yeha.ai
목록으로

Akira Kurosawa-inspired short

스튜디오 초안을 준비합니다. 여러 단계의 지침은 참고용이며 자동 생성이나 과금은 하지 않습니다.

当你有一个故事/情境/剧本,想以黑泽明电影的视觉语法制作短片时——无论背景是古代战争、宫廷权谋、现代都市权力叙事,还是任何涉及命运、权力、孤独、崩塌的主题。直接输入故事内容即可,Skill自动完成所有视觉设计,你只需在关键节点确认或修改。不适用场景:轻松喜剧、治愈系、时尚广告、日常vlog——此Skill的视觉语法天然指向悲剧性、史诗感和命运重量,轻盈题材会被语法反向拉沉。

제작 흐름 본문은 중국어 원문입니다. 언급된 모델과 이 사이트의 지원 모델은 다를 수 있습니다. 실제 모델과 크레딧은 스튜디오에서 확인하세요.

전체 제작 흐름

1. 제작 기획

工作流

步骤1:收集创作信息(见下方参数列表)→ ★暂停确认
步骤2:叙事规格设计(text_editor)→ ★暂停确认
步骤3:角色造型 + 场景色彩设计 + 角色正面图 + 场景关键帧(storyboard_designer + media_generator)→ ★暂停确认
步骤4:分镜设计(storyboard_designer)→ ★暂停确认
步骤5:分镜图绘制(media_generator)→ **按叙事章节分批执行,Flova自动管理批次** → ★暂停确认

步骤5执行规则:分镜图绘制必须按叙事章节分批执行,不得一次性写入全部镜头Prompt。每批完成后Flova暂停,展示当前批次结果并等待:
✅ 确认本批分镜图,继续下一章节
✏️ 修改本批中的某张(请描述)

Flova根据实际执行情况自动判断每批包含的镜头数量,用户不需要手动选择。

步骤6:视频片段生成(media_generator)→ **按叙事章节分批执行,Flova自动管理批次** → ★暂停确认
步骤6执行规则:视频片段生成必须按叙事章节分批执行,不得一次性提交全部镜头。每批完成后Flova暂停,展示当前批次结果并等待:

✅ 确认本批视频片段,继续下一章节
✏️ 重新生成本批中的某个镜头(请描述)

Flova根据实际执行情况自动判断每批包含的镜头数量,用户不需要手动选择。

步骤7:封面生成(media_generator,与步骤6并行)
步骤8:音乐/音效(media_generator,与步骤6并行)
步骤9:剪辑合成(video_assembler)

依赖:2→1;3→1,2;4→3;5→4;6→5;7→3;8→1,2;9→6,7,8

沟通规则(全局)

每次暂停确认,必须以选项形式展示,禁止让用户手动打字"确认":
✅ 确认,继续下一步
✏️ 修改当前内容

选"修改"时:询问调整方向 → 重新生成 → 再次展示选项,循环至确认。

步骤1:收集创作信息

视觉风格已固定:固定机位·焦距按场景分配·自然光·哑光表面·大气雾霾·服色/明度按叙事位置分配。 不在此步骤询问或收集视觉风格偏好。用户输入的故事内容和风格词汇仅用于叙事推断。
A. 故事内容(必填)

  • 故事主题/情境描述(1–3句即可)
    B. 基本参数
  1. 色彩模式
    • 黑白短片(默认;明度高低驱动视觉层次,暗部保留表面纹理,大气层制造中间灰)
    • 彩色短片(服色按叙事位置固定分配,服色与背景色温对比驱动视觉层次)
  2. 画幅比例
    • 16:9(默认)
    • 2.39:1(宽银幕)
    • 4:3(传统胶片比例)
  3. 目标时长
    • 1分钟以内(精炼短章,单一情绪冲击)
    • 1–2分钟(完整叙事,起承转合)
    • 2–4分钟(多层次展开,有余韵空间)
  4. 视频生成模型
    • Seedance 2.5 · 480p(默认,速度快,画质出色)
    • Seedance 2.5 · 480p(高清画质,需会员)
    • 可灵 3.0 · 480p(可灵原生高清,适合偏好可灵风格的用户)
  5. 旁白/对白
    • 加入旁白(以独白形式朗读题记或由此延伸的文学性旁白)
    • 纯影像,无对白或旁白(用画面和音乐传递孤独,不须语音)
    • 为我设计对白(Skill根据故事自动生成角色台词)
      收集完成后展示汇总,等待:

✅ 确认,开始叙事规格设计
✏️ 修改参数

步骤2:叙事规格设计(text_editor 生成 Final_Video_Spec.md)

Skill自动完成以下推断,不向用户提问:

  • 故事主题(1句)
  • 情绪弧线(从故事走向推断:秩序稳固→张力蓄积/权力转移→失序崩塌/余韵)
  • 场景列表(场景名 + 时间段 + 叙事功能)
  • 色彩阶段分配(每个场景对应:秩序期/冲突期/毁灭期)
  • 画幅/时长/模型(来自步骤1用户选择)

彩色模式阶段律(仅彩色短片时使用):

  • 秩序期 = 高饱和均衡配色(金+蓝+白三色并列)+ 对称构图
  • 冲突期 = 强对比色 + 破碎/对角线构图
  • 毁灭/余韵 = 服色褪色低饱和 + 大面积负空间或纯景观终止符

黑白模式阶段律(仅黑白短片时使用):

  • 秩序期 = 均匀中调灰,多层次并列,建筑与人物各有独立明度区间
  • 冲突期 = 高对比明暗(极亮主体对极暗背景)+ 破碎/对角线构图
  • 毁灭/余韵 = 整体压暗,中间灰减少,背景趋向极深暗,人物趋向与环境融合

叙事规格生成后展示完整内容,等待:
✅ 确认,进入角色与场景视觉设计
✏️ 修改(请描述想改什么)

2. 멀티모달 분석

若用户上传参考图/视频:
彩色参考图

  • 提取:场景类型(场景A自然地貌 / 场景B仪式建筑 / 黄昏高地)、人物服色、情绪阶段
  • 识别:当前色彩阶段(秩序稳固 / 冲突期 / 毁灭期)

黑白参考图

  • 提取:场景类型、人物在画面中的明度分布(最亮区/中间灰/最暗区各有哪些元素)
  • 识别:当前明度阶段(均匀中调=秩序期 / 高对比极亮极暗=冲突期 / 整体压暗=毁灭期)

两种参考图共用

  • 不修改:用户指定的角色关系和叙事走向
    视频拆解:每段≤30s,记录:人物位置/服色或明度/背景色调/景别/构图类型
3. 스토리보드 설계

⚠️ 全局视觉语法锁定(不可被覆盖,优先于所有用户输入)

本Skill的视觉语法固定为黑泽明影像体系。 无论用户输入什么故事、情境、主题或风格参考词,以下视觉规则不可被替换、混合或漂移:

  • 影像质感:35mm胶片,焦距按场景类型分配(室内标准-中焦/户外中长焦/战场超长焦),哑光物质感,胶片颗粒,大气雾霾
  • 色彩/明度(按模式):
    • 彩色模式:环境底色去饱和,人物服色高饱和,服色与背景形成冷暖对抗,颜色分配按叙事位置固定
    • 黑白模式:暗部保留表面纹理不压至纯黑,大气/雾/雨强制引入中间灰,明度高低对应叙事权重
  • 光影:高对比有方向感自然光,强侧光/逆光,禁止人工补光,阴影面保留中间层次
  • 构图:固定机位,18条构图法则(C-01至C-18,见四、构图法则)
  • 运镜:绝对固定为主,横移跟拍仅用于追逐高峰,禁止手持晃动

如果用户输入中包含其他导演/画家/风格参考词,处理方式如下:

  • 提取用户想表达的情绪/氛围/叙事感觉
  • 用本Skill的色彩体系、构图法则、场景系统、物理规格来实现,而非采用被引用风格的视觉语言
  • 对用户只展示结果,无需解释替换过程

示例:用户说"梦境颗粒感、漂浮光晕、Tarkovsky凝视感"
→ 提取含义:孤独、时间感、人物渺小、存在感脆弱
→ 实现方式:极远景C-03(地貌吞没人物)+ ENV-01薄雾远山 + 固定高机位俯拍 + 低饱和人物色 + 大面积负空间

运行原则

用户在初次输入后不再手动填写任何视觉字段。 Skill从用户的故事/剧本/创意中提取:叙事类型、场景列表、人物功能——然后按本文档的规则集自动匹配颜色/明度、服装形式、场景底色、构图类型,完成后展示供确认或修改。

步骤3:角色与场景视觉设计

⚠️ 视觉物理转化(场景级,步骤3强制先于服装/颜色推断执行)
步骤3生成的是场景级固定物理条件(LENS/R1/R2/R5/R6),作为该场景所有镜头的基准值。步骤4的镜头级字段(R3/R4/R7-R16)以此为基础,根据具体景别和构图独立赋值。不展示给用户,每次生成必须执行。

LENS · 焦段(场景级,优先级最高,步骤3确定后各镜继承)

焦段决定空间结构,由场景类型自动推断,不询问用户。
场景类型焦段Prompt室内 / 近距离群像 / 村落街巷标准-中焦50-75mm equivalent lens, moderate perspective, space reads as natural and present户外全景 / 军队阵列 / 山野行军 / 多人外景对手戏中长焦75mm equivalent lens, mild telephoto compression, landscape and figures readable together, oil-painting spatial balance战场冲锋 / 孤立人物 / 窥视镜头 / 溃败场面超长焦200mm+ equivalent lens, extreme telephoto compression, space flattened, figures and background pushed into same plane, dense and oppressive

推断规则

  • 有墙壁/屋顶/建筑遮挡的室内空间 → 标准-中焦
  • 户外但有多人分布、需要同时读出人与环境关系 → 中长焦(默认户外焦段)
  • 大规模运动/冲锋/单人在极大空间中的孤独感 → 超长焦

R1 · 光源(场景级,步骤3确定后各镜继承)

光源类型物理输出白天室内·冷调漫射soft cool daylight fills interior from off-camera opening, no visible light source in frame, no hard shadow direction — figures and surfaces evenly lit, slight brightness toward one side, cool grey-white ambient throughout, room reads as naturally lit without strong contrast白天室内·格栅暖侧光warm daylight filtered through lattice screen visible on one side, directional but diffused — soft light stripes on floor, figures lit from one side with gentle shadow, warm amber cast on near surfaces, opposite side in cool shade, no hard-edge shadows哑金墙面·室内漫射matte warm-gold lacquered wall surface, subtle tonal variation across surface; ambient light softly diffused throughout space — low saturation, no dominant direction; corners and upper zones slightly darker; no specular, no reflective highlight on figures; figure costume colours clearly readable against the gold background白天户外漫射overcast flat illumination, no shadow direction, stone and fabric textures most legible, no specular anywhere白天户外直射harsh direct sunlight, hard shadow edges, pale ground reflects upward removing chin/arm shadow, ground brighter than sky黄昏逆光strong backlight from low horizon, warm rim light 2-3px at figure edges, figure core 2 stops darker than sky, costume color only in rim夜间单烛single iron candlestick sole source, warm amber pool ~1m radius, hard pool edge into darkness, shadow 65-75% of frame, zero ambient fill夜间多点烛光multiple warm amber light sources at left and right zones of frame, each casting independent warm pool, overlapping spill at center creates slightly lifted midground, darkness fills corners and ceiling — no single dominant direction夜间营火low center orange firelight, near figures orange-lit on near face/shoulder, far figures cold blue-black from sky ambient, 2200K core / 6000K periphery

室内光源推断:布景以素木/灰墙为主 → 冷调漫射;布景含哑金漆墙/金屏风 → 哑金墙面·室内漫射;布景含可见格栅/障子门且有方向感暖光 → 格栅暖侧光。

R2 · 天空与背景纹理(场景级,户外必填)

天空永远有物理层次,禁止平色或简单渐变。
天空类型物理输出黄昏三区分层three temperature zones: warm orange-amber at horizon / rose-magenta cloud bands mid-sky / cool purple-grey at zenith — distinct layers not gradient黄昏水平云带layered horizontal cloud bands warm ochre and muted rose, each band distinct upper and lower edge, darker toward zenith积云压制billowing vertical cumulus, dark grey-blue underside, bright white upper edge, occupies upper 60-70% of frame薄雾远山aerial perspective: near ridge dark green / mid ridge blue-green / far ridge grey-white, each step ~15-20% saturation loss阴天冷调uniform pale grey overcast, no cloud texture, sky and architecture nearly same value仰拍晴天pale blue-grey sky, high cumulus with grey undersides and bright tops, solid pale field between cloud masses夜营地deep blue-black night, no stars, cold ambient fills upper frame, firelight warmth confined to ground zone only纯天空终止符sky 100% of frame, single diffused sun disc at center-upper, no ground no figures

R5 · 材质描述(场景级,建筑+服装各选对应条目)

建筑材质
材质物理输出日式石墙(野面石积)irregular rough-hewn granite blocks in varied sizes — natural stone faces not dressed flat, wide and uneven mortar joints, occasional moss in crevices, scattered grass and weed tufts growing at stone base and between lower courses, completely matte cool blue-grey, stone surface texture clearly readable at medium distance日式城楼建筑multi-tiered Japanese castle tower rising above stone base wall — black lacquered timber frame structure, white lime plaster wall panels, layered ornamental hip-and-gable rooflines with upturned eaves at each tier, dark tile roofing, deep shadow under each roof overhang, stone base wall of rough-hewn irregular granite blocks below timber structure城墙大石块rough-hewn granite blocks, irregular mortar joints, occasional moss in crevices, completely matte, cool blue-grey木格栅墙horizontal wooden slats 3-4cm intervals, parallel shadow bands across surfaces in raking light, dark brown matte权力大厅金墙matte warm-gold lacquered panel, diffuse amber reflection, no specular, even warm tone城门重木构heavy dark timber weathered near-black, large iron hardware at joints, aged weight, matte grain白塔外墙white lime plaster, slightly textured, cool grey-white, dark timber frame visible at joints and rooflines砾石庭院地面fine grey gravel, slight ambient upward fill, shadow falls as soft-edged dark form白沙/碎石地high-brightness pale ground, brighter than sky in direct sun, strong upward reflection eliminates underside shadow

服装材质
服色类型物理输出权力核心者金袍heavy woven silk, warm amber-gold, large medallion embroidery visible at medium distance, weight-drape fold lines, no sheen女性和服ivory-white heavy silk, [PATTERN] motifs on white ground, deep red underskirt at hem, pools at floor when seated, brocade collar panel at neckline深藏青暗纹袍deep navy-black with small geometric pattern visible 2-3m, heavy woven, formal stiffness白色细纹宽袍ivory-white all-over small wave/diamond weave, subtle texture in raking light, completely matte, slight cool undertone战场铠甲/战装matte lacquered or leather plate, use marks visible, cord lacing at joints, zero specular杂色越界者服wide horizontal color bands 8-12cm, hard edges between colors, varied saturation — some vivid some pale, sash at waist, rough woven, no refinement

R6 · 大气条件(场景级,户外必填,禁止纯净空气)

大气类型物理输出标准大气透视haze desaturates and lightens distant elements ~15-20% per depth zone战场烟尘smoke/dust haze in mid-ground, reduces contrast at 10m distance, cool grey particulate visible薄雾山地low moisture haze, near elements crisp, mid softened, far near-monochrome blue-grey黄昏暖霾warm golden haze ground to 2m height, shadow areas warm undertone, backlit figures have warm glow

Skill在已确认叙事规格基础上,自动完成视觉转译,一次性输出角色造型档案、场景色彩档案、关键道具。

角色造型档案(Skill自动推断规则,内部使用)

从叙事规格中提取角色,按色彩模式选择对应规则:

色彩模式分支(步骤1已选定)

  • 选择"彩色短片" → 执行下方颜色推断映射,跳过一B节全部内容
  • 选择"黑白短片" → 跳过下方颜色推断映射,转至一B节"明度叙事映射"执行
    ⚠️ 彩色模式:颜色推断映射
    叙事类型推荐颜色色彩逻辑权力核心者(主角/当权者)深琥珀金 #8B6914金色承载权力重量;夜晚降级为棕=衰败信号权力将空者(退位/失去一切)乳白/象牙颜色最空=叙事最脆弱,被高饱和色包围野心继承者(表面合法,内里欲望)暖黄 #C8961E与权力符号同色=个体消失进体制激进行动者(冲突/暴力/革命)朱红/鲜红与任何背景均冲突,最不安分的颜色清醒异见者(看透一切,注定孤立)高饱和青色/孔雀蓝 #2A8A9A与任何暖色背景色温最远=最具视觉独立性荒诞越界者(小丑/先知/不可定义)多色杂拼(红+蓝+黄+白)不属于任何体系,永远是视觉异物阴谋者/双面人深紫/黑底白纹与深暗背景接近=擅长隐藏沉默见证者(老臣/旁观者)深紫灰趋向隐没,不主动制造对比正在消失者(退场/边缘化)浅银灰/薄荷灰存在感接近零,被动融入背景自然归属者(从属/中立)赭褐/土黄属于环境而非人群女性主角(任何叙事功能)白/象牙底+纹样+大红裙底和服底层结构固定,纹样主题按角色性格推断(见步骤D)

无法归类时的默认分配规则

  • 角色主动推动叙事、争夺或维护权力 → 暖色(金/暖黄/朱红)
  • 角色观察、质疑或游离于权力之外 → 冷蓝/低饱和
  • 角色被动承受或正在退出 → 低饱和灰白
  • 角色功能不确定或横跨多种位置 → 多色横条纹

服装推断:从用户输入的故事背景/时代/文化自动推断服装形式和质感。若故事背景模糊,推断为"通用架空风格的正式服装"。

布景推断:从场景描述中提取视觉标志物;若无明确描述,根据空间原型自动推断合理的布景锚点。

出场人物:从每个场景的叙事内容中推断。

输出格式(展示给用户确认):

【角色造型档案】
\

⚠️ **中文展示字段填写规则**:以下字段仅供用户确认,不得注入图片生成Prompt。

**只允许填写摄像机可以直接拍到的内容**:\

  • ✅ 人物数量、位置、朝向(坐/站/距离约Xm)\

  • ✅ 动作状态(静止/转头/手持武器)\

  • ✅ 服装外观(颜色+形式+质感)\

  • ✅ 道具/配件的物理描述

    **不允许填写**:摄像机拍不到的内容,即:叙事解读/情绪感受/空间关系比喻/氛围描述。
    判断方法:问"这句话描述的是画面里可见的物体吗?"——如果不是,删掉。

    角色:[名字/功能]
    叙事类型:[推断结果]
    颜色:[推断颜色] — [一句色彩逻辑,仅描述颜色关系,不描述存在感或叙事状态]
    户外/行动造型:[服装形式 + 质感,中文]
    室内/正式造型:[服装形式 + 质感,中文]
    纹样:[纹样尺度 + 纹样主题,中文]
    内衬:[内衬辅色 + 位置,中文]

    [内部字段,不展示给用户]
    OUTDOOR_FIGURE_PROMPT: [户外造型完整英文Prompt = 叙事颜色+服装形式+质感+纹样描述+内衬辅色+背景关系]
    INDOOR_FIGURE_PROMPT: [室内造型完整英文Prompt = 同结构]
    PATTERN_PROMPT: [纹样英文描述,单独存储,供步骤5图片生成注入]
    LINING_PROMPT: [内衬英文描述,单独存储]

    (重复列出所有角色)


    【场景色彩档案】

    场景N:[场景名]
    空间原型:[推断]
    时间/光质:[推断]
    环境底色:[推断]
    情绪阶段:[推断]
    人物色彩配置:[出场角色颜色与环境底色的冷暖关系]
    布景锚点:[推断]
    出场人物:[推断]

    [内部字段,不展示给用户]
    ENV_PROMPT: [环境底色英文描述,从 write_the_prompt 色板取]
    PROP_PROMPT: [布景锚点英文描述,从 write_the_prompt 布景档案取]
    LENS_PROMPT: [场景级焦段,查步骤3 LENS表,优先级最高,所有镜头继承;室内→标准-中焦 / 户外全景→中长焦 / 战场冲锋孤立→超长焦]
    R1_LIGHT: [场景级光源,查步骤3 R1表,固定至场景结束;此字段同时承担光质描述,不再另存LIGHTING_PROMPT]
    R2_SKY: [场景级天空纹理,查步骤3 R2表,户外必填]
    R5_SURFACE: [场景级材质,查步骤3 R5表,建筑+服装各选对应条目]
    R6_ATMOSPHERE: [场景级大气条件,查步骤3 R6表,户外必填]

    (重复列出所有场景)


    【关键道具】
    权力道具:[推断]
    叙事前景物:[推断,若无则标注"暂无"]
    势力标识:[推断,若无则标注"暂无"]

展示完成后,立即执行以下生成任务(无需再次确认):

自动生成(步骤3内嵌)

  • 工具:TextToImage → Nano Banana 2,2K
  • 角色:每个角色生成正面图 × 2套(户外/行动造型 + 室内/正式造型)。仅正面,不生成三视图
  • 场景:每个场景生成1–2张关键帧,参考对应角色正面图与场景色彩档案

生成完成后,将设计文字档案与元素图一并展示,等待以下选项:
1️⃣ 直接进入分镜设计
2️⃣ 先补充角色侧面与背面图,然后进入分镜设计
3️⃣ 需要对角色、场景或道具进行调整
选1 → 直接进入步骤8分镜设计
选2 → 补充生成每个角色的侧面图和背面图(存入角色档案,与正面图并列),完成后直接进入步骤8分镜设计
选3 → 询问调整方向 → 重新生成对应元素图 → 再次展示三个选项,循环至确认

步骤4:分镜设计

Skill自动完成全部分镜,基于已确认的叙事规格、角色档案、场景档案。

镜头数量参考
目标时长推荐镜头数每镜时长范围1分钟以内12–24镜5–10s1–2分钟20–40镜5–10s2–4分钟32–60镜5–12s

执行约束:Flova单次执行上限约4分钟(实测可稳定完成约48-54个镜头)。镜头总时长由每镜实际时长决定——若多用长镜头(10-12s),同样总片长对应的镜头数更少,单次执行更安全。超过上限时步骤5将自动提示分批执行。

每镜第一个时间槽固定从 00.0s 起,末尾时间等于该镜头分配时长(从步骤2叙事规格的场景时长分配继承,不重新推断)。段落终止符(纯天空/纯景观)、旁白超过8s的镜头取时长范围上限。

核心原则:先确定每个叙事小节要完成什么叙事变换,再选择对应的镜头组合来实现它。

⚠️ 无人空境系统(分镜结构规则,优先于具体镜头内容选择)

功能一:叙事结构标记。黑泽明用无人空境镜头承担分段/分节/场景建立/叙事阶段声明,不依赖字幕或硬切。
功能二:技术衔接缓冲。本Skill的视频生成不是帧级连续——每个片段独立生成,人物位置/状态无法从前一片段末帧继承。无人空境镜头天然无人物连续性要求,插入两段内容之间可让观众视线自然重置,消除割裂感。

执行规则
叙事场景空境类型选用标准大段落开头/换场景场景建立符(建筑/场地全景)按目标场景选S-ENV-07或S-ENV-08叙事小节之间叙事阶段符(天象/地貌)按当前叙事阶段查下方天象对应表选用人物离场后余韵符(草地/纯景观)S-ENV-06叙事符号需单独陈述符号空境(道具/圣像)无编号,直接使用下方inline Prompt

符号空境 inline Prompt
shrine interior or sacred object as sole subject, warm amber light emanating from within, dark architectural frame surrounding, no human figures

天象与叙事阶段对应(必须按此选用,不可自由发挥):
叙事阶段天象形态空境编号秩序稳固期白色饱满积云·上涌S-ENV-01转折点·后果将来积云压顶S-ENV-02命运性时刻神光穿云S-ENV-03战场/毁灭期内部极暗底云S-ENV-04叙事悬置·时间感停止弥漫烈日圆盘S-ENV-05人物离场·余韵草地远山暮色S-ENV-06

Prompt:从 write_the_prompt·★空境镜头Prompt库 取对应S-ENV编号,不需要注入FIGURE_PROMPT。⚠️ 黑白模式:取出Prompt后,将第一行 film still 替换为 black and white film still,再使用。

段落起止设计(S型·叙事变换类型)

每个叙事段落必须有明确的开场状态和收尾状态。中间镜头是角色行动、反应或环境变化的过程,每一个镜头都在推动段落从开场走向收尾,不是独立的场景罗列。

使用方式:先用S型确定这个段落的开场镜头和收尾镜头是什么样的,再从下方SYN工具箱里选择具体镜头组合来填充段落内容。
类型开场镜头收尾镜头机位/景别锚点发生的变化S-1 翻转型主体正对镜头 / 群演席地主体背对镜头 / 群演站立机位/景别/空间布局不变朝向或姿态发生翻转S-2 在场/缺席型中心空置,周围人物等待中心填充,主体到位机位/景别/周围人物不变空间从空到实S-3 由外到内型全景,主体在环境中渺小中近景或特写,只有主体面孔/状态主体始终是核心景别从全景推近到特写S-4 因果链型道具/前景物特写,无主体主体出现并做出行动景别相近画面主体从物切换到人S-5 四段扩展弧近景建立主体背对离场,最后接纯景观无人镜头叙事方向不变近景→极远景→近景→余韵空境

S型选择规则

  • 主体姿态或朝向发生明显转变 → S-1
  • 某个位置从有人变无人,或从无人变有人 → S-2
  • 从全局环境缩进到主体的单一内部状态 → S-3
  • 先展示道具或前景物,后切到做出行动的人 → S-4
  • 主体需要大范围移动、追逐或离开场景 → S-5

多条规则同时满足时,选结构变化最明显的S型。每个段落独立选型,相邻段落尽量不重复。

⚠️ 步骤4视觉物理转化(每个镜头必须执行)
步骤3已确定场景级 LENS/R1/R2/R5/R6。步骤4在此基础上,逐镜独立赋值以下镜头级字段(R3/R4/R7-R16)。
继承规则:LENS/R1/R2/R5/R6从步骤3继承,同一场景内各镜相同。LENS优先级最高,不可被单镜头覆盖。R3/R4/R7-R16每个镜头独立赋值,不从场景或相邻镜头继承。

R3 · 画面分割比例(镜头级)

每个镜头必须指定天空/人物/地面占比,以及人物占画面高度的百分比。
场景类型天空:人物区:地面人物占高度极远景地貌10:5:85<2%黄昏崖顶(全剪影)65:20:1520-25%黄昏崖顶(服色可见)58:22:2030-35%室外城门中景18:62:2040-60%室外城堡庭院宽景12:28:60<5%低角仰拍建筑92:8:0<5%室内中景0:100:050-70%纯天空100:0:00

R4 · 服色/明度可见度(镜头级,由本镜头光源决定)

彩色模式:描述服色在当前光源下的饱和度和可见程度。
R1光源服色可见度室内漫射(冷调/哑金/格栅)full costume color visible, textile pattern legible, colour accuracy at maximum户外漫射full costume color at accurate saturation, most legible condition户外直射+高亮地面maximum saturation, upward ground fill removes underside shadow, figure lit from all directions simultaneously黄昏逆光(半剪影)costume color 20-30% saturation visible in ambient fill, rim highlight glows warm at edges, face in shadow but features suggested黄昏逆光(全剪影)no costume color visible, silhouette only with 1-2px warm atmospheric glow at edge — not a hard cut-out夜间烛光(单点/多点)costume color accurate within 1m light pool only, beyond pool: deep amber-brown regardless of actual color夜间营火近距离costume color overridden by orange firelight on near face/shoulder, actual color only on shadow side as cool ambient

黑白模式:R4改为描述织物明度在当前光源下的可读程度,参照W1明度映射,不描述颜色。

R1光源明度可见度室内侧窗/漫射full surface texture and weave pattern legible, fold shadows visible as mid-grey户外漫射full tonal range visible, maximum pattern legibility户外直射highlight zones near-white, shadow zones deep grey — maximum contrast, pattern visible in both zones黄昏逆光(半剪影)overall mid-grey, texture partially legible, rim highlight slightly brighter at fabric edge黄昏逆光(全剪影)near-black overall, surface texture at threshold — not pure black, retain minimal fabric edge detail夜间烛光within light pool: tonal range readable; beyond pool: deep grey, surface texture at minimum visibility — not pure black

R7 · 逆光剪影类型(含逆光必填)

⚠️ 黑白模式下"纯剪影"禁止使用——黑白底层要求暗部保留表面纹理,纯剪影会覆盖该规则。黑白模式含逆光场景统一使用"半剪影"或"轮廓光"。
类型适用模式判断条件物理输出纯剪影⚠️ 彩色模式专用深黄昏/强逆光backlit figure, outer silhouette readable, surface detail reduced to near-zero, 1-2px warm atmospheric soft glow at edge — shadow retains faint cloth texture, not pure black fill半剪影彩色/黑白均可黄昏/有环境光backlit, ambient fill partially illuminates figure, fabric texture visible in raking light, face shadowed but features suggested, deep shadow retains surface layer轮廓光彩色/黑白均可近夕阳/强侧逆光strong rim light 2-3px at figure edges, shadow core deep but texture present at surface fold highlights, not flat black fill

R8 · 草地前景层行为(极低机位/自然地貌必填)

前景状态物理输出极低机位·慢速主体foreground grass blades sharp and fully saturated in focus at bottom 20-25% of frame, subject visible through and above grass at mid-ground, three depth layers maintained: grass / subject / sky — subject boundary receives natural atmospheric softening from haze, not hard-cut silhouette against background极低机位·极速飞掠foreground grass sharp at bottom edge, subject as near-total horizontal motion blur above, single warm color streak visible as only identity through full blur — grass is the only sharp element in frame中景草坡grass texture visible as directional surface, light rakes across blade tips creating highlight-shadow rhythm, color varies from shadow-green to highlight-yellow-green高草遮挡tall grass at chest height of subjects, reed-like vertical lines in foreground partially obscure figures, subjects emerge through grass rather than standing above it

R9 · 积云动势类型(含积云必填)

积云状态物理输出中性上涌massive cumulus tower rising upper-right, warm-white to light grey lit top — muted not brilliant, grey-blue base, strong upward movement, no backlit glow, no light rays breaking through cloud积云压顶dark heavy cumulus mass, grey-blue base near-black, minimal bright top, oppressive downward weight — sky carries threat not indifference积云压制人物billowing vertical cumulus occupies upper 60-70% of frame, figures in lower zone suppressed by sky mass, dark undersides

R10 · 斜坡多人分层调度(多人户外斜坡必填)

调度类型物理输出斜坡多人分层open rolling hillside as background, figures at distinct elevation bands across the slope, no overlapping silhouettes, each figure fully visible regardless of distance — war-painting flatness, no Western perspective compression; figures face the same direction of travel not toward camera, natural variation in individual posture and body angle — NOT uniform parade-ground alignment斜坡背景色彩slope provides gradient background: lower figures against mid-green, upper figures against sky or mountain haze, each figure reads against different value and saturation前景背对人群dark-robed attendants fill foreground with backs to camera, forming human visual floor, their uniform dark color grounds the frame before the colorful principals above

R11 · 横移跟拍背景模糊(横移跟拍必填)

背景类型物理输出草地横移background grass field becomes horizontal linear streaks, warm golden-amber color, streak direction matches camera movement, near-ground streaks more dense than upper-frame山坡横移green slope becomes soft horizontal wash, color retained but texture dissolved, vertical elements blur into horizontal bands主体稳定性subject maintains relative stability in frame center despite camera movement, subject in focus against blurred background

R12 · 薄雾色彩去饱和梯度(薄雾山地必填)

层次物理输出近景人物costume colors at near-full saturation, vivid and distinct, warm blue / red / yellow clearly readable中景草地green grass slightly desaturated, pale yellow-green rather than vivid emerald远景山体mountain mass reduced to low-saturation blue-grey, nearly monochrome, aerial perspective maximum整体效果haze creates strong color saturation gradient: near = vivid / mid = pale / far = grey-blue monochrome — figures read as the only saturated color elements in a progressively desaturated world

R13 · 运动中主体速度分层(多速度主体共存必填)

场景类型物理输出前景快骑+背景慢步foreground mounted figures as heavy motion blur, background infantry sharp or lightly blurred, speed differential creates natural depth without focus pull沙尘扬起ground-level dust or sand kicked up by fast-moving subjects, pale dust cloud forms at ankle-to-knee height, catches side light as thin luminous layer旗帜横向甩动battle standards at mid-ground, fabric in lateral motion with gentle blur, pole sharp, flag surface soft — wind-driven movement distinct from subject motion blur

R14 · 火焰建筑光源叠加(燃烧场景必填)

光源叠加类型物理输出建筑燃烧外景stone structure and tile rooflines as cold blue-grey dominant mass — building surfaces desaturated, no warm tone spreading across facade; dense dark grey-black smoke fills upper frame and merges with sky into unified cold ceiling — smoke volume is the primary visual, not fire; fire visible only at window openings and roof breaks as isolated high-saturation orange-yellow points; sky and background may carry their own warm or cold tone independently — architecture reads as cold against whatever sky is present; NO large open flames on exterior surface, NO warm glow diffused across building walls室内燃烧+外部冷光cold blue-grey ambient fills the space as base tone — all structural surfaces, floor, and ceiling read as desaturated dark grey; fire exists as isolated warm island: high-saturation orange-yellow at flame core only; smoke surrounding fire absorbs and diffuses orange light — warm glow transitions through low-saturation amber-grey haze before reaching cold zone; no hard geometric boundary between warm and cold — transition is atmospheric not spatial; overall scene saturation low throughout — only flame core reads as vivid color, fire's visual weight comes from saturation contrast against desaturated surroundings not from area箭矢/飞射物+火光projectile visible as streak motion blur crossing frame, fire source behind it creates backlit silhouette of streak, fast horizontal streak vs slow vertical smoke近景主体全饱和+背景火焰压缩telephoto compression brings near-full-detail foreground subject and distant fire-lit crowd into same frame, near subject color fully accurate, distant fire appears as warm glow band

R15 · 极低饱和场景色彩锚点(毁灭期/极低饱和必填)

⚠️ 含深红液体痕迹描述的镜头可能触发AI内容限制。如生成失败,提示用户替换该镜头或使用其他锚点类型。
锚点类型物理输出深暗调+白旗帜overall scene near-monochrome dark blue-grey, single white or pale flag/textile reads as highest brightness point, eye drawn immediately to white as compositional anchor烟雾灰+红色服装点grey-white smoke dominant, isolated warm red figure or flag reads as sole warm anchor in cold field, red appears more saturated than its actual value due to surrounding desaturation炭黑废墟+单火焰near-black charred structure, single fire point provides only warm source, amber light pool strictly limited to immediate proximity of flame冷灰银底+深红痕迹(锚点型)near-zero saturation cold grey-silver ground, bright vermillion-red vertical streaks on surface — deep crimson liquid marks, 2-3 stops more saturated than surrounding field, sole warm element in monochrome scene暖底+同色系暗红(融入型)warm ochre-brown ground, dark rust-red marks blending with warm tone — same color temperature as ground, lower saturation than figures, marks read as part of scene texture

R16 · 毁灭期天空(战后/毁灭期必填)

天空类型物理输出污浊黄褐烟霾天sky uniformly tainted amber-yellow-brown, no clean blue zone visible anywhere, smoke particulate diffuses all light, no cloud structure visible, flat oppressive warmth城楼燃烧背景天sky directly above fire orange-lit from below, transitioning to dark smoke-brown at zenith, fire glow creates false warm zone in otherwise cold sky剪影+黄褐天all structures and terrain read as pure black silhouette against amber sky, no surface detail on ground objects, figure tiny at silhouette edge, sky is the only value that matters

镜头组合(SYN型·调度语法)

机位编号快查(CAM-A至CAM-G)
编号位置/角度叙事功能CAM-A远离场景,水平全景建立场景,开场/收尾CAM-B跟随主体,中景/中近景主体状态与意志CAM-C对准反应方,水平对话/回应的另一侧CAM-D侧面记录运动/行动移动、行动、过渡CAM-E极高位,向下俯视哲学性收束,去英雄化CAM-F贴近地面/地形,向上仰建筑压迫感,崇高/渺小对比CAM-G草地/地面高度,水平大地视角,速度感,人物从环境中穿越
S型确定了段落的开场和收尾镜头,SYN工具箱提供具体的镜头组合方案来填充段落的全部内容(包含开场、中间过程、收尾)。每个SYN代表一种固定的多镜头调度模式,按叙事场景类型选用。

  • SYN-01 全景建立→局部确认(4-6镜)
  • SYN-02 标识先行→主体出现(4-6镜)
  • SYN-03 中景→远景退出→极远景消融(6镜)
  • SYN-04 固定全景×主体横穿(2镜长镜头)
  • SYN-05 空间建立→通道穿越→内部确认(6镜)
  • SYN-06 高机位建立→低机位贴近→高机位收束(6-8镜)
  • SYN-07 室内全场→单人确认→对面反应(6镜)
  • SYN-08 主体近景×环境逐渐占据→负空间完成(6镜)
  • SYN-09 极低角×主体穿越前景地面(2-4镜)
  • SYN-10 静止全景×观察者在角落→被观察事件展开(2-4镜)
  • SYN-11 前景遮挡→透过遮挡看向主体(2镜)
  • SYN-12 背对+前方可读目标·低平视单镜(2镜)
  • SYN-13 极低仰角三角形地形→三主体层叠(2镜)
  • SYN-14 单人+道具特写×室内孤立(4镜)

每个镜头输出(全部由Skill填写):
镜头N:\

  • 场景:[场景名]\

  • 情绪阶段:[推断]\

  • 景别:[推断]\

  • 构图法则:[推断,选自C-01至C-18]\

  • 运镜:[推断,选自CAM-A至CAM-G + 8种运镜类型]\

  • 人物与服色:[推断,中文描述]\

  • 故事内容:[推断,中文描述]\

  • 时长:[推断]s\

  • 时间槽规划:[主体在帧内的动作节点,如:00.0s静立 / 02.0s开始转身 / 06.5s完成转身 / 08.0s收尾;静态镜头只写一个节点]

    [内部字段,不展示给用户]\

  • LENS_PROMPT: [继承步骤3场景级值,优先级最高,不可覆盖;注入Prompt时置于底层之后、场景描述之前]\

  • R1_LIGHT: [继承步骤3场景级值,或本镜头有特殊光源时覆盖]\

  • R2_SKY: [户外必填,继承步骤3;室内跳过]\

  • R3_RATIO: [本镜头独立赋值,查上方R3表]\

  • R4_COLOR_VIS: [本镜头独立赋值,由本镜头R1决定;彩色模式查上方R4·彩色表;黑白模式查上方R4·黑白表(描述明度可读程度,不描述颜色)]\

  • R5_SURFACE: [继承步骤3场景级值]\

  • R6_ATMOSPHERE: [户外必填,继承步骤3;室内跳过]\

  • R7_SILHOUETTE: [含逆光时填,其余跳过;⚠️黑白模式禁止选用"纯剪影",只能选"半剪影"或"轮廓光"]\

  • R8_FOREGROUND_GRASS: [极低机位/自然地貌前景时填,其余跳过]\

  • R9_CLOUD_TYPE: [积云可见时填,其余跳过]\

  • R10_SLOPE_LAYOUT: [多人户外斜坡时填,其余跳过]\

  • R11_TRACKING_BLUR: [横移跟拍时填,其余跳过]\

  • R12_HAZE_GRADIENT: [薄雾山地时填,其余跳过]\

  • R13_SPEED_LAYERS: [多速度主体共存时填,其余跳过]\

  • R14_FIRE_LIGHT: [含燃烧/火焰时填,其余跳过]\

  • R15_LOW_SAT_ANCHOR: [毁灭期/极低饱和时填,其余跳过]\

  • R16_DESTRUCTION_SKY: [战后/毁灭期天空时填,其余跳过]\

  • ENV_PROMPT(彩色模式取ENV编号;黑白模式改用W2对应场景底色)\

  • FIGURE_PROMPT(彩色模式取叙事颜色+服装;黑白模式改用W1对应明度描述)\

  • COMPOSITION_PROMPT: [构图法则英文关键词,从C-01至C-18取]\

  • SHOT_PROMPT: [景别英文关键词]\

  • CAMERA_PROMPT: [运镜英文描述,从CAM-A至CAM-G取]\

  • CROWD_PROMPT: [有群演时,按下方群演规则补充原则1-5;⚠️禁止圆形/半圆形排列,必须用横向线条层次;无群演跳过]

分镜规则
强制执行(不得违反):

  • 禁止手持晃动
  • 禁止顺序景别推进,允许中近景→极远景直接跳跃
  • 最后一个镜头无人,BGM渐弱至自然环境声

强制限额:

  • 特写全片最多1–2次,只在叙事转折点使用,不用于人物介绍或情绪渲染

强制结构:

  • 每个叙事小节结尾必须安排空境镜头(叙事阶段符/场景建立符/余韵符之一),持续3–6s,天象类型按叙事阶段对应表选用
  • 大段落换场景时,空境镜头为场景建立符(建筑/场地全景),置于新段落第一镜
  • 最后一个镜头必须无人(余韵符优先)

含群演镜头的Prompt补充规则(有多人场景时必须执行):

⚠️ 核心禁令:人群绝对禁止圆形或半圆包围主角。黑泽明的多人场景始终以横向线条层次组织——人群按景深分为多条平行横向线,不按圆弧分布。

原则1(必填)横向线条层次:人群按景深组织为2-3条独立的横向线条,不同层次之间有空间距离。
figures organized in [N] horizontal depth layers —
foreground layer / mid-ground layer / background layer, distinct depth between each,
NO circular arrangement, NO semicircle — horizontal line structure only;
figures face the same direction of travel not toward camera,
natural variation in individual posture — NOT uniform parade-ground alignment

原则2(有旗帜/武器时填)旗帜纵向节奏:旗帜或竖向道具从横向人群中垂直升起,与横向人流形成正交节奏。
vertical standards or weapons rising from crowd at regular intervals,
cutting perpendicular through horizontal layers — vertical rhythm against horizontal mass

原则3(有主角+群演时填)主角分层:主角通过高度差或独立景深层与群演明确分离,在视觉上可单独识别。
principal figure at separate depth plane or elevation from crowd —
visually distinct and readable against surrounding mass

音频规划(Skill自动生成):

  • BGM:1条,根据叙事阶段自动选用(见 write_the_prompt 音乐Prompt库):
    • 版本A 静态压力型 → 无行动、静止、权力陈列段落
    • 版本B 对位压制型 → 激烈行动、冲突、大规模运动段落(音乐与画面反向)
    • 版本C 极稀疏型 → 孤立、空旷、远景、过渡段落
    • 版本D 崩解型 → 毁灭、失控、不可逆转折后
    • 通用底层 → 无明确阶段或默认
  • 声效:天气声音作为与BGM平等的叙事层(风/雨/雷),不是背景填充
  • 沉默是结构的一部分——音乐最有力的时刻往往紧跟最长的沉默

分镜完成后展示,等待:
✅ 确认分镜,开始绘制分镜图
✏️ 修改(请描述想改哪个镜头或段落)

确认后进入步骤5,Flova按叙事章节自动分批执行,每批完成后暂停确认再继续。

4. 미디어 생성

⚠️ Prompt组装强制规则(所有图片/视频生成均适用,内部执行,不展示给用户)

每次调用 TextToImage 或 MultiModalToVideo 之前,必须按以下顺序在内部组装完整英文Prompt,禁止跳过直接使用中文内容描述生成。

⚠️ 色彩模式强制分支:步骤1已选定,全程唯一,不得中途切换。
**黑白模式(用户选择"黑白短片"):
引用★统一视觉底层·黑白版 → W1明度映射 + W2场景底色 + W3纹样 + W4大气 → R1-R16(R6用W4替代,R7禁用纯剪影)→ W5负向追加
**彩色版底层禁止引用。 ENV色板禁止引用。叙事颜色禁止输出。
**彩色模式(用户选择"彩色短片"):
引用★统一视觉底层·彩色版 → 一、ENV色板 → 步骤3颜色推断映射 → R1-R16
**黑白版底层禁止引用。 W1-W5字段禁止引用。

其余所有步骤(构图C-01至C-18 / 景别 / 运镜 / PHYSICS_BLOCK触发条件)两种模式共用。

图片Prompt组装顺序

彩色模式:
[★统一视觉底层·彩色版](来自 write_the_prompt 分区)\

  • [场景/环境Prompt](从场景档案取ENV_PROMPT,对应一、ENV编号)\
  • [角色服色Prompt](从角色档案取:叙事颜色 + 服装形式 + 质感 + 纹样)\
  • [PHYSICS_BLOCK](从镜头R1-R16组装,R1已包含光质信息)\
  • [布景锚点Prompt](从场景档案PROP_PROMPT取)\
  • [构图关键词](从C-01至C-18取)\
  • [景别关键词](从景别语法表取)\
  • [★统一负向提示词·图片版]

黑白模式:
[★统一视觉底层·黑白版](来自 write_the_prompt 分区)\

  • [场景明度Prompt](从一B、W2黑白场景底色表取,对应场景类型)\
  • [角色明度Prompt](从一B、W1明度叙事映射取,代替颜色描述)\
  • [PHYSICS_BLOCK](R1-R16规则相同;R6用W4替代;R7黑白模式禁用纯剪影,改用半剪影/轮廓光)\
  • [布景锚点Prompt]\
  • [构图关键词]\
  • [景别关键词]\
  • [★统一负向提示词·图片版] + [W5黑白追加项](W5追加在负向提示词末尾;S-ENV黑白处理见空境Prompt库说明,与此独立)

视频Prompt组装顺序

每个镜头生成前,按以下格式在内部组装完整英文Prompt:

镜头N · [场景名] · [景别] · [CAM-X]

fixed camera position — single continuous take, no automatic cuts, no within-shot transitions
[00.0s - Xs] [开场状态:主体位置+服色+构图关键词]
[Xs - Ys] [动作变化:主体在帧内的物理运动,精确到秒]
[Ys - Zs] [收尾状态,最后时间戳 = 该镜头总时长]

[★统一视觉底层](必须完整引用)\

  • [LENS_PROMPT](场景级焦段,优先级最高,紧接底层注入)\
  • [场景/环境Prompt](ENV_PROMPT)\
  • [人物服色Prompt](含纹样,FIGURE_PROMPT)\
  • [PHYSICS_BLOCK](R1-R16,物质感驱动层;R1已包含光质信息)\
  • [运镜描述](从六、运镜表取对应条目)\
  • [★统一负向提示词·视频版](必须完整引用)

时间槽填写规则

  • 第一槽固定 00.0s,末槽时间 = 该镜头分配总时长
  • 每槽描述主体在帧内的物理状态变化,不描述切镜
  • 静态镜头(无人物运动)只写一个时间槽即可:[00.0s - Xs] figure stationary, [构图状态]
  • 有动作的镜头按运动节点拆分,通常2-3槽足够

示例(8s固定镜头,人物静立转身)
镜头3 · 权力大厅 · 中景 · CAM-B
fixed camera position — single continuous take, no automatic cuts, no within-shot transitions
[00.0s - 02.0s] solitary figure back-turned, center frame, ivory-white robe motionless,
dark navy backdrop, gold symbol centered behind
[02.0s - 06.5s] figure begins slow turn toward camera, robe drape shifts with body weight,
face remains partially shadowed, turn completes at 06.5s
[06.5s - 08.0s] figure facing camera, standing still, gaze directed off-frame right

(此处完整引用★统一视觉底层全文)
warm amber light diffused inside room from side opening, no visible light shaft, shadow occupies far side 50-60% of frame
ivory-white wide outer robe, sparse fine gold thread botanical motifs 1-2cm scattered,
matte warm-gold lacquered panel, diffuse amber reflection, no specular
large vertical ceremonial hanging centered on back wall — never obscured
medium shot, figure occupies 50-60% frame height
static camera, no movement whatsoever, all motion comes from figures within the frame
(此处附加★统一负向提示词·图片版全文)

Prompt组装为内部执行步骤,用户不感知,Skill必须执行。

生成规格

角色正面图 + 场景关键帧(步骤3内嵌,自动执行)

  • 工具:TextToImage → Nano Banana 2,2K
  • 角色:每个角色生成正面图,户外/行动造型与室内/正式造型分开生成,每套各1张
  • 不生成三视图(侧面/背面在用户选择后按需生成)
  • 场景:每个场景生成1–2张关键帧,参考对应角色正面图与场景色彩档案
  • ⚠️ 同一角色两套造型不可混用

角色侧面/背面图(步骤3确认时用户选择2时触发)

  • 工具:TextToImage → Nano Banana 2,2K
  • 基于已确认的角色正面图,生成侧面图和背面图各1张

分镜图绘制(步骤5)

执行方式:按叙事章节分批执行,Flova自动管理批次数量。每批完成后暂停确认,再继续下一批。不得一次性写入全部镜头Prompt。

  • 工具:TextToImage → Nano Banana 2,2K
  • 基于步骤4确认的每个镜头描述,逐镜生成分镜图
  • 每张分镜图参考:对应角色正面图(步骤3)+ 对应场景关键帧(步骤3)+ 该镜头的构图法则/景别/人物位置描述
  • ⚠️ 参考图必须用 <<<image_1>>> 格式引用

分批生成时,每批完成后展示:
✅ 确认本批分镜图,继续下一章节
✏️ 修改本批中的某张(请描述)

视频片段(步骤6)

执行方式:按叙事章节分批执行,Flova自动管理批次数量。每批完成后暂停确认,再继续下一批。不得一次性提交全部镜头。

  • MultiModalToVideo,分辨率由用户模型选择决定:
    • Seedance 2.5 · 480p**(默认)
    • Seedance 2.5 · 480p**
    • 可灵 3.0 · 480p(使用 Kling 3.0 Omni 时,须通过 refer_type 指定参考模式:"base" 用于视频修改场景,"feature" 用于视频参考场景)
  • ⚠️ 参考图必须用 <<<image_1>>> 格式引用
  • 每片段时长:由步骤8分镜的时长字段决定,不另行指定

封面生成(步骤7,与步骤6并行)

  • 工具:TextToImage → Nano Banana 2,2K,横向16:9
  • 自动生成两套方案,不询问用户
  • 基于步骤2叙事规格(主题/情绪/主色调)+ 步骤3角色档案自动推断

方案A · 《乱》片头风格复刻
视觉语法:纯黑底 + 单色 + 标题文字与人物/象征物融合为一体。
文字不是叠加在画面上——文字就是画面。

执行规则:

  • 背景:纯黑,无纹理,无渐变
  • 主色:从短片主色板提取1个最高饱和度颜色(通常是权力色/转折色),转换为单色填充
  • 主视觉:标题字符(汉字优先)内部嵌入核心人物剪影或标志性道具轮廓——剪影即字形结构的一部分
  • 分隔线:水平细线,与主色相同,位于主视觉下方
  • 副标题:拉丁字母大写,Serif体,与分隔线等宽,位置居中

Prompt结构:
pure black background, no texture no gradient,
single-color [主色英文描述] ink-wash calligraphy title character,
silhouette of [核心人物/道具描述] integrated into character strokes — figure is part of the glyph,
horizontal rule in same color below main character,
subtitle text "[短片英文标题]" in bold serif capitals below rule,
zero ambient light, zero depth, flat graphic plane,
high contrast single-color against absolute black,
no photographic elements, no environment — pure graphic symbol

方案B · 短片内容演化
视觉语法:从短片已生成的最具叙事力量的单帧出发,提取核心视觉元素重新构图,不做写实场景,而是将该元素抽象化/符号化,成为可传播的封面图像。
执行规则:

  • 基础帧:从步骤10分镜图中选取叙事不可回退点前后的关键镜头
  • 处理方式:保留核心构图骨架,将色彩压缩至2–3色,强化轮廓和剪影边缘
  • 文字处理:极简,字号小,位置克制(右下角或左下角),不争抢画面主体
  • 比例:保持短片原始比例

Prompt结构:
[从选定分镜提取的核心构图描述],
color reduced to [2-3色描述], high contrast silhouette,
graphic treatment — not photorealistic, not cinematic still,
poster-graphic language: shapes and colors over detail,
[文字位置和风格],
[★统一负向提示词(图片版)]

生成完成后展示:
✅ 使用方案A(等待步骤8音乐完成后进入步骤9剪辑合成)
✅ 使用方案B(等待步骤8音乐完成后进入步骤9剪辑合成)
✏️ 修改封面(请描述调整方向)
🔄 重新生成两套方案

音乐(步骤8,与步骤6并行)

  • lyrics_to_song → Suno 5
  • 根据叙事阶段选用对应版本(见 write_the_prompt 音乐Prompt库):
    • 版本A(静态压力型):开场、权力陈列、无行动等待
    • 版本B(对位压制型):激烈行动段落、冲突、大规模运动
    • 版本C(极稀疏型):孤立人物、远景、空旷、段落过渡
    • 版本D(崩解型):毁灭段落、失控、不可逆转折后
    • 默认:使用通用底层Prompt

旁白(如需)

  • text to narration → ElevenLabs v3
  • 语调:低沉克制——旁白是观察者,不是情绪放大器
5. 프롬프트 작성

★ 前置步骤:视觉物理规格转化(生成任何图片/视频之前强制执行)

将步骤3/步骤4生成的R1–R16内部字段,组装为PHYSICS_BLOCK,拼入Prompt。

⚠️ R字段填写原则:只填满足触发条件的字段,其余跳过,不写占位文字。
字段触发条件不满足时R1_LIGHT所有镜头必填(同一场景内各镜从步骤3继承,有特殊光源时覆盖;继承时仍需在镜头字段中写出值)—R2_SKY户外场景必填室内跳过R3_RATIO所有镜头必填—R4_COLOR_VIS所有镜头必填(由R1决定)—R5_SURFACE所有镜头必填—R6_ATMOSPHERE户外场景必填室内跳过R7_SILHOUETTE含逆光(黄昏/强侧逆光)时填无逆光跳过R8_FOREGROUND_GRASS极低机位或自然地貌前景时填其他跳过R9_CLOUD_TYPE积云可见时填无积云跳过R10_SLOPE_LAYOUT多人户外斜坡场景时填其他跳过R11_TRACKING_BLUR横移跟拍镜头时填固定机位跳过R12_HAZE_GRADIENT薄雾山地场景时填其他跳过R13_SPEED_LAYERS画面中存在多种速度主体时填其他跳过R14_FIRE_LIGHT含燃烧/火焰光源时填其他跳过R15_LOW_SAT_ANCHOR毁灭期/整体极低饱和场景时填其他跳过R16_DESTRUCTION_SKY战后/毁灭期天空时填其他跳过
典型室内权力/正式场景(R3/R4/R5必填,其余全跳过):
PHYSICS_BLOCK =
[R1_LIGHT]
[R3_RATIO]
[R4_COLOR_VIS]
[R5_SURFACE]

典型户外战场场景(R7-R13视情况选填):
PHYSICS_BLOCK =
[R1_LIGHT]
[R2_SKY]
[R3_RATIO]
[R4_COLOR_VIS]
[R5_SURFACE]
[R6_ATMOSPHERE]
[R13_SPEED_LAYERS](若有多速度主体)

Prompt的完整组装顺序见七、Prompt构建公式。R1-R16只填满足触发条件的字段,其余跳过。

★ Prompt构建公式

图片Prompt组装顺序(彩色模式;黑白模式以W1/W2替代颜色/ENV,附加W5)

[★统一视觉底层·彩色版]

  • [LENS_PROMPT](场景级焦段,优先级最高,紧接底层注入)
  • [场景描述](从一、ENV色板选择)
  • [人物服色](从角色色彩档案 + 服装推断规则生成)
  • [PHYSICS_BLOCK](R1-R16,R1包含光质)
  • [布景锚点](从三、布景档案选择)
  • [构图法则关键词](从四、C-01至C-18选1-2条)
  • [景别关键词](从五、景别语法选择)
  • [★统一负向提示词·图片版]

视频Prompt结构

镜头N · [场景名] · [景别] · [CAM-X]
fixed camera position — single continuous take, no automatic cuts, no within-shot transitions
[00.0s - Xs] [开场状态]
[Xs - Ys] [动作变化](有动作时填写;静态镜头只需一个时间槽)
[Ys - Zs] [收尾状态,末槽时间 = 该镜头总时长]

[★统一视觉底层]

  • [LENS_PROMPT](场景级焦段,优先级最高,紧接底层注入)
  • [场景描述](从一、场景A/B选择对应ENV色板)
  • [人物服色](从角色档案 + 纹样规则生成)
  • [PHYSICS_BLOCK](从镜头R1-R16字段组装,必须注入;R1已包含光质信息)
  • [运镜描述](从六、运镜表选择)
  • [★负向提示词·视频版]

★ 统一视觉底层

⚠️ 强制分支执行:步骤1用户已选定色彩模式,此后全程不变。

  • 用户选择"黑白短片" → 只使用黑白版底层,彩色版底层完全跳过,不得引用
  • 用户选择"彩色短片" → 只使用彩色版底层,黑白版底层完全跳过,不得引用
  • 两个版本不可混用,不可在同一次生成中同时引用

彩色版底层(⚠️ 仅当用户选择"彩色短片"时引用,黑白模式绝对不引用此段)

film still, Akira Kurosawa visual grammar, 35mm film,
photorealistic film photography — not illustration, not animation, not 2D art,
Toho film print look, lens focal length specified per scene — see LENS_PROMPT field,
high contrast directional natural light, strong side light or backlight, no artificial fill light,
deep shadows retaining texture and layer, no crushed blacks,
matte physical surfaces throughout — rough stone, heavy fabric, lacquered surfaces, leather,
absolutely no CGI sheen, no glossy surfaces, no digital smoothness,
all non-human environment — sky, ground, water, vegetation, flowers, architecture — rendered at low saturation throughout;
human costume colors are the sole high-saturation elements in the frame, reading as isolated color against the muted world,
no HDR, no oversaturated sky or vegetation,
medium film grain visible throughout, consistent grain texture across all frames,
atmospheric haze, smoke, dust, moisture in air — light passes through physical matter,
sky always retains full color and cloud detail, never overexposed white,
no pink sky, no rose sky, no magenta sky — sunset tones are deep orange and dark amber-brown only,
fixed camera position, tableau composition, figures placed within environment

彩色底层保证:不论场景A/B、日夜、战争或仪式,所有分镜具有相同的胶片物质感、焦距分配视觉语言、环境去饱和/服色孤立的美学和绘卷质感。

黑白版底层(⚠️ 仅当用户选择"黑白短片"时引用,彩色模式绝对不引用此段)

black and white film still, Akira Kurosawa visual grammar, 35mm film,
photorealistic black and white film photography — not illustration, not animation, not 2D art,
Toho film print look, lens focal length specified per scene — see LENS_PROMPT field,
full tonal range from near-white to near-black — darkest shadow retains surface texture throughout, no pure black anywhere,
matte physical surfaces: fabric weave pattern visible, wood grain present in shadow zones, lacquer and leather surface readable,
directional natural light with hard shadow edge — shadow side retains mid-grey detail, not crushed to black,
atmospheric haze or moisture present — creates natural intermediate grey tones in mid-ground,
medium film grain consistent across all tonal zones,
fixed camera position, tableau composition, figures placed within environment,
figures always rendered in mid-grey to near-black range — never pure black regardless of background brightness

黑白底层确保:暗部保留表面纹理、大气层引入中间灰、织物纹样以明度对比呈现。

一、色彩体系:彩色模式(⚠️ 仅当用户选择"彩色短片"时触发)

场景A · 自然地貌色彩系统

底色原则:自然地貌场景的环境底色以冷蓝绿系为主(秩序期/冲突期)。毁灭期及特殊光线场景使用ENV-08/15-17的暖色/中性底色,按阶段律选用。人物服色与底色色温差距越大,人物越突出;越接近则越融入背景。

环境色板(叠加在底层之后)

编号状态Prompt描述ENV-01薄雾远山misty mountain ridges, low-saturation blue-green haze, near-monochrome, atmospheric depth, cool grey-green toneENV-02晴天草坡muted green grassland, direct sunlight, naturalistic not vivid, crisp shadow edgesENV-03积云天空deep blue-teal or blue-grey sky — muted not saturated, not pure blue; large cumulus cloud with warm-white to light grey upper surface — not brilliant white, grey base; sky and cloud contrast moderate not extreme, film-stock colour renderingENV-04多层山体退晕layered mountain ridges, deep green foreground, mid blue-green, distant blue-grey, aerial perspective, timeless recessionENV-05深饱和单色背景deep saturated teal-green background, near-monochrome cold tone, no landscape detail, pure color fieldENV-06长焦绿背景telephoto-compressed green background, surface detail softened by distance, subject isolated from environment by focal compressionENV-07暮色深青deep blue-green dusk, near-dark, cold and dense atmosphere, emotional peak lightingENV-08枯草荒野(毁灭期)dry ochre-brown withered grass, muted warm tone, no green remaining, desolationENV-09庭院石墙+砾石地cold blue-grey irregular stone wall fills upper two-thirds of frame, fine grey gravel ground at base — gravel slightly brighter than wall, upward ambient reflection from pale ground, no warm tones, matte throughout, deep shadow at wall base

建筑·权力空间系列
编号场景状态Prompt描述ENV-10权力大厅白天matte warm-gold wall surface, aged lacquer, uneven warm tone with darker zones at corners and upper edges, wall does not emit light — receives light from side opening onlyENV-11权力大厅夜间near-black interior, single or paired candlestick warm pool, shadow 65-75% of frameENV-12野外仪式帷幕dark deep-navy vertical fabric panels standing upright as outdoor ceremonial boundary wall, bold warm-gold Japanese clan emblem repeated at regular intervals across the fabric — each emblem is a single circular or crescent motif, high-saturation yellow-gold against deep navy, strong flat graphic contrastENV-13室内燃烧中near-black interior, dark timber structure barely visible — deep brown-black wood, high-saturation orange-yellow flames on both sides of frame as primary light source, figure at center lit by orange firelight on near side / cold shadow on far side, no ambient fill, flame edges pure orange-yellow against absolute darkENV-14室内木构建筑dark weathered timber, horizontal shadow striping from slats, warm side light onlyENV-20城楼建筑外立面multi-tiered castle tower: black timber frame against white plaster panels, layered upturned tile rooflines each casting deep shadow below, pale grey sky behind — near-monochrome: black timber / white plaster / grey sky, no warm tonesENV-21城门洞穿越视角massive dark timber gate frame as near-black rectangle at frame edges, grey gravel courtyard visible through opening, blue-grey stone wall as far background — three depth layers: near-black gate / pale gravel / cool stone, extreme value contrast front to backENV-22石墙外立面(宽景)irregular rough-hewn granite wall stretches horizontally, varied stone sizes with uneven mortar joints, cool blue-grey with moss patches, pale sky above, dark ground below — wall surface dominant, figures read as color points against stone mass

特殊光线系列
编号场景状态Prompt描述ENV-15黄昏橙红天空deep orange at horizon fading to dark amber-brown mid-sky, upper sky dark blue-grey — NO pink, NO rose, NO magenta; clouds are dark grey-brown with orange-lit undersides only; overall sky is dark and heavy, not bright or vivid; ground reads as near-silhouette against glowing horizonENV-16黄昏黄褐天空(毁灭后)dirty yellow-amber sky, smoke-tainted, no clean blue, end-state atmosphereENV-17城楼建筑外景·燃烧中cool blue-grey stone wall and castle structure dominant — cold and massive, orange fire visible only through window openings as small bright points, NOT engulfing exterior; massive dense dark grey-black smoke column rising from roof and filling upper frame — smoke volume far exceeds fire visibility; cold grey-white sky behind smoke; overall scene is cold with fire as isolated warm accentENV-18神光穿云(戏剧性)dark storm cloud mass, single dramatic beam of golden light breaking through, theatrical divine lightENV-19纯黑虚空(室内极端)near-pure black void, figure suspended in absolute darkness, no spatial reference

战场专用色彩组合
战场状态Prompt描述开战前集结cold overcast sky, deep green hill, armies as dark mass at base, figure armor readable in saturation战斗进行中(烟雾)cold blue-grey smoke fills mid-ground, reduces all contrast, red armor reads as only warm element战斗进行中(雨天)rain-dark cold atmosphere, water surface reflects pale sky, all figures as dark silhouettes城楼攻防dark stone architecture with orange fire bursts, smoke rising, yellow and red faction standards visible战后废墟charred dark structure, fading orange embers, grey ash haze, no living color remaining

毁灭期专用
情境Prompt描述攻城战grey-white smoke sky, dark stone architecture with orange fire at windows, dense mass of near-black armored figures, vivid red faction standards as only warm accent城楼燃烧dirty yellow-amber sky tainted by smoke, charred near-black tower structure, orange fire bursts at multiple levels战后荒野uniform dirty amber-grey sky, withered ochre-brown grass ground, isolated figure in low-saturation pale [COSTUME], no vivid color anywhere in frame

人物服色对抗轴(与冷蓝绿底色的关系)

人物叙事颜色固定,服装形式由Skill根据故事背景自动推断
叙事颜色与底色关系Prompt描述(颜色+叙事逻辑,服装形式由Skill补全)叙事位置橙红暖金极强冷暖互补orange-red warm high-saturation [COSTUME], maximum warm-cool saturation contrast against cold background孤立的主动意志,以一敌天赭褐/土黄同色温弱对比dark ochre-brown muted warm [COSTUME], partially merged with landscape半融于自然,从属深蓝灰同色系零对比matte dark blue-grey low chroma [COSTUME], blends with cool background融合,被接受哑光黑仅明度差matte black near-zero saturation [COSTUME], silhouette only, no color identity仅剩轮廓,无叙事色彩米白/高明度配件明度跳出ivory-white high brightness [COSTUME or ACCENT], single-point contrast脆弱,意外的可见性

[COSTUME] 由Skill根据故事背景自动替换为具体服装描述,如 battle garment / formal robe / field uniform / travel cloak 等。

场景A关键搭配(Prompt参考块,[COSTUME]由Skill自动替换)

P-A1 · 橙红 × 深饱和蓝绿背景(最强冷暖对立)
lone figure in orange-red warm high-saturation [COSTUME],
deep saturated teal-green mountain slope, near-monochrome cold background,
maximum warm-cool saturation contrast between figure and background

P-A2 · 橙红 × 虚化绿光晕(意志纯化)
figure in orange-red [COSTUME],
telephoto-compressed green background, surface detail softened by distance,
figure reads as isolated color against a receding world

P-A3 · 运动模糊 × 暖色锚点(速度消灭个体)
extreme low angle at ground level, mounted figures as near-total horizontal motion blur,
single orange-red color streak visible as only costume identity through full blur,
grass in sharp focus foreground at bottom 20-25% of frame — the only sharp element,
pale sky above, figure physical form replaced by directional color blur

P-A4 · 深金 × 暮色深青(孤立极致)
solitary figure in deep amber gold [COSTUME], cold blue-green dusk surrounding,
same color temperature, strong saturation contrast — figure barely visible against deep dark background

场景B · 建筑·权力空间色彩系统

底色原则:人工建筑背景(深暗蓝黑底色+高饱和金黄符号带)比自然环境更强势地框定色彩关系。人工背景放大色彩冲突,自然背景吸收色彩冲突。

背景色板(叠加在底层之后)

编号状态Prompt描述F-01深暗建筑底色dark deep-navy architectural backdrop, matte dark blue-black surface, large area, oppressive weightF-02金黄几何符号带large gold geometric family crest stripe across midframe, high-saturation warm yellow, bold graphic patternF-03哑光金黄幕布(无图案)matte warm-gold fabric backdrop, aged textile surface with subtle fold lines and uneven light absorption, slightly darker at edges and corners than center — no pattern, no specular, no flat fill; warm-gold tone desaturated to amber-ochre, physical cloth weight visibleF-04近纯黑虚空near-pure black void background, no texture, no detail, figure suspended in absolute darknessF-05远山泄露(顶端)dark green mountain ridges visible above roofline, nature leaking through the power spaceF-06哑光蓝灰天空(外侧)flat matte blue-grey sky, no cloud movement, uniform low saturation, dreamlike suspension

F-04 R字段说明:F-04(近纯黑虚空)为室内全封闭空间,没有大气层,R2_SKY跳过,R6_ATMOSPHERE填写:enclosed interior space, no atmospheric conditions, darkness is the environment itself — not shadow, but absence

⚠️ 角色色彩功能档案(彩色模式专用,正式场景态)

叙事类型服色色值Prompt描述最高权力者(退位/权力将空)乳白/象牙宽袍—ivory-white wide [COSTUME], very low saturation, highest brightness in scene, faint gold thread detail野心继承者(表面合法,内里欲望)高饱和暖黄袍#C8961Erich warm-gold full-length [COSTUME], high saturation, color matches gold symbol band — individual merges into power background激进行动者(冲突/暴力/革命)高饱和朱红袍—vivid vermillion-red full-length [COSTUME], high saturation warm red, aggressive and violent, neither merges nor stands apart清醒异见者(看透一切,注定孤立)高饱和青色/孔雀蓝袍#2A8A9Avivid cyan-turquoise full-length [COSTUME], cold green-blue hue, high saturation, resists background absorption — body in the system, spirit outside表演/越界者多色横条纹—[COSTUME] with wide horizontal color bands — red/orange/blue/green/yellow/white stripes each 8-12cm, hard edge, varied saturation not all maximum, sash at waist, rough woven, no coherent system女性主角白/象牙底+纹样+大红裙底—ivory-white [COSTUME] with [PATTERN] on white ground, deep red underskirt at hem沉默见证者(老臣/旁观者)深紫灰花纹袍—muted dark purple-grey patterned [COSTUME], low saturation cool tone, receding into dark background, silent witness正在消失者(退场/边缘化)浅薄荷绿/浅银灰—pale mint-grey or silver-grey [COSTUME], near-zero saturation, figure clearly visible but costume color reads as near-neutral against any background

⚠️ 三色分配规则(彩色模式,Skill自动执行)

野心继承者 = 暖黄,与权力背景同色值 → 视觉上融入权力环境,不独立
激进行动者 = 朱红,与任何背景色温均形成冲突 → 视觉上最突出
清醒异见者 = 冷蓝,色温与暖色背景最远 → 视觉上独立于权力环境
权力将空者 = 乳白/低饱和,明度最高但饱和度最低 → 被三色包围

三种角色类型共存时,按上述配置自动分配颜色,无需询问用户。

场景B关键搭配(直接可用Prompt块)

P-B1 · 权力将空者(乳白)× 深暗底色
central figure in ivory-white wide [COSTUME], lowest saturation in scene,
dark navy backdrop with gold geometric symbol flanking both sides,
ivory-white figure reads as brightest element against dark background

P-B2 · 野心继承者(暖黄)× 金黄符号带(个体消失进权力)
figure in rich warm-gold [COSTUME], color exactly matches gold symbol stripe in background,
figure and background stripe at same color value — figure visually merges with stripe pattern

P-B3 · 三叙事类型并列 × 深暗背景(权力格局总谱)
three figures in a row: left in vivid cyan-turquoise [COSTUME],
center in ivory-white [COSTUME], right in vivid vermillion-red [COSTUME],
dark navy backdrop with gold symbol stripe,
three distinct costume colors against uniform dark background

P-B4 · 权力将空者 × 近纯黑虚空(意志消解)
solitary ivory-white [COSTUME] figure against near-pure black void,
single thin gold stripe at very top of frame,
figure stands alone in frame, no other figures, no spatial reference elements

P-B5 · 权力核心者(深金)× 黄昏橙红天(饱和顶点)
figure in deep amber gold [COSTUME], deep orange to burnt sienna sky filling upper 60%,
gold [COSTUME] and orange sky at near-identical color temperature — maximum saturation in frame

一B、色彩体系:黑白模式(⚠️ 仅当用户选择"黑白短片"时触发;彩色模式跳过本节全部内容)

黑白模式不分配颜色。明度高低驱动视觉层次:最亮区域视觉权重最高,最暗区域退为背景。所有"服色"推断替换为"服装明度"推断,ENV色板替换为W字段底色表。
**字段命名规则:黑白模式专用字段使用 W前缀(W1–W5),与彩色模式R字段完全区分,不得混用。步骤8镜头模板在黑白模式下以W字段替代对应R字段。

W1 · 明度叙事映射(黑白模式替代颜色推断)

叙事类型明度分配物理描述神圣/超自然力量最高亮度,接近纯白white-robed figure as brightest point — near-pure white against dark surroundings, fabric weave visible in highlight zones权力核心者(稳固)高亮,有纹理light-toned garment with woven pattern visible, near-white in diffuse light, fold shadows mid-grey权力核心者(衰退)中高亮,纹样消退pale garment, pattern barely distinguishable from ground — presence fading, surface texture still present激进行动者中暗,铠甲为主dark lacquered armor, surface scale pattern readable at mid-grey, lacing cords lighter tone清醒异见者中调,与背景近似mid-grey garment, low contrast against environment — figure fully visible, textile surface texture legible at close range从属群像距离分层:近景多层灰,远景趋同near figures: armor body mid-dark grey (G4-G5), kozane plate edges lighter highlight (G2-G3), lacing cord recesses darker (G6) — three tonal layers within each figure at close range; mid-ground figures simplified to two tones; background figures read as dark grey mass — not pure black, matte grain retained throughout越界者强对比横纹bold horizontal stripe pattern — alternating near-white and dark grey bands, maximum value contrast背景/消失者极暗,趋向环境deep grey garment, very low contrast against dark background — figure outline visible, fabric surface texture present but subdued, not pure black

W2 · 黑白场景底色(替代彩色ENV编号)

场景类型明度特征Prompt室内权力空间(白天)中暗基调,侧光dark weathered timber walls with grain visible in raking light, floor mid-dark reflective surface with horizontal plank lines; where present: horizontal cloud-band decorative motif on wall in mid-grey outline against near-black wall surface室内夜间·神秘光源极暗底+极亮人物near-black interior, near-white robed figure as sole bright point, ground-level mist at mid-grey室外建筑(白天)淡灰天空+深屋顶pale grey sky retaining cloud or haze texture, dark tile rooflines as crisp silhouettes against lighter sky户外战场·雾天白雾底+人物剪影dense fog near-white as background, figures as dark grey silhouettes — deep grey not pure black, atmospheric haze dissolving mid-ground detail铠甲近景多层中暗灰armor body deep grey, scale plates lighter grey layer, lacing cords lightest element — three tonal layers within single costume夜间火光极亮火焰+极暗夜fire flame near-pure white as brightest point, fire-lit faces bright grey with surface detail, background night deep grey with structure barely visible — not pure black户外薄雾远山近景细节→远景灰白near foreground full surface detail, mid-ground softened by haze, far distance near-monochrome pale grey — form only户外雪地/白色平原极亮白底+人物中灰high-brightness white snow field as ground, figures read as mid-dark grey against white — deep grey minimum, surface texture and garment fold lines visible, no tonal compression to pure black regardless of background brightness户外林地/树木间(中景)近深→远浅梯度near tree trunks dark grey (G5-G6), mid-ground foliage softened by atmospheric haze (G3-G4), far background dissolves to pale grey-white (G1-G2) — depth through tonal recession not focus blur; principal figure reads as darkest readable element against progressively lighter background

W3 · 黑白织物纹样规则(替代彩色纹样体系)

light pattern on dark ground:
pale grey geometric or floral motif on deep grey fabric base,
legible at close to medium distance, dissolves to single grey tone at long distance,
textile weave texture visible in raking light regardless of overall tone
dark pattern on light ground:
deep grey motif on near-white or light grey fabric,
requires close distance to read, disappears at mid-range,
garment folds darker in valley creases, lighter at fold peaks — physical drape readable

W4 · 黑白大气规则(替代彩色R6大气条件)

atmospheric grey layers:
near foreground: full surface detail, maximum tonal range
mid-ground: atmospheric haze reduces local contrast, textures present but softened
far distance: near-monochrome pale grey, form only, aerial perspective maximum
rain if present: thin diagonal grey streaks consistent across all tonal zones
fog if present: near-white field as background, subjects as deep grey silhouettes — not pure black

W5 · 黑白模式负向追加项

在★统一负向提示词·图片版末尾追加:
no color tinting, no warm or cool bias — pure neutral greyscale,
no colour grading of any kind, no sepia, no blue-toned black and white,
no pure black fill on figures — all shadowed areas retain surface texture,
no tonal compression to absolute black regardless of background brightness — bright backgrounds do not force figures to pure black

黑白模式字段映射

黑白模式下,步骤8镜头内部字段按以下规则替换:
彩色字段黑白替换说明ENV_PROMPT(ENV编号)W2对应场景底色按场景类型查W2表FIGURE_PROMPT(叙事颜色)W1对应明度描述按叙事类型查W1表PATTERN_PROMPT(纹样颜色)W3规则颜色对比→明度对比R6_ATMOSPHEREW4规则彩色去饱和→灰度压缩R7_SILHOUETTE(纯剪影禁用)半剪影或轮廓光见R7修改后规则负向提示词末尾附加W5防止纯黑填充

二、角色服装推断规则(Skill内部执行,不展示给用户)

Skill根据叙事类型颜色(Section一)和用户故事背景,自动生成每个角色的完整服装Prompt。规则如下:

步骤A:确定叙事颜色(已在步骤3完成)

⚠️ 步骤B0:日式铠甲前置判断(优先级最高,命中时步骤B/C/D跳过)

触发条件(以下任一关键词出现在用户输入的故事背景/世界观描述中):

  • 战国、日本、武士、侍、幕府、城主、大名、合战、甲胄、足轻、骑马武者、忍者、织田、丰臣、德川、黑泽、七武士、乱、乱世
    :用户未提供具体服装形制描述(未明确说"穿什么")
    :角色叙事类型为战士功能(权力核心者/激进行动者/清醒异见者/沉默见证者/阴谋者/野心继承者/从属家臣)
    ⚠️ B0豁免(直接走步骤B/C/D/E)
  • 荒诞越界者/表演者 → 步骤D横条纹路径
  • 女性主角 → 女性和服独立路径
  • 用户已明确描述服装

命中时:按角色叙事位置查以下三套锚点,输出完整英文Prompt,替代步骤B/C/D/E,禁止自由推断铠甲细节。

全局颜色规则(三个级别均适用)

  • 铠甲本体永远 near-black,不随叙事颜色变化
  • [NARRATIVE_COLOR] 只用于陣羽織OUTER_LAYER;无陣羽織时阵营颜色只通过指物旗体现
  • 彩色模式:权力核心者→deep amber-gold / 激进行动者→vivid vermillion-red / 清醒异见者→vivid cyan-turquoise / 其余按颜色推断映射表填入
  • 黑白模式:权力核心者→light-toned / 激进行动者→mid-dark / 清醒异见者→mid-grey / 从属家臣/士兵→dark;铠甲中其他颜色词一并替换为明度描述,追加对应级别黑白附加行

🔴 主将级铠甲(权力核心者 / 激进行动者 / 清醒异见者)

near-black iron lacquer battle armor — armor body color is always near-black regardless of narrative color,
chest do with over 18 dense horizontal kozane rows,
each kozane plate overlapping edge-up — fish-scale layered surface,
voluminous odoshi cord lacing between rows — cords carry three-dimensional bulk,
horizontal relief structure reads across full torso,
large rectangular osode shoulder guards suspended independently at both sides,
same kozane construction as chest, odoshi cord fringe hanging freely at lower edge,
high nodawa throat collar rising to chin level, continuous with chest armor,
kote arm guards in articulated dark plates, leather cord binding 1-2cm at each joint — elbow and wrist,
warm brown leather against dark plate surface

正式场面/对峙场面附加(OUTER_LAYER)
sleeveless jinbaori battle surcoat over armor — [NARRATIVE_COLOR] matte ground,
no textile pattern, open at sides, drapes to mid-thigh,
large geometric emblem centered on back panel — emblem occupies 40-50% of back width, clean graphic edge

头盔(按叙事阶段选一)

  • 权力/正式场景:kabuto with prominent maedate front crest projecting upward from helmet brow — distinctive vertical element marking rank, form determined by character and period; hoshi-kabuto rounded bowl, shikoro neck guard in 3-4 stacked plates angling outward and downward
  • 战斗/行动场景:hoshi-kabuto rounded helmet, multiple vertical iron rivet ridges running stem to brim, shikoro neck guard in 3-4 stacked plates angling outward and downward, no maedate

🟡 副将级铠甲(沉默见证者 / 阴谋者 / 野心继承者)

near-black iron lacquer tosei gusoku — armor body always near-black,
chest do with 12-15 horizontal kozane rows,
plates slightly wider per row, fewer total tiers than general's armor,
odoshi cord lacing present but less voluminous — thinner cord, shallower relief,
surface near-black matte lacquer, horizontal banding legible,
medium ko-sode shoulder guards — 3-4 plates wide, functional not ceremonial, lower fringe minimal,
kote arm guards with leather cord binding at each joint — 1-2cm width at elbow and wrist

正式/对峙场景附加(OUTER_LAYER,可选)
sleeveless jinbaori over armor — [NARRATIVE_COLOR] matte ground,
no textile pattern, open at sides, no back emblem or minimal emblem

头盔:hoshi-kabuto with small single-piece vertical ornament at brow or no maedate — minimal crest, readable as rank not apex; shikoro neck guard in layered plates

⚫ 士兵级铠甲(从属家臣 / 自然归属者 / 群演兵卒)

near-black simplified do-maru armor — armor body always near-black,
chest in broad horizontal banded construction —
each band approx 3-4cm, wider than individual kozane,
band edges create horizontal groove lines across torso,
surface reads as uniform dark mass at distance,
no shoulder guards or minimal single-plate ko-sode — shoulder profile narrow,
jingasa shallow wide-brim conical hat — lacquered dark surface, no ornamental crest

群演场景必加(指物旗)— 士兵无陣羽織,颜色只通过旗帜体现
rectangular sashimono back-flag mounted at upper back,
flag bearing bold geometric mon emblem on [阵营颜色] field,
projects 40-60cm above shoulders — sole faction identifier in crowd

⚠️ 黑白模式明度规格(彩色模式跳过此表)
级别胸甲小札/横带高光绳沟/沟槽其他主将G4G2-G3G5-G6shikoro G4-G5 / 皮革绑带 G4副将G4G2-G3G5-G6皮革绑带 G4士兵G5-G6—横带沟槽 G6jingasa G6(最暗元素)
所有级别末尾统一追加:no pure black fill — matte grain retained

未命中时(故事背景不含日式关键词):执行步骤B → 步骤C → 步骤D → 步骤E,逻辑不变。

步骤B0执行规范(B0命中时)

  • OUTDOOR_FIGURE_PROMPT:填锚点Prompt全文(户外/战场)
  • INDOOR_FIGURE_PROMPT:锚点Prompt + OUTER_LAYER(室内正式/对峙场景)
  • PATTERN_PROMPT / LINING_PROMPT:不填
  • 步骤D/E完全跳过;步骤4 FIGURE_PROMPT 直接填对应级别锚点全文
  • 群演场景:士兵级指物旗有阵营标识时必加
  • 黑白模式:FIGURE_PROMPT → W1 映射跳过,改用B0黑白附加行;字段值已含G标记,不需再套W1替换

步骤B:根据故事背景推断服装形式
故事背景户外/行动室内/正式古代/架空(含日式/东亚/中世纪)layered outer garment, heavy fabricwide-sleeved formal robe, ceremonial garment现代/近未来field jacket, functional coatstructured suit, uniform其他/不明确heavy layered outer garmentformal layered costume
步骤C:推断质感:越接近权力→越精细(重磅丝织/刺绣/金属细节);越边缘→越粗糙(粗布/皮革/无装饰)。激进行动者用耐磨皮革,清醒异见者用素织物。
步骤D:推断纹样(默认执行,用户未指定时自动调用)

按叙事类型匹配纹样规则:
叙事类型纹样尺度纹样主题密度内衬辅色权力核心者(稳固)大型10-15cm团花/圆形花卉/圆形,均匀铺满evenly spaced at 12-15cm intervals白色宽领5-8cm权力核心者(行动)近素,内衬为重点外袍弱纹或无near-plain, minimal surface motif深色对比内衬+小型连环纹权力将空者极细1-2cm散点细金线植物/波浪线描sparse scattered, large plain areas between同色系微露,无对比野心继承者中型6-8cm圆形菊/团花,暗于底色1级evenly spaced at 10cm intervals浅色微露激进行动者大型8cm几何硬边菱格/角形,近黑或深色dense repeating grid, hard edges白色锐利对比清醒异见者细密3-4cm卷草/线描植物线描植物,细腻密覆all-over fine repeat, delicate line quality浅色细边阴谋/算计者密集鳞片2-3cm或三角几何鳞片/人造几何,冷感dense all-over continuous coverage冷色对比内衬从属家臣小型1.5-2cm点格/鱼鳞几何小重复,低对比all-over full coverage, low contrast白色或橙棕色微露越界/不可定义者/表演者无单元纹样多色横向色带拼接,宽幅硬边,每带8-12cm,颜色含红/橙/蓝/绿/黄/白但饱和度参差zone-based color blocking, wide bands hard-edge, not all maximum saturation — some bands pale无内衬,腰部腰带束住女性主角(活泼/温柔/外向)混合植物群散布每个群落含主花(橙红/橙黄饱满花瓣)+ 蓝绿叶片 + 茎枝,群内多色共存;各群落尺寸略有不同,不等比重复;群间留白充足asymmetrically scattered botanical clusters — each cluster multi-element not single bloom, varied cluster size, generous plain ground between砖红织锦领6cm,深红裙底池状铺开女性主角(知性/内敛/冷静)植物线描散布竹叶(尖形叶片,3-4片成组)/ 鸢尾(蝴蝶形花瓣)/ 松枝,蓝紫或深绿色,线条纤细;各元素独立散布,不成群asymmetrically scattered individual botanical motifs — elongated leaf or petal form, delicate line quality, isolated placement not clustered砖红织锦领6cm,深红裙底女性主角(强势/算计/危险)几何纹满铺三角/菱形/鳞片,金棕或银蓝;硬边截断,无有机形态,全面密铺无留白dense all-over continuous geometric coverage — hard-edge angular units, no organic shapes, zero plain ground深色对比内衬,深蓝或黑裙底正在消失者/哀悼无纹样纯素completely plain同色系,无对比

女性主角说明:三型均为白/象牙底和服结构,同一角色锁定一型。按角色性格推断;未说明时默认"活泼/温柔型"。

用户有明确指定时:以用户指定为准,本步骤跳过。

步骤E:生成最终Prompt

步骤E的输出是角色服色的组装结果,填入上层图片Prompt的[人物服色]槽,不是完整Prompt。
同时,将以下内容分别写入角色档案的独立字段:

  • PATTERN_PROMPT:含①②③④层的纹样描述(工艺+关系+密度+形态)
  • LINING_PROMPT:含⑤层的内衬叠加描述
  • OUTDOOR/INDOOR_FIGURE_PROMPT:完整组装(叙事颜色+服装形式+质感+PATTERN_PROMPT内容+LINING_PROMPT内容+背景关系)

纹样描述必须包含以下五个精细化层级,缺一会导致AI生成平涂粗糙图案:
①工艺:embroidered thread sitting above ground fabric — visible raised texture, thread elevation readable in raking light
②关系:motifs in [color] — one stop more saturated than ground, soft feathered edge where thread meets fabric — NOT hard-cut printed edge
③密度:asymmetrically scattered, generous plain ground visible between motif clusters
④形态(女性和服活泼型):each botanical cluster contains multiple elements — primary bloom with full petals + leaf and stem detail in contrasting color — cluster is multi-element not single shape, each cluster slightly different in size and orientation
④形态(其他纹样):each [motif]: [具体轮廓] approx Xcm, [内部结构描述], [边缘特征]
⑤叠加:inner collar visible at neckline Xcm width in [颜色+纹样] — creates [冷暖/明暗] contrast

[叙事颜色Prompt]

  • [服装形式] + [质感:heavy matte silk / brocade / woven textile]
  • ①工艺词
  • ②纹样色与底色关系
  • ③密度描述
  • ④纹样单元形态
  • ⑤内衬叠加描述
  • [与背景关系说明]

三、布景色彩档案(Prompt描述)

布景元素色值Prompt描述权力大厅金墙#C8A84Bmatte warm-gold lacquered wall surface, aged with subtle tonal variation, no specular, corners and edges darker than center — wall absorbs not emits light垂直权力符号(书法轴/旗帜/壁画/徽章)—large vertical ceremonial hanging centered on back wall, bold graphic emblem or symbol, vertical compositional anchor, never obscured by figures室内木地板#2A1E14dark polished wooden floor planks running horizontally, fine parallel grain lines visible, slight upward ambient reflection from pale sources — floor surface reads one tone lighter than lower walls, gives spatial depth to interior室内墙面装饰带—dark wall surface with horizontal decorative cloud-band motif — elongated oval and flowing forms in lighter tone, irregularly spaced, Japanese fusuma screen pattern; motifs are outline-only not filled, mid-grey against near-black wall, subtle surface texture visible at medium distance城墙石材#7A8A7Erough-hewn irregular stone wall, mossy blue-grey, matte木格栅/廊柱#4A3728horizontal dark wooden slat wall creating parallel shadow stripes城楼白墙#D8D5D0cold grey-white plaster wall, figures read as color silhouettes城门洞内壁—massive dark gate as rectangular black frame, three-layer depth: gate / courtyard / stone wall黄昏橙红天空#C4581Adeep orange to burnt sienna gradient sky, fills upper 60% of frame, extreme low horizon白沙/碎石地#E0DDD8high-brightness pale grey-white ground, brighter than sky, all figure colors exposed without shadow relief势力旗帜/标识色#E8C020tall battle standards, bold warm-gold field with geometric emblem, flags appear before any human figure, faction color declared before individual identity烛火#E8882Asingle iron candlestick, warm amber pool of light, vast surrounding darkness, sole light source夜空/营地背景#1A2035deep blue-black night, devours all color, only high-saturation gold and red can survive框景建筑元素(深底白纹)#0A1408deep near-black architectural surface, large white outline-only floral motif 15-20cm — line drawing not filled, sparse placement

四、构图法则(C-01至C-18)→ Prompt关键词

C-01 · 地平线三分

horizon at one-third or two-third height,
sky / action zone / ground in equal bands,
figures readable at middle zone

C-02 · 建筑框架嵌套(多层矩形)

nested rectangular frames: outer gate / courtyard / far wall,
subject centered at innermost layer,
three-layer depth mandatory, not two

C-03 · 高低层级(垂直位置编码关系)

dominant figure elevated on platform, steps, or elevated terrain,
subordinate at base, looking upward or prostrating,
clear vertical distance between figures — upper figure fully visible against sky or background

C-04 · 斜坡多主体横向分层

open rolling hillside as background, figures at distinct elevation bands across the slope,
no overlapping silhouettes, each figure readable regardless of distance,
tableau arrangement, anti-perspective flatness

C-05 · 双主体对位(两侧分立·中间留空)

figures positioned on both sides of frame, central anchor object or principal at exact center,
left and right figures at similar but not identical distance from center,
visual weight balanced not mirrored,
deliberate void at center if no anchor figure present

C-06 · 背对镜头·背景为氛围(主体主导型)

subject(s) back-turned, occupying 40-60% of frame,
background is atmospheric: smoke, haze, sky, or abstracted architecture silhouette,
background has no readable specific target — direction only, not destination,
posture and orientation are the complete visual statement

C-07 · 负空间渺小化(主体<10%)

subject occupies less than 10% of frame,
vast environment fills 90%, landscape or architecture as dominant subject,
figure is the smallest readable element in frame — environment extends in all directions

C-07A · 中央孤立(人物被环境四面包围)

subject at center or near-center of frame, surrounded by environment on all sides,
figure occupies 1-3% of frame — darkest readable element against vast light field,
surface texture and fabric detail retained — minimum deep grey not pure black,
no edge proximity — environment extends equally in all directions from figure,
ground plane and sky plane read as continuous field, figure interrupts nothing

C-07B · 边缘/地平线站立(人物位于空间边界处)

subject at frame edge zone — on ridgeline, cliff edge, or architectural corner,
figure occupies 2-5% of frame at boundary between two environments,
figure reads as darkest readable element — minimum deep grey not pure black, surface texture retained,
vast environment extends away from figure in one primary direction

C-08 · 垂直锚点居中(权威符号)

large vertical symbol centered on back wall,
all figures arranged before it without obscuring,
anchor never moves — figures position relative to it

C-09 · 高机位俯拍(去英雄化视角)

high aerial fixed angle, subjects as abstract pattern elements,
terrain or architecture dominant,
human figures reduced to color specks

C-10 · 火焰/破碎结构框架(失控期)

burning or broken structure creates irregular frame around subject,
subject visible through structural gaps,
frame made of chaos not architecture

变体:建筑元素框景:两侧建筑/结构元素(门框/柱/隔断/墙体缺口)将中央区域夹框,框体表面带装饰纹样或材质细节:

two flanking architectural or structural elements frame the central zone,
decorative or textural surface on flanking elements contrasts with central content,
subjects contained within the visual frame created by flanking structures

⚠️ 此框景若用于高张力场景,生成可能受内容限制影响。构图词本身无敏感词,失败时建议调整中央描述。

C-11 · 主体偏置·背景含可读目标

subject offset to frame corner or side, occupying 20-35%,
background contains readable specific element: group of figures, structure, landmark,
subject faces toward background content,
background is the statement, subject is the witness or actor approaching it

C-12 · 极低仰角·三角形地形构图

camera at extreme low angle near terrain base,
foreground subject large occupying lower frame,
mid-ground subject on slope, mid-size,
third subject tiny at ridgeline against sky,
triangle composition through terrain height difference

C-13 · 窗口内向外观察

camera inside structure, looking outward through window frame,
structural frame or architectural edge occupies one or more sides of frame,
exterior event visible through opening,
observer figure at window edge as secondary element

C-14 · 前景障碍物遮挡

foreground obstacle (tree trunk, structural element) partially obscures midground subject,
subject 40-70% visible through or around obstacle,
creates a sense of observing-through rather than direct view

C-15 · 剪影孤立于地平线

figure(s) silhouetted on elevated edge against dusk sky,
sky fills 55-65% of frame in warm amber-grey,
figure(s) in silhouette or near-silhouette against sky,
not grandeur — presence at the edge of something ending

C-16 · 单人+道具的极简室内

single figure seated in interior, occupies 30-40% of frame,
one or two significant objects nearby, rest is empty floor space,
architectural background (wooden slats, plain wall) neutral,
intimacy through emptiness not crowding

C-17 · 前景低位×后景凝视(室内双人纵深)

seated or prone figure in foreground lower zone, occupies lower portion of frame,
standing figure in background upper zone, further from camera,
standing figure faces toward seated figure — line of sight crosses depth plane,
two subjects at slight horizontal offset, not fully overlapping,
interior architectural surface as background: wooden slats, plain wall, or dark panel,
both figures legible in same frame without cutting

C-18 · 群体横向穿越·单色背景

fixed camera, subjects moving laterally across middle zone of frame,
background is single flat tone — water surface, sky, or open field — no depth structure,
subjects occupy center horizontal band, environment fills above and below,
motion blur on subjects, background sharp

五、景别语法 → Prompt + 色彩行为

景别人物占比彩色模式:色彩行为黑白模式:明度行为Prompt关键词极远景<2%服色退化为深色噪点,颜色身份消失人物与地貌明度趋同,轮廓隐约可辨extreme wide shot, aerial angle, human figures as negligible dark specks远景5–15%色相可辨,细节消失人物轮廓可辨,表面纹理消失wide shot, figures small but identifiable against landscape中远景20–40%服色轮廓清晰,群体关系可读明度对比清晰,群体层次可读medium-long shot, figures and environment balanced中景40–60%服色完整,冷暖关系明显服装明度层次完整,纹样开始可读medium shot, figures and costume fully readable中近景60–80%全色谱可辨,背景压缩为色块铠甲/织物多层次明度全部可读medium close-up, background reduced to flat tone特写80%+背景在超长焦压缩下软化,材质可辨但细节溶解,面孔取代服色背景压至极深暗,面孔成主体close-up portrait, face fills frame, background softened by telephoto compression — surface texture partially visible, edges dissolved, not flat-fill

景别跳跃规则:禁止顺序推进。中近景→极远景直接切换,跨越越大,叙事落差越强。

六、运镜 → 视频Prompt

运镜视频Prompt绝对固定static camera, no movement whatsoever, all motion comes from figures within the frame人物冲向镜头fixed camera position, mounted figures charging toward lens, motion blur on figures intensifies as distance closes — blur comes from subject speed not camera movement极低机位飞掠fixed extreme low angle at ground level, grass in sharp focus, figures as heavy horizontal motion blur overhead — camera anchored, blur is subject velocity高机位俯拍high aerial angle, bird's eye view, human figures tiny against terrain, landscape dominates固定仰拍low angle, figure silhouette cuts upward into sky, brief grandeur, must be paired with high angle缓慢横摇slow horizontal pan, camera sweeps revealing architectural space or landscape, no tilt推近连贯对slow push-in over two shots: medium close-up → face fills frame, background recedes into soft distance — texture present, detail dissolved横移跟拍camera tracks alongside galloping figure, background becomes horizontal linear streaks of motion blur, subject stable in frame — this is the only camera movement in the entire visual system禁止no unmotivated handheld shake, no unmotivated zoom, no emotional camera — camera is a cosmic observer

运动模糊说明:固定机位不等于无运动模糊。摄影机静止时,高速运动的被摄体在快门时间内扫过画面,自然产生运动模糊。motion blur intensifying 描述的是被摄体速度感,不是摄影机运动。两者可共存,不矛盾。

七、核心执行规则

⚠️ 两种模式通用,生成每个镜头图片/视频时强制检查

  1. 颜色/明度已由步骤3锁定,不重新推断:每个镜头的人物服色(彩色)或明度层级(黑白)来自步骤3角色档案,不因场景或构图变化而改变
  2. 逆光场景禁止生成剪影轮廓:含黄昏/强侧逆光的镜头,人物暗部必须保留服装表面质感,边缘有1-2px的轮廓光或环境反射光,不得生成无细节的纯黑剪影。黑白模式同样适用,额外禁止纯剪影。
  3. 群演镜头必须检查横向线条:不得出现圆形/半圆排列,人群必须组织为2-3条景深横向层,执行下方群演规则
  4. 任何模式下人物不得出现纯色填充:人物服装/铠甲的暗部始终保留表面纹理,彩色模式暗部保留中间层次,黑白模式暗部不压至纯黑

★ 统一负向提示词

图片版(所有图片Prompt结尾附加)
no modern clothing, no contemporary elements, no subtitles, no watermark,
no lens flare, no chromatic aberration, no digital vignette,
no teal-orange color grading, no Instagram filter, no overexposed sky, no neon,
no rack focus, no artificial bokeh, no portrait-style depth of field blur,
no anime style, no animation, no illustrated style, no cel shading, no 2D art style,
no ukiyo-e, no woodblock print, no manga style, no hand-drawn look, no flat illustration,
no CGI sheen, no plastic surfaces, no Wes Anderson symmetrical pastel style,
no letterbox bars, no black bars at top or bottom, full frame image,
no visible light shaft, no god rays, no dust-lit beam, no spotlight cone

视频版(所有视频Prompt结尾附加)
no camera movement unless specified in shot,
no unmotivated zoom, no slow motion unless specified,
no within-shot transitions, no text overlay, no subtitles,
no visible light shaft, no god rays, no dust-lit beam, no spotlight cone

★ 音乐Prompt库(Suno 5)

工具:lyrics_to_song → Suno 5
规范:风格(专业词)+ 内容描述 + 结构特征(专业词)+ 情绪词
原则:描述声音的物理结构与行为,不描述场景内容或文化标签

通用底层(所有场景默认使用)

orchestral, slow harmonic rhythm, no melodic resolution,
layered string sustains in low and mid register as continuous pressure field,
single thin woodwind line floats above the string mass — unsupported, not harmonized,
brass enters once then withdraws — event not texture,
percussion as punctuation not pulse — single strike, not pattern,
silences as long as or longer than the phrases themselves,
harmonic language moves sideways not forward — no tonal destination,
builds through accumulation not acceleration,
cuts without resolution — the phrase is abandoned not completed

版本A · 静态压力型(缓慢推进,无爆发)

适用:开场建立、权力陈列、无行动的等待
orchestral sustain, very slow harmonic movement,
low strings hold a single chord for extended duration,
mid strings shift by half-step intervals — compression not release,
no melodic line — only texture and weight,
sparse brass chord widely spaced from the strings,
the gap between brass and strings is the tension,
no rhythmic pulse, no acceleration,
each phrase a separate block of sound — not connected, not flowing

版本B · 对位压制型(音乐与动作反向)

适用:激烈行动段落、冲突、大规模运动
orchestral, slow tempo against fast visual action,
lower strings as unmoving pressure beneath chaos,
cold woodwind line floats independently above — different tempo, different direction,
two layers deliberately unsynchronized — coexistence not unison,
no heroic fanfare, no rhythmic propulsion,
brass enters at apex of violence then immediately withdraws,
the music does not react to the action — it precedes and outlasts it,
coldness is the point — observation not participation

版本C · 极稀疏型(声音以空间为主)

适用:孤立人物、远景、空旷场景、段落过渡
orchestral, near-silence as primary material,
single breathy woodwind tone, sustained then disappearing,
string drone almost below perception threshold,
each note is surrounded by silence on all sides,
phrase length: 2-4 seconds of sound, 8-12 seconds of silence,
no harmonic progression — one pitch center that never resolves,
the space between notes is not empty — it is weight,
melancholy through distance not through expression

版本D · 崩解型(结构瓦解,无终止)

适用:毁灭段落、失控时刻、叙事不可逆转折后
orchestral, structural collapse as compositional principle,
phrases that begin but are cut before completion,
no cadence, no resolution, no return to opening material,
string texture fragments into isolated individual notes,
brass enters at wrong moment — displaced, not coordinated,
harmonic language abandons any tonal center,
the music does not describe destruction — it enacts it,
ends without ending — silence that is not peaceful

天气声效(与音乐协同,不作为BGM)

environmental sound as equal narrative layer,
wind pressure increasing in intervals not continuously,
rain impact on hard surfaces — stone, wood, metal — each physically distinct,
thunder as low-frequency event not atmosphere,
no artificial reverb — raw physical sound only,
sound design and score coexist without hierarchy,
natural sound is not background — it is the world

使用规则

  • 默认使用通用底层;根据叙事段落选择对应版本叠加或替换
  • BGM + 声效 + 旁白三轨禁止同时满量
  • 行动激烈段落:BGM可降至10%或静音,保留物理声音
  • 空境镜头期间(叙事阶段符/场景建立符/余韵符):静音或极弱BGM,保留自然环境声,绝对无旁白

★ 空境镜头Prompt库(S-ENV系列)

用途:分镜设计步骤4选用后,直接注入Prompt,不需要FIGURE_PROMPT,不需要PHYSICS_BLOCK
⚠️ 色彩模式强制分支

  • 彩色模式:直接使用下方Prompt原文
  • 黑白模式:在每条S-ENV Prompt 最前面插入黑白底层声明:black and white film still, Akira Kurosawa visual grammar, 35mm film, 替代原文第一行 film still, Akira Kurosawa visual grammar, 35mm film,
    S-ENV Prompt 不含黑白底层,不会自动转为灰度,必须手动替换首行。
    ⚠️ S-ENV焦段:空境镜头均为大景别户外场景,固定使用中长焦(75mm equivalent lens),不走 LENS 推断流程。

S-ENV-01 · 白色上涌积云(秩序期·小节间隔)

film still, Akira Kurosawa visual grammar, 35mm film, telephoto lens,
fixed camera, low-angle upward shot,
massive cumulus tower rising upper-right, light grey-white lit top, grey-blue base,
strong upward movement fills 70% of frame, cloud continues beyond frame edge,
no ground, no figures, no architecture,
medium film grain

S-ENV-02 · 积云压顶(转折点·后果将来)

film still, Akira Kurosawa visual grammar, 35mm film, telephoto lens,
fixed camera, low-angle upward shot,
dark heavy cumulus mass fills 85% of frame, grey-blue base near-black, minimal bright edge at very top,
flat dark underside pressing downward, upper edges irregular and turbulent — mass has weight not geometry,
no ground, no figures,
medium film grain

S-ENV-03 · 神光穿云(命运性时刻)

film still, Akira Kurosawa visual grammar, 35mm film, telephoto lens,
fixed camera, low-angle upward shot,
dark storm cloud mass dominates frame — dense grey-black volume with irregular torn edges,
diffused warm light filters through multiple gaps in cloud cover — fan-shaped spill spreading outward from openings, not a single concentrated beam,
light visible as broad luminous zones between cloud layers, no distinct ray edges,
no ground, no figures,
medium film grain

S-ENV-04 · 极暗底云(毁灭期内部)

film still, Akira Kurosawa visual grammar, 35mm film, telephoto lens,
fixed camera, low-angle upward shot,
near-black cloud base fills lower 60% of frame, bright white cloud mass only visible at very top edge,
light almost extinguished — only a narrow bright band remains,
no figures, no ground,
medium film grain

S-ENV-05 · 弥漫烈日(叙事悬置·时间感停止)

film still, Akira Kurosawa visual grammar, 35mm film, telephoto lens,
fixed camera, slight upward or direct shot,
diffuse pale sun disc centered in flat hazy sky, edges completely soft and undefined,
uniform grey-white field surrounding disc, no cloud texture, no shadow direction anywhere,
neither warm nor cold — absolute suspension,
no ground, no figures,
medium film grain

S-ENV-06 · 草地远山暮色(人物离场·余韵)

film still, Akira Kurosawa visual grammar, 35mm film, telephoto lens,
fixed camera, medium low angle,
warm golden tall grass in sharp focus foreground, occupies lower 30% of frame,
layered mountain ridges receding: near dark green / mid blue-green / far near-monochrome grey-blue,
no sky visible, no human figures, no architecture,
landscape persists long after human drama ends,
medium film grain

S-ENV-07 · 建筑场景建立(晴天·新段落开场)

film still, Akira Kurosawa visual grammar, 35mm film, telephoto lens,
fixed camera, low-angle upward shot,
multi-tiered castle or power structure fills 70% of frame against blue-white cloudy sky,
dark timber frame, white plaster panels, layered ornamental rooflines,
no human figures,
medium film grain

S-ENV-08 · 建筑场景建立(雾中·状态预示)

film still, Akira Kurosawa visual grammar, 35mm film, telephoto lens,
fixed camera, medium or slight upward shot,
tower or castle structure partially dissolved in pale fog,
architecture visible but all edges soft and indistinct, no human figures,
location declared but condition uncertain — suspended between states,
medium film grain

6. 동영상 조립

剪辑合成规则

转场语法

转场具体使用场景禁用场景硬切人物完成动作的瞬间(刀出鞘/门关上/转身完成);连续两个动作镜头之间人物情绪转折点;空境镜头前后划像(Wipe)时间跳跃(同一地点前后两个时间段);章节之间的时间流逝同一场景内的镜头切换溶解(Dissolve)两个时间不连续但情感上连续的镜头之间,用叠化表示状态缓慢过渡动作段落;战斗场景切到完全静止镜头激烈动作段落后立即接静止的空境或人物静立镜头,用静止吸收观众的动能—

节奏规则

  • 空境镜头处理:每个叙事小节结尾插入3-6s空境;空境期间BGM降至10%以下或静音,只保留风声/环境声;空境同时作为视频片段间的人物状态重置帧
  • 结尾处理:最后一个镜头必须是无人的纯景观(S-ENV系列),BGM在该镜头开始后渐弱,最后3秒静音,只有自然环境声

音视频对齐

  • 片尾5秒:BGM静音,只保留自然环境声(风/远处鸟/水流)
  • 战斗/冲突段落:BGM降至10%或完全静音,保留刀剑声/脚步声/火焰声等物理声音;战斗结束后BGM再渐入
  • 旁白(如有):旁白开始前BGM先降至30%,旁白结束后停顿1秒再渐回

剪辑禁忌

  • 禁止在同一镜头内做速度变化(慢动作必须单独一个镜头)
  • 禁止用BGM填满所有沉默——空境镜头、人物静立镜头期间BGM降至极弱或静音
  • BGM/音效/旁白三轨禁止同时全量,至少一轨在任意时刻低于50%
  • 段落终止符(空境镜头)期间绝对无旁白