绕运镜 / 慢动作 / 焦点锁定
  input:
    image: file?
    video: file?
    vfx_type: string
  output:
    video_path: string
  params_schema:
    vfx_type:
      type: enum
      options:
      - bullet_time
      - orbit
      - slow_motion
      - focus_lock
      - custom
      default: orbit
      label: 特效类型
    pan:
      type: number
      min: -10
      max: 10
      default: 0
      label: pan(水平摇)
    tilt:
      type: number
      min: -10
      max: 10
      default: 0
      label: tilt(垂直摇)
    roll:
      type: number
      min: -10
      max: 10
      default: 0
      label: roll(旋转)
    zoom:
      type: number
      min: -10
      max: 10
      default: 0
      label: zoom(变焦)
  providers:
  - id: kling
    name: 可灵运镜
    api_key_ref: kling.api_key
    default: true
  - id: tianmu
    name: 万兴天幕
    api_key_ref: wondershare.api_key
  - id: hailuo
    name: 海螺AI
    api_key_ref: hailuo.api_key
  - id: comfyui
    name: ComfyUI
    api_key_ref: runninghub.api_key
  defaults:
    provider: kling
    vfx_type: orbit
  execution: remote
  impl: capabilities/kling_video.py:270 build_camera_control
  status: partial
  source: 🆕 待接（S5）；参数来自 DCS 第七节
- id: character_3d
  name: 3D 角色
  category: external
  description: 概念图 → 多视图 3D 模型 + 骨骼绑定
  input:
    image: file
    prompt: string?
  output:
    model_file: string
    preview: string
  params_schema:
    format:
      type: enum
      options:
      - glb
      - fbx
      - obj
      default: glb
      label: 格式
    rigging:
      type: boolean
      default: true
      label: 自动绑骨
  providers:
  - id: hunyuan_3d
    name: 混元3D
    api_key_ref: hunyuan_3d.secret_id
    default: true
  defaults:
    provider: hunyuan_3d
    format: glb
    rigging: true
  execution: remote
  impl: ''
  status: planned
  source: 🆕 待接（S5）；接口见 DCS 第六节
- id: frame_chain
  name: 首尾帧衔接
  category: video
  description: 上一镜尾帧 → 作为下一镜首帧/尾帧，长镜头分段生成后无缝接续
  input:
    tail_image: file
    prompt: string
  output:
    video_path: string
  params_schema:
    mode:
      type: enum
      options:
      - tail
      - head
      default: tail
      label: 衔接方式
  providers:
  - id: kling
    name: 可灵（image_tail）
    api_key_ref: kling.api_key
    param: image_tail
    default: true
  defaults:
    provider: kling
    mode: tail
  execution: remote
  impl: capabilities/kling_video.py:143-152
  status: partial
  source: 代码已实现 image_tail；⚠️ 分镜流程尚未接线（P1 待做）
- id: voice_lock
  name: 声线锁
  category: audio
  description: 一个角色 = 一个固定音色（跨镜/跨集逐字复用，外观变体不换锁）
  input:
    cast: list
    text: string
  output:
    audio_path: string
  providers:
  - id: local_lib
    name: 音色库（演员表绑定）
    api_key_ref: aliyun.api_key
    default: true
  defaults:
    provider: local_lib
  execution: local
  impl: composer/dubbing.py（resolve_voice_map / pick_voice_for_shot）
  status: ready
  source: 2026-10-03 配音链路实测；音色库见 config/cosyvoice_voices.yaml
- id: text_to_image
  name: 文生图
  category: image
  description: 文本 → 图片（角色定妆照 / 概念图 / 场景图 / 关键帧草图）
  input:
    prompt: string
  output:
    image_path: string
  params_schema:
    size:
      type: enum
      options:
      - 1K
      - 2K
      - 4K
      default: 2K
      label: 分辨率档
    n:
      type: number
      min: 1
      max: 12
      default: 1
      label: 输出张数
    negative_prompt:
      type: string
      default: ''
      label: 负向提示词（排除不要的）
  providers:
  - id: qwen
    name: 千问AI平台·万相2.7
    api_key_ref: aliyun.api_key
    model: wan2.7-image-pro
    default: true
  - id: qwen
    name: 千问AI平台·Qwen-Image
    api_key_ref: aliyun.api_key
    model: qwen-image-3.0-pro
  - id: qwen
    name: 千问AI平台·Z-Image（快/便宜）
    api_key_ref: aliyun.api_key
    model: z-image-turbo
  - id: kling
    name: 可灵
    api_key_ref: kling.api_key
    model: kling-v3
  - id: jimeng
    name: 即梦
    api_key_ref: volcano.ark_api_key
    model: seedream
  defaults:
    provider: qwen
    model: wan2.7-image-pro
    size: 2K
    n: 1
  execution: remote
  status: ready
  impl: character_sheet.py / 待接 qwen-image
  docs:
  - source: qwen
    slug: api-reference-image-generation-wan27-image-gen-edit-synchronous
    title: Wan 2.7 — 同步调用
- id: ref_to_image
  name: 参考图生图（多图融合）★
  category: image
  description: 1~9 张参考图（角色多角度 + 场景多机位 + 道具）+ 文本指令 → 融合成一张画面 —— 关键帧创作的核心能力：角色与场景不再二选一
  input:
    prompt: string
    ref_images: list[file]
  output:
    image_path: string
  params_schema:
    size:
      type: enum
      options:
      - 1K
      - 2K
      - 4K
      default: 2K
      label: 输出档
    n:
      type: number
      min: 1
      max: 6
      default: 1
      label: 输出张数
    mode:
      type: enum
      options:
      - single
      - sequential
      default: single
      label: 单张 / 顺序模式（一次多张且角色一致）
  providers:
  - id: qwen
    name: 千问AI平台·万相2.7 Pro（最多 9 图 + 顺序 12 张）
    api_key_ref: aliyun.api_key
    model: wan2.7-image-pro
    default: true
  - id: qwen
    name: 千问AI平台·Qwen-Image-2.1 Pro（最多 10 图 + 透明通道）
    api_key_ref: aliyun.api_key
    model: qwen-image-2.1-pro
  - id: kling
    name: 可灵（主体1-4 + 场景 + 风格分离）
    api_key_ref: kling.api_key
    model: kling-v2-1
  defaults:
    provider: qwen
    model: wan2.7-image-pro
    mode: single
    n: 1
  execution: remote
  status: partial
  impl: 待接（capabilities/kling_video.submit_multi_image2image 已有可灵版）
  note: 关键帧创作的主力；可灵版接口已实现但从未接入流程
  docs:
  - source: qwen
    slug: developer-guides-image-generation-wan-image-editing
    title: 图像编辑 - 万相2.7/2.6/2.5（多图输入多图输出）
  - source: kling
    slug: multiImage2Image
    title: 可灵·多图参考生图
- id: image_edit
  name: 图像编辑（指令）
  category: image
  description: 文本指令修改图片（换主体 / 换环境 / 改细节），保留其余部分
  input:
    prompt: string
    ref_images: list[file]
  output:
    image_path: string
  params_schema:
    size:
      type: enum
      options:
      - 1K
      - 2K
      default: 2K
      label: 分辨率档
  providers:
  - id: qwen
    name: 千问AI平台·Qwen-Image-Edit
    api_key_ref: aliyun.api_key
    model: qwen-image-3.0-pro
    default: true
  - id: qwen
    name: 千问AI平台·Wan 2.5 图像编辑
    api_key_ref: aliyun.api_key
    model: wan2.5-i2i-preview
  defaults:
    provider: qwen
    model: qwen-image-3.0-pro
  execution: remote
  status: ready
  impl: 待接（同步接口 ImageGeneration.call 支持 text+image 混排）
  docs:
  - source: qwen
    slug: developer-guides-image-generation-image-editing
    title: 图像编辑
- id: image_inpaint
  name: 局部重绘
  category: image
  description: 给定原图 + 涂抹掩码 + 描述，只重绘指定区域
  input:
    prompt: string
    image: file
    mask: file
  output:
    image_path: string
  params_schema: {}
  providers:
  - id: qwen
    name: 千问AI平台·wanx-x-painting
    api_key_ref: aliyun.api_key
    model: wanx-x-painting
    default: true
  defaults:
    provider: qwen
  execution: remote
  status: ready
  impl: 待接
  docs:
  - source: qwen
    slug: developer-guides-image-generation-image-inpainting
    title: 图像局部重绘
- id: image_outpaint
  name: 画面扩展（扩图）
  category: image
  description: 按宽高比 / 比例 / 方向扩展画面，或旋转扩图
  input:
    image: file
  output:
    image_path: string
  params_schema:
    expand_ratio:
      type: number
      default: 1.5
      label: 扩展比例
  providers:
  - id: qwen
    name: 千问AI平台·image-out-painting
    api_key_ref: aliyun.api_key
    model: image-out-painting
    default: true
  defaults:
    provider: qwen
  execution: remote
  status: ready
  impl: 待接
  docs:
  - source: qwen
    slug: developer-guides-image-generation-image-out-painting
    title: 图像画面扩展
- id: image_erase
  name: 擦除补全
  category: image
  description: 保留背景的同时擦除指定区域（去杂物 / 去路人）
  input:
    image: file
    mask: file
    keep_mask: file?
  output:
    image_path: string
  params_schema: {}
  providers:
  - id: qwen
    name: 千问AI平台·图像擦除补全
    api_key_ref: aliyun.api_key
    model: image-erase-completion
    default: true
  defaults:
    provider: qwen
  execution: remote
  status: ready
  impl: 待接
  docs:
  - source: qwen
    slug: developer-guides-image-generation-image-erase-completion
    title: 图像擦除补全
- id: image_style_transfer
  name: 图像风格迁移
  category: image
  description: 把写实图转成指定美术风格（漫画 / 3D / 国风 / 纸艺 …）
  input:
    image: file
    style: string
  output:
    image_path: string
  params_schema: {}
  providers:
  - id: qwen
    name: 千问AI平台·万相2.1 通用编辑（风格迁移）
    api_key_ref: aliyun.api_key
    model: wanx2.1-imageedit
    default: true
  - id: qwen
    name: 千问AI平台·Wan（背景生成/风格）
    api_key_ref: aliyun.api_key
    model: wan2.7-image-pro
  defaults:
    provider: qwen
  execution: remote
  status: ready
  impl: 待接
  docs:
  - source: qwen
    slug: developer-guides-image-generation-wan21-image-editing
    title: 万相2.1 通用图像编辑（含风格迁移）
- id: sketch_to_image
  name: 涂鸦作画
  category: image
  description: 手绘草图 + 文字描述 → 成品画面
  input:
    image: file
    prompt: string
  output:
    image_path: string
  params_schema: {}
  providers:
  - id: qwen
    name: 千问AI平台·涂鸦作画
    api_key_ref: aliyun.api_key
    model: wanx-sketch-to-image
    default: true
  defaults:
    provider: qwen
  execution: remote
  status: ready
  impl: 待接
  docs:
  - source: qwen
    slug: developer-guides-image-generation-sketch-to-image
    title: 涂鸦作画
- id: background_generate
  name: 背景生成
  category: image
  description: 为商品/人物图生成高质量背景（文本引导 / 图像引导 / 边缘引导）
  input:
    image: file
    prompt: string
  output:
    image_path: string
  params_schema: {}
  providers:
  - id: qwen
    name: 千问AI平台·图像背景生成
    api_key_ref: aliyun.api_key
    model: wanx-background-generation
    default: true
  defaults:
    provider: qwen
  execution: remote
  status: ready
  impl: 待接
  docs:
  - source: qwen
    slug: developer-guides-image-generation-background-generation
    title: 图像背景生成
- id: virtual_model
  name: 虚拟模特 / 换装
  category: image
  description: 保持真人站姿不变，替换背景与模特（服装试穿 / 模特图批量生成）
  input:
    image: file
    template: file
  output:
    image_path: string
  params_schema: {}
  providers:
  - id: qwen
    name: 千问AI平台·虚拟模特
    api_key_ref: aliyun.api_key
    model: wanx-virtualmodel
    default: true
  defaults:
    provider: qwen
  execution: remote
  status: ready
  impl: 待接
  docs:
  - source: qwen
    slug: developer-guides-image-generation-virtual-model
    title: 虚拟模特生成
- id: person_segment
  name: 人物分割（抠图）
  category: image
  description: 识别图中人物并输出像素级掩码（抠人 / 合成前置）
  input:
    image: file
  output:
    mask_path: string
  params_schema: {}
  providers:
  - id: qwen
    name: 千问AI平台·人物实例分割
    api_key_ref: aliyun.api_key
    model: person-instance-segmentation
    default: true
  defaults:
    provider: qwen
  execution: remote
  status: ready
  impl: 待接
  docs:
  - source: qwen
    slug: developer-guides-image-generation-person-instance-segmentation
    title: 人物实例分割
- id: ref_to_video
  name: 参考驱动生视频（R2V）★
  category: video
  description: 参考图/视频（最多 5 个混合）+ 文本 → 视频；可同时给角色图 + 场景图 + 首帧，还可绑音色（reference_voice）
  input:
    prompt: string
    media: list[{type,url}]
  output:
    video_path: string
  params_schema:
    resolution:
      type: enum
      options:
      - 720P
      - 1080P
      default: 720P
      label: 清晰度
    duration:
      type: number
      min: 2
      max: 10
      default: 5
      label: 时长(秒)
    ratio:
      type: enum
      options:
      - '16:9'
      - '9:16'
      - '1:1'
      default: '9:16'
      label: 宽高比
  providers:
  - id: qwen
    name: 千问AI平台·Wan2.7-R2V（5 参考 + 音色）
    api_key_ref: aliyun.api_key
    model: wan2.7-r2v
    default: true
  - id: qwen
    name: 千问AI平台·Wan2.6-R2V-Flash（¥0.3/秒）
    api_key_ref: aliyun.api_key
    model: wan2.6-r2v-flash
  defaults:
    provider: qwen
    model: wan2.6-r2v-flash
    resolution: 720P
    duration: 5
  execution: remote
  status: ready
  impl: 待接
  docs:
  - source: qwen
    slug: api-reference-video-generation-wan27-image-to-video-create-task
    title: Wan 2.7 参考生视频
- id: keyframe_video
  name: 首尾帧生视频（KF2V）★
  category: video
  description: 首帧图 + 尾帧图 + 过渡描述 → 平滑过渡视频（固定 5 秒）—— 关键帧 → 视频的主力
  input:
    prompt: string?
    first_frame_url: file
    last_frame_url: file
  output:
    video_path: string
  params_schema:
    resolution:
      type: enum
      options:
      - 480P
      - 720P
      - 1080P
      default: 480P
      label: 清晰度
    duration:
      type: number
      default: 5
      label: 时长(秒) · 固定 5
    template:
      type: string
      default: ''
      label: 特效模板（填了则只需首帧）
  providers:
  - id: qwen
    name: 千问AI平台·Wan2.2-KF2V-Flash（¥0.1/秒·480P）
    api_key_ref: aliyun.api_key
    model: wan2.2-kf2v-flash
    default: true
  - id: qwen
    name: 千问AI平台·Wan2.1-KF2V-Plus
    api_key_ref: aliyun.api_key
    model: wanx2.1-kf2v-plus
  defaults:
    provider: qwen
    model: wan2.2-kf2v-flash
    resolution: 480P
  execution: remote
  status: ready
  impl: 待接
  docs:
  - source: qwen
    slug: api-reference-video-generation-wan-image-to-video-first-last-frames-create-task
    title: Wan — 首尾帧图生视频
- id: video_style_transform
  name: 视频风格重绘
  category: video
  description: 把视频转成 8 种艺术风格（日式漫画 / 美式漫画 / 清新漫画 / 3D卡通 / 国风 / 纸艺 / 简易插画 …）
  input:
    video_url: file
  output:
    video_path: string
  params_schema:
    style:
      type: enum
      options:
      - 0 日式漫画
      - 1 美式漫画
      - 2 清新漫画
      - 3 3D卡通
      - 4 国风卡通
      - 5 纸艺风格
      - 6 简易插画
      - '7'
      default: 3 3D卡通
      label: 目标风格
    animate_emotion:
      type: bool
      default: false
      label: 情绪表情迁移
    use_SR:
      type: bool
      default: false
      label: 超分辨率增强
    min_len:
      type: enum
      options:
      - 720
      - 540
      default: 540
      label: 短边分辨率
  providers:
  - id: qwen
    name: 千问AI平台·视频风格重绘
    api_key_ref: aliyun.api_key
    model: video-style-transform
    default: true
  defaults:
    provider: qwen
  execution: remote
  status: ready
  impl: 待接
  docs:
  - source: qwen
    slug: api-reference-video-generation-video-style-transform-create-task
    title: 视频风格重绘
- id: video_edit
  name: 统一视频编辑
  category: video
  description: 一个接口靠 function 切换五大能力：多图参考 / 视频重绘 / 局部编辑 / 视频续写 / 画面扩展
  input:
    function: enum
    prompt: string
    ref_images_url: list[file]
    video_url: file?
    mask_*: any
  output:
    video_path: string
  params_schema:
    function:
      type: enum
      options:
      - 多图参考
      - 视频重绘
      - 局部编辑
      - 视频续写
      - 画面扩展
      default: 局部编辑
      label: 编辑类型
    strength:
      type: number
      default: 0.8
      label: 强度
  providers:
  - id: qwen
    name: 千问AI平台·Wan 统一视频编辑
    api_key_ref: aliyun.api_key
    model: wan-video-editing
    default: true
  defaults:
    provider: qwen
  execution: remote
  status: ready
  impl: 待接
  docs:
  - source: qwen
    slug: openapi-wan-video-editing
    title: Wan 统一视频编辑
- id: character_swap
  name: 视频换人（保留场景光线）
  category: video
  description: 把我的角色图换到别人的视频里，保留原视频的场景、光线与色调
  input:
    image_url: file
    video_url: file
  output:
    video_path: string
  params_schema:
    mode:
      type: enum
      options:
      - wan-std
      - wan-pro
      default: wan-std
      label: 模式（std 快/便宜 · pro 专业）
    check_image:
      type: bool
      default: true
      label: 图片合规检测
  providers:
  - id: qwen
    name: 千问AI平台·视频换脸
    api_key_ref: aliyun.api_key
    model: wan-character-swap
    default: true
  defaults:
    provider: qwen
    mode: wan-std
  execution: remote
  status: ready
  impl: 待接
  docs:
  - source: qwen
    slug: openapi-wan-character-swap
    title: Wan 视频换脸