@ninkch/dsh-artcraft 0.3.2 → 0.3.4

package/skills/artcraft-director-lens/references/director-styles/05_scorsese.md

@@ -0,0 +1,81 @@
+---
+name: artcraft-lens-scorsese
+description: >-
+  导演透镜（ArtCraft 适配版）：马丁·斯科塞斯导演风格 Skill。把该导演的工作方式翻译成 ArtCraft 可执行参数——FilmCraft 时间线与转场、LightCraft 影调分区、SoundCraft 声音规则、EffectCraft 机位。当用户点名该导演、要求该风格，或项目美学圣经引用该透镜时加载。
+---
+
+# 马丁·斯科塞斯导演风格 Skill（ArtCraft 适配版）
+
+## 风格参数 / Style parameters
+
+```yaml
+style_id: scorsese
+lens_kit_mm: [18, 27, 40, 75]   # 18/27 撑稳定器长镜头，让背景始终贴着人物；40 拍自辩的中近景；75 是急推的落点——急推由一支约 27-75mm 的变焦头在 0.5-1 秒内推完，机位不动
+camera:
+  prefer: ["steadicam long take entering a space", "dolly-in that accelerates then stops hard", "snap zoom on a face", "lateral tracking along a bar or table", "locked frontal when the character addresses the lens"]
+  avoid: ["unmotivated handheld shake", "drifting float with no endpoint", "slow drone-like flyover", "symmetrical locked wide", "static coverage of a moving scene"]
+  motion_budget: high
+  height: "眼平；入场长镜头维持 1.5-1.6m 恒定，暴力节拍可掉到腰高"
+shot_size_bias: ["steadicam medium tracking", "close-up on excitement or shame", "wide of the room as a social map", "insert of money/hands/glass", "frontal medium for direct address"]
+composition: ["subject centred with the world streaming past behind", "layered background of people who all have business", "practical light source burning in frame", "over-shoulder into a room the character is not yet allowed in", "frontal symmetrical only for direct address"]
+lighting:
+  key_quality: hard
+  key_ratio: "5:1，方向混乱（多个实用光源互相打架）"
+  motivation: practical
+  signature: "入画且允许过曝的霓虹与灯罩，常驻烟雾让每个光源都有可见光锥"
+palette:
+  dominant: 酒吧红、后厨绿、钠黄街灯、包厢金
+  accent: 白炽过曝的高光点、单一冷蓝（警灯或电视）
+  saturation: high
+  contrast: high
+editing:
+  avg_shot_length_s: "1.5-3 音乐蒙太奇；8-15 崩塌段（无音乐）"
+  transition_bias: ["hard cut on the downbeat", "jump cut", "freeze frame", "whip pan"]
+  coverage_style: "长镜头承担入场，碎切承担上升；同一空间在上升与坠落时用相同机位、不同节奏拍两次"
+sound:
+  music: 场景内可见来源的流行歌（点唱机、车载电台、楼上派对），可被人物关掉；关掉即转折
+  ambience: 三层都市噪音——低频车流、中频人声餐具、高频笑声玻璃警笛
+  dialogue: 重叠、打断、语速快；群戏只需听清一条线，其余是能量
+  silence_use: 崩塌段完全无音乐，只留房间底噪，持续 15 秒以上
+performance:
+  register: 魅力与危险并存；演"我在解释"，不许演"我知道我错了"
+  blocking: 人物不断向房间更深处移动；被排除时用一道门或一群背影挡住他
+aspect_bias: "1.85:1 / 16:9"
+ai_video:
+  clip_duration_s: "2-4 蒙太奇碎片；8-12 入场长镜头"
+  prompt_bias: 把摄影机运动写成有终点的句子（"tracks with him until he stops at the bar, then holds"），不要写"camera moves through the scene"。速度曲线必须显式：slow start, faster middle, hard stop。
+  risk: 模型把"跟拍"做成无止境的漂移，人物与背景脱节，能量变成眩晕。对策是每条 clip 指定一个停止点与停止时的构图，并把背景人物写成"各自有事在做"而不是"人群"。
+negative_prompt_adds: ["handheld jitter", "aimless floating camera", "slow motion", "empty background", "clean even lighting", "modern smartphone screens", "text or watermark", "symmetrical composition"]
+```
+
+## ArtCraft 落点（本插件适配）
+
+### 剪辑（FilmCraft）
+- 单镜均长目标 **"1.5-3 音乐蒙太奇；8-15 崩塌段（无音乐）"s**（montage / 快切（1.2–2.5s 档））→ `timeline.place` 铺轨后按分配修长：`timeline.trim {mode:"ripple"}`、`timeline.rateStretch` 微调
+- 转场偏好：**["hard cut on the downbeat", "jump cut", "freeze frame", "whip pan"]** → `sequence.applyVideoTransition {effect}` / `sequence.setTransition {duration, align}`
+- 覆盖方式："长镜头承担入场，碎切承担上升；同一空间在上升与坠落时用相同机位、不同节奏拍两次"
+- 音画切点：对白/音乐切点参考 SoundCraft `event.beat_detective` 的节拍位置
+
+### 影调（LightCraft / FilmCraft Lumetri）
+- 光比目标 **"5:1，方向混乱（多个实用光源互相打架）"** → LightCraft `light.contrast` + `grading.shadows/highlights` 分离
+- 主色：酒吧红、后厨绿、钠黄街灯、包厢金 → `mixer.{}.*` 定点 + `grading.*` 方向
+- 点缀：白炽过曝的高光点、单一冷蓝（警灯或电视）
+- 饱和/对比：high / high
+- 光线签名："入画且允许过曝的霓虹与灯罩，常驻烟雾让每个光源都有可见光锥"
+- 落点命令：`develop.set control=<id> value=<v>` 逐项；定稿 `preset.create` 固化，整批 `preset.apply`；FilmCraft 侧 `lumetri.setLook` + `presets.save`
+
+### 声音（SoundCraft / FilmCraft）
+- 音乐：场景内可见来源的流行歌（点唱机、车载电台、楼上派对），可被人物关掉；关掉即转折
+- 环境：三层都市噪音——低频车流、中频人声餐具、高频笑声玻璃警笛
+- 对白：重叠、打断、语速快；群戏只需听清一条线，其余是能量
+- 静默：崩塌段完全无音乐，只留房间底噪，持续 15 秒以上
+- 落点命令：`essentialSound.setType` + `essentialSound.generateDucking`（避让对白）+ `essentialSound.autoMatch {target:-14}`；SoundCraft `mix.insert` 挂处理链
+
+### 机位与运动（EffectCraft）
+- 优先：["steadicam long take entering a space", "dolly-in that accelerates then stops hard", "snap zoom on a face", "lateral tracking along a bar or table", "locked frontal when the character addresses the lens"]
+- 避免：["unmotivated handheld shake", "drifting float with no endpoint", "slow drone-like flyover", "symmetrical locked wide", "static coverage of a moving scene"]
+- 运动预算：high
+- 落点命令：`camera.orbit/pan/dolly` + `view.3d.*` 机位；2D 素材用 `layer.*` + `keys.easyEase*` 缓动
+
+### 质检挂钩（artcraft_qc / 人工复核）
+- 逐镜抽帧复核 + `scopes.read` 示波器核对影调是否落在本透镜的光比与色相档位

package/skills/artcraft-director-lens/references/director-styles/06_fellini.md

@@ -0,0 +1,81 @@
+---
+name: artcraft-lens-fellini
+description: >-
+  导演透镜（ArtCraft 适配版）：费德里科·费里尼导演风格 Skill。把该导演的工作方式翻译成 ArtCraft 可执行参数——FilmCraft 时间线与转场、LightCraft 影调分区、SoundCraft 声音规则、EffectCraft 机位。当用户点名该导演、要求该风格，或项目美学圣经引用该透镜时加载。
+---
+
+# 费德里科·费里尼导演风格 Skill（ArtCraft 适配版）
+
+## 风格参数 / Style parameters
+
+```yaml
+style_id: fellini
+lens_kit_mm: [18, 25, 32]   # 18 装下整支队伍与前景穿行；25 做正面近景的漫画式透视夸张（不要用长焦修正）；32 是人群中跟移的默认焦段
+camera:
+  prefer: ["drifting track through a crowd at 0.4-0.8 m/s", "pan handing off from one face to the next", "frontal wide close-up with mild distortion", "crane rising over a procession", "follow that loses the subject and finds another"]
+  avoid: ["locked symmetrical wide", "long-lens compression", "shallow-focus isolation", "static shot-reverse-shot", "drone-style detached flyover"]
+  motion_budget: high
+  height: "眼平游动为主；垂直事件时升到 2.5-4m 让人群变成图案"
+shot_size_bias: ["wide of the whole procession", "frontal close-up of a single strange face", "medium tracking through bodies", "over-shoulder past a foreground passer-by", "high angle turning the crowd into a pattern"]
+composition: ["figures entering continuously from both frame edges", "foreground passers-by occupying 10-30%", "reframed every 2-3 seconds, never settled", "one vertical event lifting someone above the crowd", "direct address to the lens"]
+lighting:
+  key_quality: mixed
+  key_ratio: "4:1，阴影保留细节"
+  motivation: stylized
+  signature: "硬边舞台光区被常驻的雾柔化半米；人物走进走出光区，明暗切换本身是表演"
+palette:
+  dominant: 粉、金、酒红、天蓝
+  accent: 纯白与漆黑的服装对撞
+  saturation: high
+  contrast: mid
+editing:
+  avg_shot_length_s: "6-14；结尾段拉到 15-30"
+  transition_bias: ["hard cut on a musical phrase end", "match cut on a recurring image", "dissolve"]
+  coverage_style: "不做正反打；一个游动镜头交接多个人物，离题镜头占段落 15-25%"
+sound:
+  music: 圆舞曲、进行曲、马戏团铜管——先于动作确定节拍，人物跟着音乐走路
+  ambience: 允许非写实：笑声、叫卖、祈祷、海浪出现在物理上不该出现的地方
+  dialogue: 互相压过、听不清，是人群的质地；重要的话在音乐间隙由一个人单独说出
+  silence_use: 极少；只用在某个意象第二次回来时，留一种声音（风、海、铃）
+performance:
+  register: 夸张、怪诞、可笑又可怜；选角标准是"一张脸能说完一个故事"
+  blocking: 主角站着不动被环绕，人群绕他运动；群众演员每人有一句话身份与一个正在做的动作
+aspect_bias: "1.85:1；黑白段落可用 1.37:1"   # 9:16 交付时的构图改写见「镜头语言」末条
+ai_video:
+  clip_duration_s: "6-12"
+  prompt_bias: 先写人群的行为清单（谁在做什么），再写摄影机；把"持续有人从画面边缘进出"写成硬要求，把音乐节拍写进动作描述（walking on the beat of a waltz）。
+  risk: 模型会把人群做成对称、同步、面目模糊的背景板，怪异感消失。对策是点名 3-5 个具体人物及其各自动作与服装，并写 no two people doing the same thing、faces distinct and exaggerated。
+negative_prompt_adds: ["symmetrical composition", "synchronised crowd", "blurred faceless background people", "shallow depth of field", "long-lens compression", "cold desaturated palette", "text or watermark", "empty space"]
+```
+
+## ArtCraft 落点（本插件适配）
+
+### 剪辑（FilmCraft）
+- 单镜均长目标 **"6-14；结尾段拉到 15-30"s**（conventional（5–12s 档））→ `timeline.place` 铺轨后按分配修长：`timeline.trim {mode:"ripple"}`、`timeline.rateStretch` 微调
+- 转场偏好：**["hard cut on a musical phrase end", "match cut on a recurring image", "dissolve"]** → `sequence.applyVideoTransition {effect}` / `sequence.setTransition {duration, align}`
+- 覆盖方式："不做正反打；一个游动镜头交接多个人物，离题镜头占段落 15-25%"
+- 音画切点：对白/音乐切点参考 SoundCraft `event.beat_detective` 的节拍位置
+
+### 影调（LightCraft / FilmCraft Lumetri）
+- 光比目标 **"4:1，阴影保留细节"** → LightCraft `light.contrast` + `grading.shadows/highlights` 分离
+- 主色：粉、金、酒红、天蓝 → `mixer.{}.*` 定点 + `grading.*` 方向
+- 点缀：纯白与漆黑的服装对撞
+- 饱和/对比：high / mid
+- 光线签名："硬边舞台光区被常驻的雾柔化半米；人物走进走出光区，明暗切换本身是表演"
+- 落点命令：`develop.set control=<id> value=<v>` 逐项；定稿 `preset.create` 固化，整批 `preset.apply`；FilmCraft 侧 `lumetri.setLook` + `presets.save`
+
+### 声音（SoundCraft / FilmCraft）
+- 音乐：圆舞曲、进行曲、马戏团铜管——先于动作确定节拍，人物跟着音乐走路
+- 环境：允许非写实：笑声、叫卖、祈祷、海浪出现在物理上不该出现的地方
+- 对白：互相压过、听不清，是人群的质地；重要的话在音乐间隙由一个人单独说出
+- 静默：极少；只用在某个意象第二次回来时，留一种声音（风、海、铃）
+- 落点命令：`essentialSound.setType` + `essentialSound.generateDucking`（避让对白）+ `essentialSound.autoMatch {target:-14}`；SoundCraft `mix.insert` 挂处理链
+
+### 机位与运动（EffectCraft）
+- 优先：["drifting track through a crowd at 0.4-0.8 m/s", "pan handing off from one face to the next", "frontal wide close-up with mild distortion", "crane rising over a procession", "follow that loses the subject and finds another"]
+- 避免：["locked symmetrical wide", "long-lens compression", "shallow-focus isolation", "static shot-reverse-shot", "drone-style detached flyover"]
+- 运动预算：high
+- 落点命令：`camera.orbit/pan/dolly` + `view.3d.*` 机位；2D 素材用 `layer.*` + `keys.easyEase*` 缓动
+
+### 质检挂钩（artcraft_qc / 人工复核）
+- 逐镜抽帧复核 + `scopes.read` 示波器核对影调是否落在本透镜的光比与色相档位

package/skills/artcraft-director-lens/references/director-styles/07_bergman.md

@@ -0,0 +1,81 @@
+---
+name: artcraft-lens-bergman
+description: >-
+  导演透镜（ArtCraft 适配版）：英格玛·伯格曼导演风格 Skill。把该导演的工作方式翻译成 ArtCraft 可执行参数——FilmCraft 时间线与转场、LightCraft 影调分区、SoundCraft 声音规则、EffectCraft 机位。当用户点名该导演、要求该风格，或项目美学圣经引用该透镜时加载。
+---
+
+# 英格玛·伯格曼导演风格 Skill（ArtCraft 适配版）
+
+## 风格参数 / Style parameters
+
+```yaml
+style_id: bergman
+lens_kit_mm: [75, 100, 150]   # 75 中近景；100/150 拍大特写，长焦不给脸任何变形借口；单脸收到 f/11-f/16 让整张脸落在景深内，双脸构图开到 f/2.8-f/4 留给变焦点
+camera:
+  prefer: ["locked frontal close-up on the eyeline axis", "two-face composition with one profile in foreground", "focus pull between two faces instead of a cut", "imperceptible push-in at 0.02-0.05 m/s", "hold past the emotional peak"]
+  avoid: ["handheld", "shot-reverse-shot on fixed A/B setups", "camera orbit", "wide establishing of the room", "any move the audience can notice"]
+  motion_budget: low   # 实际接近 none：唯一允许的运动是观众察觉不到的推进
+  height: "眼平，精确对齐演员瞳孔高度；全场不变"
+shot_size_bias: ["extreme close-up filling 60-80% of frame height", "two-face composition", "medium close-up at 75mm", "hands or object insert", "one wide only if the room must be established once"]
+composition: ["camera on the eyeline axis, deviation under 10 degrees", "two faces overlapping at their edges, one frontal one profile", "near-black or near-white flat background", "negative space on the side the character is avoiding", "whole face inside depth of field"]
+lighting:
+  key_quality: hard
+  key_ratio: "6:1，暗面不补光但保留轮廓"
+  motivation: practical
+  signature: "正侧方 90 度单光源，鼻影把脸分成两半；背景压成单色平面"
+palette:
+  dominant: 灰蓝、灰绿、近黑、近白
+  accent: 皮肤的暖色是画面里唯一的暖
+  saturation: low
+  contrast: high
+editing:
+  avg_shot_length_s: "20-90；一场 4 分钟的戏只用 5-8 个镜头"
+  transition_bias: ["hard cut", "fade"]
+  coverage_style: "不做标准正反打；用画内移动与变焦点代替切换，情绪高点之后再拍 3-5 秒"
+sound:
+  music: 极少；只在场景之间的空白处，不与对白同时存在
+  ambience: 单一持续元素（钟摆、海浪、风穿窗缝），让沉默可被计量
+  dialogue: 干、近、无混响；呼吸与吞咽必须清晰可闻
+  silence_use: 每场指定 2-3 处 4-8 秒沉默，并写清沉默时镜头在拍谁的脸
+performance:
+  register: 高度抑制；最高强度是"没有哭出来但脸正在塌"，指令写成抑制目标
+  blocking: 人物在画内起身、转身、靠近窗；谁离镜头近谁在这一刻更有权力
+aspect_bias: "1.37:1 / 4:3——接近方形的画幅让脸在竖向被填满；16:9 会在脸两侧留下无用空间"
+ai_video:
+  clip_duration_s: "8-15（在工具能稳住身份的前提下取最长）"
+  prompt_bias: 一条 clip 只写一次微表情变化 + 一次呼吸事件，摄影机写成 locked 或 imperceptible push-in；显式写 the whole face stays in focus。唯一例外是双脸变焦点那一条——它要从 negative 里拿掉 shallow depth of field，改写成 focus shifts from the front face to the face behind over 1.5 seconds。20 秒以上的长镜头用同参数分段续接拼出来，接点选在动作完全静止的那一帧。
+  risk: 模型会给静止的脸补上多余动作（转头、眨眼过频、微笑）并把背景虚化成柔焦氛围，抑制感立刻消失。对策是写 no head turn, blinks only once, background flat and dark, no bokeh，并把 clip 控制在一次心理变化点之内。
+negative_prompt_adds: ["shallow depth of field or bokeh", "soft glamour lighting", "head turning away", "smiling", "camera movement", "warm colour grade", "background detail or props", "text or watermark"]
+```
+
+## ArtCraft 落点（本插件适配）
+
+### 剪辑（FilmCraft）
+- 单镜均长目标 **"20-90；一场 4 分钟的戏只用 5-8 个镜头"s**（observational / 长镜（8–15s 档））→ `timeline.place` 铺轨后按分配修长：`timeline.trim {mode:"ripple"}`、`timeline.rateStretch` 微调
+- 转场偏好：**["hard cut", "fade"]** → `sequence.applyVideoTransition {effect}` / `sequence.setTransition {duration, align}`
+- 覆盖方式："不做标准正反打；用画内移动与变焦点代替切换，情绪高点之后再拍 3-5 秒"
+- 音画切点：对白/音乐切点参考 SoundCraft `event.beat_detective` 的节拍位置
+
+### 影调（LightCraft / FilmCraft Lumetri）
+- 光比目标 **"6:1，暗面不补光但保留轮廓"** → LightCraft `light.contrast` + `grading.shadows/highlights` 分离
+- 主色：灰蓝、灰绿、近黑、近白 → `mixer.{}.*` 定点 + `grading.*` 方向
+- 点缀：皮肤的暖色是画面里唯一的暖
+- 饱和/对比：low / high
+- 光线签名："正侧方 90 度单光源，鼻影把脸分成两半；背景压成单色平面"
+- 落点命令：`develop.set control=<id> value=<v>` 逐项；定稿 `preset.create` 固化，整批 `preset.apply`；FilmCraft 侧 `lumetri.setLook` + `presets.save`
+
+### 声音（SoundCraft / FilmCraft）
+- 音乐：极少；只在场景之间的空白处，不与对白同时存在
+- 环境：单一持续元素（钟摆、海浪、风穿窗缝），让沉默可被计量
+- 对白：干、近、无混响；呼吸与吞咽必须清晰可闻
+- 静默：每场指定 2-3 处 4-8 秒沉默，并写清沉默时镜头在拍谁的脸
+- 落点命令：`essentialSound.setType` + `essentialSound.generateDucking`（避让对白）+ `essentialSound.autoMatch {target:-14}`；SoundCraft `mix.insert` 挂处理链
+
+### 机位与运动（EffectCraft）
+- 优先：["locked frontal close-up on the eyeline axis", "two-face composition with one profile in foreground", "focus pull between two faces instead of a cut", "imperceptible push-in at 0.02-0.05 m/s", "hold past the emotional peak"]
+- 避免：["handheld", "shot-reverse-shot on fixed A/B setups", "camera orbit", "wide establishing of the room", "any move the audience can notice"]
+- 运动预算：low
+- 落点命令：`camera.orbit/pan/dolly` + `view.3d.*` 机位；2D 素材用 `layer.*` + `keys.easyEase*` 缓动
+
+### 质检挂钩（artcraft_qc / 人工复核）
+- 逐镜抽帧复核 + `scopes.read` 示波器核对影调是否落在本透镜的光比与色相档位

package/skills/artcraft-director-lens/references/director-styles/08_tarkovsky.md

@@ -0,0 +1,81 @@
+---
+name: artcraft-lens-tarkovsky
+description: >-
+  导演透镜（ArtCraft 适配版）：安德烈·塔可夫斯基导演风格 Skill。把该导演的工作方式翻译成 ArtCraft 可执行参数——FilmCraft 时间线与转场、LightCraft 影调分区、SoundCraft 声音规则、EffectCraft 机位。当用户点名该导演、要求该风格，或项目美学圣经引用该透镜时加载。
+---
+
+# 安德烈·塔可夫斯基导演风格 Skill（ArtCraft 适配版）
+
+## 风格参数 / Style parameters
+
+```yaml
+style_id: tarkovsky
+lens_kit_mm: [28, 35, 50]   # 广角到标准，深焦分层；靠景深而非切镜建立层次
+camera:
+  prefer: [locked hold, very slow lateral dolly, slow push-in on an object, slow tilt down to water, walking-pace follow from behind]
+  avoid: [handheld shake, whip pan, orbit, crash zoom, drone reveal]
+  motion_budget: low          # 10 秒内位移 ≤ 1 米，恒速，无加减速
+  height: "waist to chest, level or slightly low, planted like a witness"
+shot_size_bias: [wide, medium wide, insert or detail, over-shoulder from behind, close-up]
+composition: [deep staging with three or more planes, reflective foreground, off-center or back-turned figure, vertical layering from ceiling to floor, threshold framing]
+lighting:
+  key_quality: soft
+  key_ratio: "2:1 to 3:1"
+  motivation: naturalistic
+  signature: "阴天窗光加单一火光点；无硬边阴影，湿表面自己发光"
+palette:
+  dominant: "mud brown, wet stone grey, moss green, sepia"
+  accent: "fire orange, milk white, rust"
+  saturation: low
+  contrast: low
+editing:
+  avg_shot_length_s: "30-90"
+  transition_bias: [hard cut, slow dissolve, fade]
+  coverage_style: "单机位单角度，一个镜头解决一场戏，不拍反打"
+sound:
+  music: "极少，出现时为宗教性合唱或单一持续音，不跟画面情绪走"
+  ambience: "可数的滴水、风穿空屋、火焰爆裂、远处狗叫；环境声即主音轨"
+  dialogue: "低音量慢速，常在画外或背对镜头说出，每场 40 字以内"
+  silence_use: "连续 5-15 秒静默，用来让观众听出空间尺寸"
+performance:
+  register: "内敛疲惫，情绪落后事件 2-3 秒，反应幅度小于事实要求"
+  blocking: "长时间静止或缓慢走动，常背对镜头或被前景遮挡"
+aspect_bias: "1.66:1 / 4:3"   # 竖向余量留给天花板、树冠和垂落的水
+ai_video:
+  clip_duration_s: "8-15"     # 取所用工具允许的最长单条时长；宁可一个长镜头，不要三个短镜头
+  prompt_bias: "主体动作写到最小（呼吸、转头、放下一件东西），描述预算全部给环境运动和材质：水、火、雾、布料、光的缓慢变化。"
+  risk: "模型把'慢'理解成'静止'，输出一张会呼吸的照片。缓解：显式命名两个持续的环境运动源，并给镜头一个恒速微移。"
+negative_prompt_adds: [saturated colors, hard-edged shadows, lens flare, fast camera move, quick cuts, glossy clean surfaces, stage-symmetric framing, orchestral swell]
+```
+
+## ArtCraft 落点（本插件适配）
+
+### 剪辑（FilmCraft）
+- 单镜均长目标 **"30-90"s**（observational / 长镜（8–15s 档））→ `timeline.place` 铺轨后按分配修长：`timeline.trim {mode:"ripple"}`、`timeline.rateStretch` 微调
+- 转场偏好：**[hard cut, slow dissolve, fade]** → `sequence.applyVideoTransition {effect}` / `sequence.setTransition {duration, align}`
+- 覆盖方式："单机位单角度，一个镜头解决一场戏，不拍反打"
+- 音画切点：对白/音乐切点参考 SoundCraft `event.beat_detective` 的节拍位置
+
+### 影调（LightCraft / FilmCraft Lumetri）
+- 光比目标 **"2:1 to 3:1"** → LightCraft `light.contrast` + `grading.shadows/highlights` 分离
+- 主色："mud brown, wet stone grey, moss green, sepia" → `mixer.{green}.*` 定点 + `grading.*` 方向
+- 点缀："fire orange, milk white, rust"
+- 饱和/对比：low / low
+- 光线签名："阴天窗光加单一火光点；无硬边阴影，湿表面自己发光"
+- 落点命令：`develop.set control=<id> value=<v>` 逐项；定稿 `preset.create` 固化，整批 `preset.apply`；FilmCraft 侧 `lumetri.setLook` + `presets.save`
+
+### 声音（SoundCraft / FilmCraft）
+- 音乐："极少，出现时为宗教性合唱或单一持续音，不跟画面情绪走"
+- 环境："可数的滴水、风穿空屋、火焰爆裂、远处狗叫；环境声即主音轨"
+- 对白："低音量慢速，常在画外或背对镜头说出，每场 40 字以内"
+- 静默："连续 5-15 秒静默，用来让观众听出空间尺寸"
+- 落点命令：`essentialSound.setType` + `essentialSound.generateDucking`（避让对白）+ `essentialSound.autoMatch {target:-14}`；SoundCraft `mix.insert` 挂处理链
+
+### 机位与运动（EffectCraft）
+- 优先：[locked hold, very slow lateral dolly, slow push-in on an object, slow tilt down to water, walking-pace follow from behind]
+- 避免：[handheld shake, whip pan, orbit, crash zoom, drone reveal]
+- 运动预算：low
+- 落点命令：`camera.orbit/pan/dolly` + `view.3d.*` 机位；2D 素材用 `layer.*` + `keys.easyEase*` 缓动
+
+### 质检挂钩（artcraft_qc / 人工复核）
+- 逐镜抽帧复核 + `scopes.read` 示波器核对影调是否落在本透镜的光比与色相档位

package/skills/artcraft-director-lens/references/director-styles/09_wong_kar_wai.md

@@ -0,0 +1,81 @@
+---
+name: artcraft-lens-wong_kar_wai
+description: >-
+  导演透镜（ArtCraft 适配版）：王家卫导演风格 Skill。把该导演的工作方式翻译成 ArtCraft 可执行参数——FilmCraft 时间线与转场、LightCraft 影调分区、SoundCraft 声音规则、EffectCraft 机位。当用户点名该导演、要求该风格，或项目美学圣经引用该透镜时加载。
+---
+
+# 王家卫导演风格 Skill（ArtCraft 适配版）
+
+## 风格参数 / Style parameters
+
+```yaml
+style_id: wong_kar_wai
+lens_kit_mm: [18, 25, 85]   # 广角贴身拍变形的亲密，85mm 隔街压缩人流；中焦段刻意回避
+camera:
+  prefer: [small-amplitude handheld, step-print slow motion, reframe-in-shot, shoot through foreground obstruction, low-angle from a corner]
+  avoid: [locked tripod, symmetric center framing, smooth gimbal glide, drone, full clean wide]
+  motion_budget: medium       # 频率高、幅度小：几乎每秒都在微调，但整镜位移 < 1 米。写提示词时给的是"重新构图"，不是"运动"
+  height: "chest to shoulder, often from a corner or across an obstruction, rarely level with the eyes"
+shot_size_bias: [medium close-up, close-up of hands and objects, obstructed medium, over-shoulder, extreme close-up]
+composition: [foreground obstruction in every frame, subject pushed to frame edge, frame within frame via door or window, mirror or glass reflection, head cropped at top]
+lighting:
+  key_quality: hard
+  key_ratio: "6:1 to 10:1"
+  motivation: practical
+  signature: "裸灯泡与招牌直射给硬边和高光比，霓虹管只负责染色勾边；空气里有烟和水汽，光柱可见"
+palette:
+  dominant: "neon green, deep red, tungsten amber"
+  accent: "magenta, jade, gold"
+  saturation: high
+  contrast: high
+editing:
+  avg_shot_length_s: "1.5-4"   # 每段允许一个 6-12 秒的慢动作长挂
+  transition_bias: [hard cut, jump cut, whip pan, dissolve]
+  coverage_style: "同一动作跳接三次，位置微移；关键进展留在两个镜头之间"
+sound:
+  music: "一首歌即一个主题，全片出现 2-3 次，每次对应关系的一次退步"
+  ambience: "雨、电话铃、收音机、餐馆碗碟、楼下摩托、隔壁电视"
+  dialogue: "短、跳跃、答非所问；被环境声盖掉一半是可接受的"
+  silence_use: "极少完全静默；用歌声顶替静默，让人物在音乐里失语"
+performance:
+  register: "克制迷人，台词与意图错开一层，不说真正的需求"
+  blocking: "两人同框不得正对；一个背对或隔玻璃，距离以厘米计量"
+aspect_bias: "1.85:1"        # 窄边框加强前景遮挡的压迫；9:16 竖幅同样成立
+ai_video:
+  clip_duration_s: "4-8"      # 短，靠碎片累积；慢动作段落单独出一条
+  prompt_bias: "先写前景挡什么，再写霓虹的两种颜色，最后才写人做什么；动作只留一个，且必须是'收回'而不是'伸出'。"
+  risk: "模型把手持理解成整体抖动，把慢动作理解成平滑插帧的升格。缓解：写 'small handheld reframing, not shake'；步进印片要同时点名两件事——'long-shutter trails behind the moving hand' 和 'repeated-frame stutter, not smooth slow motion'，只写其一都拿不到；再锁死一个静止的前景遮挡物做位移参照。"
+negative_prompt_adds: [locked tripod stillness, symmetric center framing, even soft fill, desaturated palette, clean empty space, smooth gimbal glide, full frontal face, clear unobstructed view]
+```
+
+## ArtCraft 落点（本插件适配）
+
+### 剪辑（FilmCraft）
+- 单镜均长目标 **"1.5-4"s**（tightening / 紧凑对话（3–6s 档））→ `timeline.place` 铺轨后按分配修长：`timeline.trim {mode:"ripple"}`、`timeline.rateStretch` 微调
+- 转场偏好：**[hard cut, jump cut, whip pan, dissolve]** → `sequence.applyVideoTransition {effect}` / `sequence.setTransition {duration, align}`
+- 覆盖方式："同一动作跳接三次，位置微移；关键进展留在两个镜头之间"
+- 音画切点：对白/音乐切点参考 SoundCraft `event.beat_detective` 的节拍位置
+
+### 影调（LightCraft / FilmCraft Lumetri）
+- 光比目标 **"6:1 to 10:1"** → LightCraft `light.contrast` + `grading.shadows/highlights` 分离
+- 主色："neon green, deep red, tungsten amber" → `mixer.{green,magenta,red,orange,aqua}.*` 定点 + `grading.*` 方向
+- 点缀："magenta, jade, gold"
+- 饱和/对比：high / high
+- 光线签名："裸灯泡与招牌直射给硬边和高光比，霓虹管只负责染色勾边；空气里有烟和水汽，光柱可见"
+- 落点命令：`develop.set control=<id> value=<v>` 逐项；定稿 `preset.create` 固化，整批 `preset.apply`；FilmCraft 侧 `lumetri.setLook` + `presets.save`
+
+### 声音（SoundCraft / FilmCraft）
+- 音乐："一首歌即一个主题，全片出现 2-3 次，每次对应关系的一次退步"
+- 环境："雨、电话铃、收音机、餐馆碗碟、楼下摩托、隔壁电视"
+- 对白："短、跳跃、答非所问；被环境声盖掉一半是可接受的"
+- 静默："极少完全静默；用歌声顶替静默，让人物在音乐里失语"
+- 落点命令：`essentialSound.setType` + `essentialSound.generateDucking`（避让对白）+ `essentialSound.autoMatch {target:-14}`；SoundCraft `mix.insert` 挂处理链
+
+### 机位与运动（EffectCraft）
+- 优先：[small-amplitude handheld, step-print slow motion, reframe-in-shot, shoot through foreground obstruction, low-angle from a corner]
+- 避免：[locked tripod, symmetric center framing, smooth gimbal glide, drone, full clean wide]
+- 运动预算：medium
+- 落点命令：`camera.orbit/pan/dolly` + `view.3d.*` 机位；2D 素材用 `layer.*` + `keys.easyEase*` 缓动
+
+### 质检挂钩（artcraft_qc / 人工复核）
+- 逐镜抽帧复核 + `scopes.read` 示波器核对影调是否落在本透镜的光比与色相档位

package/skills/artcraft-director-lens/references/director-styles/10_nolan.md

@@ -0,0 +1,81 @@
+---
+name: artcraft-lens-nolan
+description: >-
+  导演透镜（ArtCraft 适配版）：克里斯托弗·诺兰导演风格 Skill。把该导演的工作方式翻译成 ArtCraft 可执行参数——FilmCraft 时间线与转场、LightCraft 影调分区、SoundCraft 声音规则、EffectCraft 机位。当用户点名该导演、要求该风格，或项目美学圣经引用该透镜时加载。
+---
+
+# 克里斯托弗·诺兰导演风格 Skill（ArtCraft 适配版）
+
+## 风格参数 / Style parameters
+
+```yaml
+style_id: nolan
+lens_kit_mm: [14, 21, 35, 50]   # 广角为主：14-21 给尺度，21 贴到 60-90cm 给物理重量，50 是长焦上限
+camera:
+  prefer: [controlled handheld follow, wide tracking alongside action, slow push-in on a mechanism, crane up to reveal scale, locked insert on an instrument]
+  avoid: [orbit for its own sake, whip pan, slow motion for beauty, dreamy diffusion, drone flythrough]
+  motion_budget: high         # 镜头几乎总在移动，但移动是"跟上"，不是表演
+  height: "eye level, camera close to the body, occasionally low to give machinery mass"
+shot_size_bias: [medium close-up, wide tracking, extreme wide for scale, mechanism insert, over-shoulder]
+composition: [figure against oversized real structure, horizon low to give sky weight, mechanism readable in frame, two timelines matched by shape, foreground hardware giving depth]
+lighting:
+  key_quality: mixed
+  key_ratio: "3:1 to 4:1"
+  motivation: naturalistic
+  signature: "真实天光加可见的工作灯；夜景只有钠灯暖橙一个暖点"
+palette:
+  dominant: "concrete grey, steel blue, black, canvas khaki"
+  accent: "sodium amber, warning red"
+  saturation: low
+  contrast: mid
+editing:
+  avg_shot_length_s: "2-5"
+  transition_bias: [hard cut, match cut, cut on sound]
+  coverage_style: "交叉剪辑为结构；轮转周期从 12 秒收紧到 3 秒，每次切线带走一个新信息"
+sound:
+  music: "持续低频脉冲与无解决的上升音型，功能是报时而非煽情"
+  ambience: "金属应力、气压、低频冲击、按空间尺寸变化的混响"
+  dialogue: "规则说明必须在动作压力下说出，禁止坐下来解释"
+  silence_use: "只在归零前一拍使用，长度 0.5-1.5 秒，用来放大下一次冲击"
+performance:
+  register: "理性克制，情绪藏在选择里；崩溃上限是一次呼吸失控"
+  blocking: "人物永远在执行一件有时限的任务，走位由机制而非情绪决定"
+aspect_bias: "1.90:1 / 2.20:1"   # 比 2.39:1 高的画幅：同样的画面宽度下多出的纵向余量留给天空和头顶，尺度靠竖向撑起来
+ai_video:
+  clip_duration_s: "3-6"       # 每条线单独生成，交叉在剪辑台完成，不要求模型在一条 clip 里切线
+  prompt_bias: "一条 clip 只讲一个机制事实：一个动作、一个镜头运动、一个可见的计时器变化。把'尺度'翻译成具体材料与尺寸，不要写'宏大'。"
+  risk: "模型把'高概念'渲染成发光全息与赛博界面，并把交叉剪辑压成一镜。缓解：负面词里点名 holographic UI 与 neon；三条线各出一条 clip，用同一组材料词、同一色温、同一颗粒描述锁一致性。"
+negative_prompt_adds: [glowing holographic interface, floating blue data, neon signage, obvious CGI sheen, slow motion, dreamy diffusion, saturated color grade, lens flare]
+```
+
+## ArtCraft 落点（本插件适配）
+
+### 剪辑（FilmCraft）
+- 单镜均长目标 **"2-5"s**（tightening / 紧凑对话（3–6s 档））→ `timeline.place` 铺轨后按分配修长：`timeline.trim {mode:"ripple"}`、`timeline.rateStretch` 微调
+- 转场偏好：**[hard cut, match cut, cut on sound]** → `sequence.applyVideoTransition {effect}` / `sequence.setTransition {duration, align}`
+- 覆盖方式："交叉剪辑为结构；轮转周期从 12 秒收紧到 3 秒，每次切线带走一个新信息"
+- 音画切点：对白/音乐切点参考 SoundCraft `event.beat_detective` 的节拍位置
+
+### 影调（LightCraft / FilmCraft Lumetri）
+- 光比目标 **"3:1 to 4:1"** → LightCraft `light.contrast` + `grading.shadows/highlights` 分离
+- 主色："concrete grey, steel blue, black, canvas khaki" → `mixer.{red,orange,blue}.*` 定点 + `grading.*` 方向
+- 点缀："sodium amber, warning red"
+- 饱和/对比：low / mid
+- 光线签名："真实天光加可见的工作灯；夜景只有钠灯暖橙一个暖点"
+- 落点命令：`develop.set control=<id> value=<v>` 逐项；定稿 `preset.create` 固化，整批 `preset.apply`；FilmCraft 侧 `lumetri.setLook` + `presets.save`
+
+### 声音（SoundCraft / FilmCraft）
+- 音乐："持续低频脉冲与无解决的上升音型，功能是报时而非煽情"
+- 环境："金属应力、气压、低频冲击、按空间尺寸变化的混响"
+- 对白："规则说明必须在动作压力下说出，禁止坐下来解释"
+- 静默："只在归零前一拍使用，长度 0.5-1.5 秒，用来放大下一次冲击"
+- 落点命令：`essentialSound.setType` + `essentialSound.generateDucking`（避让对白）+ `essentialSound.autoMatch {target:-14}`；SoundCraft `mix.insert` 挂处理链
+
+### 机位与运动（EffectCraft）
+- 优先：[controlled handheld follow, wide tracking alongside action, slow push-in on a mechanism, crane up to reveal scale, locked insert on an instrument]
+- 避免：[orbit for its own sake, whip pan, slow motion for beauty, dreamy diffusion, drone flythrough]
+- 运动预算：high
+- 落点命令：`camera.orbit/pan/dolly` + `view.3d.*` 机位；2D 素材用 `layer.*` + `keys.easyEase*` 缓动
+
+### 质检挂钩（artcraft_qc / 人工复核）
+- 逐镜抽帧复核 + `scopes.read` 示波器核对影调是否落在本透镜的光比与色相档位

package/skills/artcraft-director-lens/references/director-styles/11_villeneuve.md

@@ -0,0 +1,81 @@
+---
+name: artcraft-lens-villeneuve
+description: >-
+  导演透镜（ArtCraft 适配版）：丹尼斯·维伦纽瓦导演风格 Skill。把该导演的工作方式翻译成 ArtCraft 可执行参数——FilmCraft 时间线与转场、LightCraft 影调分区、SoundCraft 声音规则、EffectCraft 机位。当用户点名该导演、要求该风格，或项目美学圣经引用该透镜时加载。
+---
+
+# 丹尼斯·维伦纽瓦导演风格 Skill（ArtCraft 适配版）
+
+## 风格参数 / Style parameters
+
+```yaml
+style_id: villeneuve
+lens_kit_mm: [21, 40, 75]   # 21 给纪念碑尺度，40 给程序化中景，75 压缩疲惫的脸；不用超广也不用超长
+camera:
+  prefer: [locked wide, constant-speed lateral track, constant-speed push toward vanishing point, high overhead, slow tilt up a structure]
+  avoid: [handheld, orbit, whip pan, speed ramp, crash zoom]
+  motion_budget: low          # 有移动但恒速无表情，像在丈量空间
+  height: "eye level or extreme overhead; nothing in between"
+shot_size_bias: [extreme wide, wide, overhead, medium close-up, close-up]
+composition: [center symmetry, single strong horizon at one-eighth or seven-eighths height, human at 5-15 percent of frame height, silhouette against a lit plane, negative space as the subject]
+lighting:
+  key_quality: soft
+  key_ratio: "5:1 to 8:1"     # 逆光面对正面补光，不是常规主光对补光
+  motivation: naturalistic
+  signature: "整面墙或整片天当光源的逆光；人物半剪影，无硬边阴影"
+palette:
+  dominant: "one hue per sequence: ochre sand, teal fog, copper interior, ink green, grey-white ice"
+  accent: "a single small warm point, usually a lamp or a fire"
+  saturation: low
+  contrast: mid
+editing:
+  avg_shot_length_s: "8-18"
+  transition_bias: [hard cut, fade, dissolve]
+  coverage_style: "一场 3-9 个镜头；人物进画前留 2 秒，出画后留 3 秒；事件后必给余波镜头"
+sound:
+  music: "40-80Hz 轰鸣、持续合成音、共振铜管；无旋律记忆点，只有质地与压强"
+  ambience: "分层的风、沙粒打在织物上、呼吸、远处金属应力"
+  dialogue: "被环境声压低，关键台词只说一次，'听不清'是主题的一部分"
+  silence_use: "全段 20-40% 的时长无音乐；低频比画面早进 2-4 秒"
+performance:
+  register: "内敛专业疲惫，情绪靠呼吸与姿态的微小塌陷"
+  blocking: "人物穿过巨大空间，方向单一，速度恒定；停下来的那一次就是转折"
+aspect_bias: "2.39:1"        # 宽画幅强化水平线与横向丈量；9:16 交付时的构图改写见「镜头语言」末条
+ai_video:
+  clip_duration_s: "8-15"     # 越长越好；给模型时间把雾的层次渲染出来
+  prompt_bias: "先写几何体与材料，再写雾从第几米开始，最后才写人；人的描述不超过一句，且必须给出他占画面高度的比例。"
+  risk: "模型把'宏大'渲染成拥挤的赛博城市，或把雾渲染成整体模糊。缓解：负面词点名 neon 与 holographic；把雾写成 depth gradient（近处清晰、30 米外淡 2-3 档），并给一个已知尺寸的参照物。"
+negative_prompt_adds: [holographic interface, neon signage, cluttered cyberpunk street, hard-edged shadows, handheld shake, split warm-cool grade, busy background detail, visible lens flare]
+```
+
+## ArtCraft 落点（本插件适配）
+
+### 剪辑（FilmCraft）
+- 单镜均长目标 **"8-18"s**（observational / 长镜（8–15s 档））→ `timeline.place` 铺轨后按分配修长：`timeline.trim {mode:"ripple"}`、`timeline.rateStretch` 微调
+- 转场偏好：**[hard cut, fade, dissolve]** → `sequence.applyVideoTransition {effect}` / `sequence.setTransition {duration, align}`
+- 覆盖方式："一场 3-9 个镜头；人物进画前留 2 秒，出画后留 3 秒；事件后必给余波镜头"
+- 音画切点：对白/音乐切点参考 SoundCraft `event.beat_detective` 的节拍位置
+
+### 影调（LightCraft / FilmCraft Lumetri）
+- 光比目标 **"5:1 to 8:1"** → LightCraft `light.contrast` + `grading.shadows/highlights` 分离
+- 主色："one hue per sequence: ochre sand, teal fog, copper interior, ink green, grey-white ice" → `mixer.{green,aqua}.*` 定点 + `grading.*` 方向
+- 点缀："a single small warm point, usually a lamp or a fire"
+- 饱和/对比：low / mid
+- 光线签名："整面墙或整片天当光源的逆光；人物半剪影，无硬边阴影"
+- 落点命令：`develop.set control=<id> value=<v>` 逐项；定稿 `preset.create` 固化，整批 `preset.apply`；FilmCraft 侧 `lumetri.setLook` + `presets.save`
+
+### 声音（SoundCraft / FilmCraft）
+- 音乐："40-80Hz 轰鸣、持续合成音、共振铜管；无旋律记忆点，只有质地与压强"
+- 环境："分层的风、沙粒打在织物上、呼吸、远处金属应力"
+- 对白："被环境声压低，关键台词只说一次，'听不清'是主题的一部分"
+- 静默："全段 20-40% 的时长无音乐；低频比画面早进 2-4 秒"
+- 落点命令：`essentialSound.setType` + `essentialSound.generateDucking`（避让对白）+ `essentialSound.autoMatch {target:-14}`；SoundCraft `mix.insert` 挂处理链
+
+### 机位与运动（EffectCraft）
+- 优先：[locked wide, constant-speed lateral track, constant-speed push toward vanishing point, high overhead, slow tilt up a structure]
+- 避免：[handheld, orbit, whip pan, speed ramp, crash zoom]
+- 运动预算：low
+- 落点命令：`camera.orbit/pan/dolly` + `view.3d.*` 机位；2D 素材用 `layer.*` + `keys.easyEase*` 缓动
+
+### 质检挂钩（artcraft_qc / 人工复核）
+- 逐镜抽帧复核 + `scopes.read` 示波器核对影调是否落在本透镜的光比与色相档位