feat(monitor): 抖音 Web 接口客户端 —— 绕开爬虫子进程，直接发 HTTP
移植自 mac-agent-os 的 mediacrawler_adapter：不起子进程、不开页面，用浏览器里那份
登录态直接调抖音 Web 接口。产物键名照抄 store/douyin，所以 ingest 那条链路一个字不用改。
**目前能用的（真环境实测，非推断）**：
profile/other : 200, 7075 字节 —— 博主主页指标（粉丝/获赞/作品数/昵称）
aweme/detail : 200, 45425 字节 —— 单条作品详情（含点赞/评论/收藏/分享）
**目前不能用的：作品列表 `aweme/post`。** 两个互相独立的原因：
1. 这个接口被抖音单独升级成了真校验：不带 x-tt-argus 回 403「Uifid Not Found」，
带上 dummy 值回 200 + **空 body**。也就是说「头在不在」骗得过，「真校验」过不了。
同一套头打 profile/other 和 aweme/detail 都是通的 —— 抖音是挑着接口加保护的，
挑中的恰好是「批量拉作品列表」这个最敏感的动作。
2. 改走页面截获也不行：CDP 浏览器打开博主主页会落到「验证码中间页」（当天大量探测的
代价，过几小时要重测）。
所以现在的边界是：**已知作品的指标刷新能做，自动发现新作品做不了**。
**排查中控住变量后得到的两条事实**（都写进注释了）：
· `Accept` / `Accept-Language` / `Referer` 才是主页接口能返回真数据的原因 —— 只有
UA+client hints+Cookie 时是 200 但仅 121 字节的空壳，补上这三个头变 7074 字节。
（我先前猜的 sec-ch-ua 不是关键。）
· 因此 UA 与 client hints 必须**成套地取自同一个浏览器**，所以 BrowserIdentity 一次
从 CDP 取齐 cookie + UA + hints，而不是各自写死。
「200 + 空 body 必须当场报错」也是刻意写死的：放过去它会在下游变成「这个博主没作品」，
把一次失败伪装成一条正常结果 —— 爬虫那条路正是这么栽的，还被翻译成「账号被封」。
测试 +11：cookie 解析、请求头成套性（含 uifid 缺失/回退）、产物键名与 store 对齐、
以及 _get 的三条失败路径（空 body / 403 带网关原话 / 正常返回）。
This commit is contained in:
+53
@@ -0,0 +1,53 @@
|
||||
---
|
||||
|
||||
id: KB-20260517-003
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-05-17
|
||||
date_modified: 2026-05-17
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
collected: true
|
||||
collected_date: 2026-05-20
|
||||
---
|
||||
|
||||
# 2026-05-16_AI提示词驱动文生视频__李导_Agent版_10条视频知识提取
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# AI提示词驱动文生视频——李导（Agent版）10条视频知识提取
|
||||
# AI提示词驱动文生视频——李导（Agent版）知识合集
|
||||
---
|
||||
## 1. 60提示词做 Ai 短视频第三讲 用三分法构图控制主体精准站位
|
||||
### 完整字幕
|
||||
### 提示词/指令原文
|
||||
### 操作步骤
|
||||
1. 确定画面主体与人物站位需求
|
||||
2. 采用三分法构图，将画面九等分
|
||||
3. 把主体放置在三分法的四个视觉锚点上
|
||||
4. 结合景别控制，调整主体的远近与大小
|
||||
5. 多人物场景时，用三分法规定每个人的位置，用景别控制远近层次
|
||||
### 关键参数
|
||||
### 核心技巧
|
||||
1. 用三分法构图实现 AI 画面主体精准站位，解决自然语言命令不准的问题
|
||||
2. 四个交叉点为视觉锚点，是画面视觉重心舒适区
|
||||
3. 构图 + 景别结合，精准控制主体位置、远近、大小
|
||||
4. 多人物场景下，用三分法规定每个人位置，用景别控制层次
|
||||
5. 摄影构图对影片叙事表达起重要作用
|
||||
---
|
||||
## 2. 64用提示词创作短视频教程 2 用视角调动观众的心理预期
|
||||
### 完整字幕
|
||||
### 提示词/指令原文
|
||||
### 操作步骤
|
||||
1. 确定视频创作主体与场景
|
||||
2. 选择提示词所需的景别类型
|
||||
+44
@@ -0,0 +1,44 @@
|
||||
---
|
||||
|
||||
id: KB-20260517-027
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-05-17
|
||||
date_modified: 2026-05-17
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
collected: true
|
||||
collected_date: 2026-05-20
|
||||
---
|
||||
|
||||
# 2026-05-16_character-consistency-reference-sheet
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# character-consistency-reference-sheet
|
||||
# 角色一致性定妆照方案（参考图/多视角 Character Sheet/九宫格）
|
||||
---
|
||||
## 一、核心原则
|
||||
### 从"Lights, Camera, Consistency"论文中提取的核心结论
|
||||
- **I2I 种子帧是角色一致性的架构支点**（ablation 证明：没有 I2I 种子帧，一致性评分从 7.99 暴跌至 0.55）
|
||||
- Asset-First（先出角色定妆照，再出场景）—— 把身份与上下文解耦
|
||||
- Temporal Bridge：上一场景的最后一帧作为下一场景第一帧的视觉条件
|
||||
- CLIPSeg+DINO 用于评估角色一致性的客观指标
|
||||
---
|
||||
## 二、主流定妆照方案
|
||||
### 方案 A：单参考图锚定（Kling 官方推荐）
|
||||
**适用场景**：角色不复杂、场景变化不大的快速生产
|
||||
- 用 **同一张参考图** 锚定每一次生成
|
||||
- 每张图必须清晰展示角色的 **面部、发型、服装**
|
||||
- 参考图要求：**中性光照 + 干净背景**，让模型聚焦主体
|
||||
- 切忌在不同场景间切换参考图——哪怕姿态/光线的微小差异都会导致面部重构
|
||||
+52
@@ -0,0 +1,52 @@
|
||||
---
|
||||
|
||||
id: KB-20260517-030
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-05-17
|
||||
date_modified: 2026-05-17
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
collected: true
|
||||
collected_date: 2026-05-20
|
||||
---
|
||||
|
||||
# 2026-05-16_lip-sync-solutions-comparison
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# lip-sync-solutions-comparison
|
||||
# AI 视频唇形同步（Lip-Sync）方案调研与成本对比
|
||||
---
|
||||
## 一、需求背景
|
||||
- **省去** 视频生成后单独做口型对齐的步骤
|
||||
- 让数字人的口播更自然（语音与唇形天然匹配）
|
||||
- 部分方案支持从 **文本直接到带语音+口型的视频**（Text→Voice→Video 一步完成）
|
||||
---
|
||||
## 二、方案全景对比
|
||||
### 商业方案
|
||||
### 开源方案（可本地部署）
|
||||
---
|
||||
## 三、Kling LipSync 详细分析
|
||||
### API 端点
|
||||
### 输入要求
|
||||
### 处理时间与成本
|
||||
- **固定推理时间**：~12 分钟/次（不随视频时长变化）
|
||||
- **成本**：**$0.014 / 5秒**（按 5 秒单位向上取整）
|
||||
- 3s → 收 5s → $0.014 ≈ ¥0.10
|
||||
- 7s → 收 10s → $0.028 ≈ ¥0.20
|
||||
- 10 个镜头 → 约 $0.14 ≈ ¥1.00
|
||||
- **注意**：通过 fal.ai 代理，非 Kling 官方直连
|
||||
### 限制
|
||||
- 输入视频必须 720p 或 1080p
|
||||
- Audio→Video 视频最长 10s（太短）
|
||||
+51
@@ -0,0 +1,51 @@
|
||||
---
|
||||
|
||||
id: KB-20260517-039
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-05-17
|
||||
date_modified: 2026-05-17
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
collected: true
|
||||
collected_date: 2026-05-20
|
||||
---
|
||||
|
||||
# 2026-05-16_邵氏电影美学AI提示词技巧__依士曼胶片_高调硬光_影棚置景
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 邵氏电影美学AI提示词技巧——依士曼胶片+高调硬光+影棚置景
|
||||
# 邵氏电影美学AI提示词技巧
|
||||
## 视频地址
|
||||
---
|
||||
## 完整字幕
|
||||
- **00:20 — 使用依士曼彩色胶片**：使用依士曼彩色胶片，强制AI还原高饱和度的复古影像特征。
|
||||
- **00:38 — 采用强烈的舞台戏剧、高调硬光**：采用强烈的舞台戏剧、高调硬光，塑造侠客风骨和戏剧光影张力。
|
||||
- **00:58 — 加入高度人工化的影棚置景**：加入高度人工化的影棚置景和材质混搭，呈现古典舞台意境。
|
||||
- **01:20 — 结语**：内容由AI生成。
|
||||
---
|
||||
## 提示词/指令原文
|
||||
---
|
||||
## 操作步骤
|
||||
1. **确定复刻目标**为邵氏电影美学风格
|
||||
2. **加入依士曼彩色胶片**风格关键词
|
||||
3. **加入舞台戏剧感、高调硬光**光影风格描述
|
||||
4. **加入人工化影棚置景、材质混搭**场景描述
|
||||
5. 输入以上风格关键词给到**AI视频生成工具**
|
||||
---
|
||||
## 关键参数
|
||||
---
|
||||
## 核心技巧
|
||||
1. **胶片质感**：套用"依士曼彩色胶片"特质，强制AI还原高饱和度复古影像质感（避免AI默认的数码感/柔和色调）
|
||||
2. **光影塑造**：运用"高调硬光 + 舞台戏剧化光影"，打造武侠侠客风骨与画面张力（邵氏标志性的高反差打光）
|
||||
+48
@@ -0,0 +1,48 @@
|
||||
---
|
||||
|
||||
id: KB-20260601-008
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-01
|
||||
date_modified: 2026-06-01
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-02---
|
||||
|
||||
# 2026-05-17_2026-05-16_AI视频素材生成__提示词工程与分镜脚本方法汇总
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-05-16_AI视频素材生成__提示词工程与分镜脚本方法汇总
|
||||
# AI视频素材生成——提示词工程与分镜脚本方法汇总
|
||||
# AI视频素材生成——提示词工程与分镜脚本方法汇总
|
||||
---
|
||||
## 一、核心流程：文案 → 分镜 → 素材
|
||||
---
|
||||
## 二、分镜脚本提示词模板
|
||||
### 2.1 文案→分镜生成模板
|
||||
1. 镜号 (Shot Number)
|
||||
2. 画面描述 (Visual Description) — 详细描述画面内容
|
||||
3. 旁白文案 (Narration)
|
||||
4. 建议的画面类型 (Type of Visual): 实景 / 动画 / 图示
|
||||
5. 建议的镜头景别 (Shot Scale): 全景 / 中景 / 特写
|
||||
### 2.2 分镜提示词结构（五要素）
|
||||
### 2.3 完整提示词组装示例
|
||||
# → "水墨国风，一位老者在山水间打太极，云雾缭绕的山峰背景，镜头缓慢推进，柔和自然光"
|
||||
---
|
||||
## 三、风格统一方法
|
||||
### 3.1 三原则
|
||||
1. **定义核心风格锚点**: 整期视频定义一个核心风格词（如 "超写实电影感" / "水墨国风"）
|
||||
2. **所有分镜共用风格词**: 每条 prompt 开头/结尾强制加入统一风格描述
|
||||
+44
@@ -0,0 +1,44 @@
|
||||
---
|
||||
|
||||
id: KB-20260601-010
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-01
|
||||
date_modified: 2026-06-01
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-02---
|
||||
|
||||
# 2026-05-17_2026-05-16_character-consistency-reference-sheet
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-05-16_character-consistency-reference-sheet
|
||||
# character-consistency-reference-sheet
|
||||
# 角色一致性定妆照方案（参考图/多视角 Character Sheet/九宫格）
|
||||
---
|
||||
## 一、核心原则
|
||||
### 从"Lights, Camera, Consistency"论文中提取的核心结论
|
||||
- **I2I 种子帧是角色一致性的架构支点**（ablation 证明：没有 I2I 种子帧，一致性评分从 7.99 暴跌至 0.55）
|
||||
- Asset-First（先出角色定妆照，再出场景）—— 把身份与上下文解耦
|
||||
- Temporal Bridge：上一场景的最后一帧作为下一场景第一帧的视觉条件
|
||||
- CLIPSeg+DINO 用于评估角色一致性的客观指标
|
||||
---
|
||||
## 二、主流定妆照方案
|
||||
### 方案 A：单参考图锚定（Kling 官方推荐）
|
||||
**适用场景**：角色不复杂、场景变化不大的快速生产
|
||||
- 用 **同一张参考图** 锚定每一次生成
|
||||
- 每张图必须清晰展示角色的 **面部、发型、服装**
|
||||
- 参考图要求：**中性光照 + 干净背景**，让模型聚焦主体
|
||||
+50
@@ -0,0 +1,50 @@
|
||||
---
|
||||
|
||||
id: KB-20260601-013
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-01
|
||||
date_modified: 2026-06-01
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-02---
|
||||
|
||||
# 2026-05-17_2026-05-16_lip-sync-solutions-comparison
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-05-16_lip-sync-solutions-comparison
|
||||
# lip-sync-solutions-comparison
|
||||
# AI 视频唇形同步（Lip-Sync）方案调研与成本对比
|
||||
---
|
||||
## 一、需求背景
|
||||
- **省去** 视频生成后单独做口型对齐的步骤
|
||||
- 让数字人的口播更自然（语音与唇形天然匹配）
|
||||
- 部分方案支持从 **文本直接到带语音+口型的视频**（Text→Voice→Video 一步完成）
|
||||
---
|
||||
## 二、方案全景对比
|
||||
### 商业方案
|
||||
### 开源方案（可本地部署）
|
||||
---
|
||||
## 三、Kling LipSync 详细分析
|
||||
### API 端点
|
||||
### 输入要求
|
||||
### 处理时间与成本
|
||||
- **固定推理时间**：~12 分钟/次（不随视频时长变化）
|
||||
- **成本**：**$0.014 / 5秒**（按 5 秒单位向上取整）
|
||||
- 3s → 收 5s → $0.014 ≈ ¥0.10
|
||||
- 7s → 收 10s → $0.028 ≈ ¥0.20
|
||||
- 10 个镜头 → 约 $0.14 ≈ ¥1.00
|
||||
- **注意**：通过 fal.ai 代理，非 Kling 官方直连
|
||||
+47
@@ -0,0 +1,47 @@
|
||||
---
|
||||
|
||||
id: KB-20260601-015
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-01
|
||||
date_modified: 2026-06-01
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-02---
|
||||
|
||||
# 2026-05-17_2026-05-16_stuck-intervention
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-05-16_stuck-intervention
|
||||
# stuck-intervention
|
||||
# 卡壳干预模板
|
||||
---
|
||||
## 执行步骤
|
||||
1. 立即停止当前尝试，不自行继续
|
||||
2. 写入 04_memory/logs/errors.log（格式: timestamp | context | error_type | attempted_solutions）
|
||||
3. 按以下模板组织输出
|
||||
---
|
||||
## 输出模板
|
||||
**已尝试方案**:
|
||||
- 方案A: [简述] → 失败原因: [简述]
|
||||
- 方案B: [简述] → 失败原因: [简述]
|
||||
**候选方案**:
|
||||
- **A)** [方案简述] — 预估耗时: [X分钟] — 风险: [简述]
|
||||
- **B)** [方案简述] — 预估耗时: [X分钟] — 风险: [简述]
|
||||
**建议**: [倾向 A/B，简述理由]
|
||||
**等待你的决定** → 收到后按选定方案继续执行。
|
||||
---
|
||||
## 退出
|
||||
+42
@@ -0,0 +1,42 @@
|
||||
---
|
||||
|
||||
id: KB-20260601-019
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-01
|
||||
date_modified: 2026-06-01
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-02---
|
||||
|
||||
# 2026-05-17_AI提示词驱动文生视频__AICG造梦局10条视频知识提取
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# AI提示词驱动文生视频——AICG造梦局10条视频知识提取
|
||||
# AI提示词驱动文生视频——知识合集
|
||||
---
|
||||
## 1. Image2+Sd2.0故事版玩法
|
||||
### 📋 操作步骤
|
||||
1. 视频中演示的具体操作流程，分步骤详细记录
|
||||
### ⚙️ 关键参数
|
||||
- 视频中提到的工具名称、参数设置、具体数字等
|
||||
- 平台：B 站
|
||||
- 盘点对象：Top 10 大模型学习 UP 主
|
||||
- 涉及 UP 主名称：deepthoughts、Coggle 数据科学、王木头学科学、良睦路程序员、霹雳吧啦 Wz、跟李沐学 AI
|
||||
- 适用人群：资深开发、技术发烧友、视觉算法工程师、全栈开发者、AI 技术爱好者、职场研发人员、职场技术人员、数学基础薄弱的初学者、有 Python 基础的开发工程师、非科班新手、产品运营、后端工程师、算法工程师、Python 开发、测试工程师、零基础 AI 入门者、计算机专业学生、AI 研发工程师
|
||||
- 技术方向：高阶源码解析、多模态大模型实战、前沿技术解读、新款模型与新技术解读、数学原理讲解、工程开发、热门应用、模型量化、本地部署、大模型全栈实战
|
||||
### 💡 核心技巧
|
||||
- 视频中传授的具体方法、技巧、经验
|
||||
+54
@@ -0,0 +1,54 @@
|
||||
---
|
||||
|
||||
id: KB-20260601-043
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-01
|
||||
date_modified: 2026-06-01
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-02---
|
||||
|
||||
# 2026-05-17_video-shooting-cool-techniques
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# video-shooting-cool-techniques
|
||||
# 视频拍摄酷炫技巧完全手册——转场·灯光·动作衔接
|
||||
---
|
||||
## 第一部分：转场衔接——让视频"丝滑"的10大技巧
|
||||
### 1.1 相似动作转场
|
||||
**操作**：
|
||||
- 模特在两个场景做相同姿势（转身、挥手、跳跃）
|
||||
- 保持相同构图和拍摄角度
|
||||
- 直接硬切，不加转场
|
||||
**效果**：爆款变装、瞬间场景切换
|
||||
### 1.2 黑色物体转场
|
||||
**操作**：
|
||||
1. 前镜头结尾：镜头推进直到画面被黑色物体完全遮挡
|
||||
2. 后镜头开头：从同一黑色物体向后拉出
|
||||
**适用**：需要极简过渡的画面切换
|
||||
### 1.3 甩镜头转场（Whip Pan）
|
||||
**操作**：前结尾快速右甩 → 后开头从右快速甩入
|
||||
**效果**：穿越时空、快速地点切换
|
||||
**注意**：两段速度方向必须一致
|
||||
### 1.4 下坠/坠落转场
|
||||
**操作**：
|
||||
- 前镜头：物体下落，镜头跟随向下
|
||||
- 后镜头：继续保持同方向（向下）运镜
|
||||
- 使用稳定器全域模式实现灵活移动
|
||||
### 1.5 动作衔接法（最高频使用）
|
||||
**操作步骤**：
|
||||
1. **找点**：找到前一段中动作的最高点（手抬最高、脚刚离地）
|
||||
+57
@@ -0,0 +1,57 @@
|
||||
---
|
||||
|
||||
id: KB-20260601-045
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-01
|
||||
date_modified: 2026-06-01
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-02---
|
||||
|
||||
# 2026-05-17_可灵AI提示词完全知识手册__文生视频_图生视频提示词工程
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 可灵AI提示词完全知识手册——文生视频/图生视频提示词工程
|
||||
# 可灵AI提示词完全知识手册
|
||||
---
|
||||
## 第一章：可灵AI提示词核心公式
|
||||
### 1.1 通用公式结构
|
||||
### 1.2 5W1H 提示词法
|
||||
### 1.3 六大核心要素（编剧视角）
|
||||
1. 主体：谁在画面中？主角/配角的特征
|
||||
2. 构图与角度：黄金分割点、低角度、仰视视角
|
||||
3. 照明与色彩：柔和的侧光、赛博朋克冷色调
|
||||
4. 运动：摄像机镜头运动 + 画面中物体的运动
|
||||
5. 背景：场景环境、时间、氛围
|
||||
---
|
||||
## 第二章：可灵提示词进阶技巧
|
||||
### 2.1 语言风格原则
|
||||
- 语言简洁：用简单的词语和句子，避免复杂长句
|
||||
- 顺序清晰：主体→场景→运动→镜头语言的顺序
|
||||
- 避免歧义：不省略关键信息
|
||||
- 细节优先：先保证核心内容，再逐步添加细节
|
||||
### 2.2 十大万能镜头运动提示词
|
||||
### 2.3 穿越机效果关键词
|
||||
---
|
||||
## 第三章：可灵模型版本与参数
|
||||
### 3.1 版本对照
|
||||
### 3.2 核心参数
|
||||
### 3.3 核心功能
|
||||
- 文生视频 / 图生视频
|
||||
- 首尾帧（关键帧控制）
|
||||
- 运动笔刷（指定局部运动方向）
|
||||
- 对口型功能
|
||||
+47
@@ -0,0 +1,47 @@
|
||||
---
|
||||
id: KB-20260602-002
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-02
|
||||
date_modified: 2026-06-02
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-03---
|
||||
|
||||
# 2026-06-01_2026-05-17_2026-05-16_AI视频素材生成__提示词工程与分镜脚本方法汇总
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-05-17_2026-05-16_AI视频素材生成__提示词工程与分镜脚本方法汇总
|
||||
# 2026-05-16_AI视频素材生成__提示词工程与分镜脚本方法汇总
|
||||
# AI视频素材生成——提示词工程与分镜脚本方法汇总
|
||||
# AI视频素材生成——提示词工程与分镜脚本方法汇总
|
||||
---
|
||||
## 一、核心流程：文案 → 分镜 → 素材
|
||||
---
|
||||
## 二、分镜脚本提示词模板
|
||||
### 2.1 文案→分镜生成模板
|
||||
1. 镜号 (Shot Number)
|
||||
2. 画面描述 (Visual Description) — 详细描述画面内容
|
||||
3. 旁白文案 (Narration)
|
||||
4. 建议的画面类型 (Type of Visual): 实景 / 动画 / 图示
|
||||
5. 建议的镜头景别 (Shot Scale): 全景 / 中景 / 特写
|
||||
### 2.2 分镜提示词结构（五要素）
|
||||
### 2.3 完整提示词组装示例
|
||||
# → "水墨国风，一位老者在山水间打太极，云雾缭绕的山峰背景，镜头缓慢推进，柔和自然光"
|
||||
---
|
||||
## 三、风格统一方法
|
||||
### 3.1 三原则
|
||||
1. **定义核心风格锚点**: 整期视频定义一个核心风格词（如 "超写实电影感" / "水墨国风"）
|
||||
+41
@@ -0,0 +1,41 @@
|
||||
---
|
||||
id: KB-20260602-004
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-02
|
||||
date_modified: 2026-06-02
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-03---
|
||||
|
||||
# 2026-06-01_2026-05-17_2026-05-16_character-consistency-refere
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-05-17_2026-05-16_character-consistency-reference-sheet
|
||||
# 2026-05-16_character-consistency-reference-sheet
|
||||
# character-consistency-reference-sheet
|
||||
# 角色一致性定妆照方案（参考图/多视角 Character Sheet/九宫格）
|
||||
---
|
||||
## 一、核心原则
|
||||
### 从"Lights, Camera, Consistency"论文中提取的核心结论
|
||||
- **I2I 种子帧是角色一致性的架构支点**（ablation 证明：没有 I2I 种子帧，一致性评分从 7.99 暴跌至 0.55）
|
||||
- Asset-First（先出角色定妆照，再出场景）—— 把身份与上下文解耦
|
||||
- Temporal Bridge：上一场景的最后一帧作为下一场景第一帧的视觉条件
|
||||
- CLIPSeg+DINO 用于评估角色一致性的客观指标
|
||||
---
|
||||
## 二、主流定妆照方案
|
||||
### 方案 A：单参考图锚定（Kling 官方推荐）
|
||||
**适用场景**：角色不复杂、场景变化不大的快速生产
|
||||
+44
@@ -0,0 +1,44 @@
|
||||
---
|
||||
id: KB-20260602-005
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-02
|
||||
date_modified: 2026-06-02
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-03---
|
||||
|
||||
# 2026-06-01_2026-05-17_2026-05-16_cross-domain
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-05-17_2026-05-16_cross-domain
|
||||
# 2026-05-16_cross-domain
|
||||
# cross-domain
|
||||
# 跨域联想协议
|
||||
---
|
||||
## 检索路径（按序执行，命中即止）
|
||||
1. **L2 历史洞见**: 查询 L2 facts.db → 是否有相关问题的历史思考和解决方案？
|
||||
2. **知识库跨域模型**: 检索 03_knowledge/10_concepts/ 下各子目录 → 是否有可类比的核心概念？
|
||||
3. **经典理论参考**: 检索 03_knowledge/40_references/ → 是否有可引用的经典理论？
|
||||
4. **无匹配处理**: 以上均无高价值发现 → 回复"本次未找到高价值跨域类比"，不编造
|
||||
---
|
||||
## 输出格式
|
||||
**跨界视角**: [领域]的[核心概念] → [一句精要解释这个类比如何启发当前问题]
|
||||
---
|
||||
## 约束
|
||||
- 宁缺毋滥。找不到就跳过，不编造
|
||||
- 此协议独立于高阶思维协议，两者可先后触发但分别执行
|
||||
- 本轮回答结束后，此模块退出上下文
|
||||
+55
@@ -0,0 +1,55 @@
|
||||
---
|
||||
id: KB-20260602-006
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-02
|
||||
date_modified: 2026-06-02
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-03---
|
||||
|
||||
# 2026-06-01_2026-05-17_2026-05-16_knowledge-review
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-05-17_2026-05-16_knowledge-review
|
||||
# 2026-05-16_knowledge-review
|
||||
# knowledge-review
|
||||
# 知识审查流程
|
||||
---
|
||||
## 审查四步
|
||||
### 1. 来源可靠性评分
|
||||
- 官方文档 / 学术论文：+0.3
|
||||
- 知名技术博客 / 行业报告：+0.1
|
||||
- 个人博客 / 论坛帖子：-0.1
|
||||
- 未知来源 / 社交媒体：-0.3
|
||||
- 基础可靠性基线：0.5
|
||||
### 2. 一致性检查
|
||||
- 完全一致 → 旧知识 confidence 轻微提升
|
||||
- 补充细节 → 合并为新条目
|
||||
- 矛盾 → 进入冲突消解（新事实 > 旧事实，用户明确修正 > 系统推断）
|
||||
### 3. 时效性标记
|
||||
- 技术类知识：180 天后自动失效
|
||||
- 法律/法规：按原文有效期
|
||||
- 个人偏好/观点：不过期，标记为 opinion
|
||||
- 物理/数学公理：永久有效
|
||||
### 4. 日志记录
|
||||
---
|
||||
## 输出格式
|
||||
- 决策：[通过 / 待确认 / 拒绝]
|
||||
- 入库路径：[具体目录路径]
|
||||
- 理由：[一句话说明决策依据]
|
||||
---
|
||||
## 退出
|
||||
+48
@@ -0,0 +1,48 @@
|
||||
---
|
||||
id: KB-20260602-007
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-02
|
||||
date_modified: 2026-06-02
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-03---
|
||||
|
||||
# 2026-06-01_2026-05-17_2026-05-16_lip-sync-solutions-compariso
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-05-17_2026-05-16_lip-sync-solutions-comparison
|
||||
# 2026-05-16_lip-sync-solutions-comparison
|
||||
# lip-sync-solutions-comparison
|
||||
# AI 视频唇形同步（Lip-Sync）方案调研与成本对比
|
||||
---
|
||||
## 一、需求背景
|
||||
- **省去** 视频生成后单独做口型对齐的步骤
|
||||
- 让数字人的口播更自然（语音与唇形天然匹配）
|
||||
- 部分方案支持从 **文本直接到带语音+口型的视频**（Text→Voice→Video 一步完成）
|
||||
---
|
||||
## 二、方案全景对比
|
||||
### 商业方案
|
||||
### 开源方案（可本地部署）
|
||||
---
|
||||
## 三、Kling LipSync 详细分析
|
||||
### API 端点
|
||||
### 输入要求
|
||||
### 处理时间与成本
|
||||
- **固定推理时间**：~12 分钟/次（不随视频时长变化）
|
||||
- **成本**：**$0.014 / 5秒**（按 5 秒单位向上取整）
|
||||
- 3s → 收 5s → $0.014 ≈ ¥0.10
|
||||
- 7s → 收 10s → $0.028 ≈ ¥0.20
|
||||
+52
@@ -0,0 +1,52 @@
|
||||
---
|
||||
id: KB-20260602-008
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-02
|
||||
date_modified: 2026-06-02
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-03---
|
||||
|
||||
# 2026-06-01_2026-05-17_2026-05-16_meta-thinking
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-05-17_2026-05-16_meta-thinking
|
||||
# 2026-05-16_meta-thinking
|
||||
# meta-thinking
|
||||
# 高阶思维协议
|
||||
---
|
||||
## 执行步骤
|
||||
1. **问题本质**
|
||||
2. **前提挑战**
|
||||
**绝不无条件接受用户的预设。**
|
||||
3. **知识溯源**
|
||||
- 优先级1：L2 facts.db 中是否有相关历史洞见？
|
||||
- 优先级2：03_knowledge/10_concepts/ 下是否有可类比的跨领域模型？
|
||||
- 优先级3：03_knowledge/40_references/ 中是否有经典理论支撑？
|
||||
- 若仅为个人推断，须明确声明置信度。
|
||||
4. **反直觉锚点**
|
||||
---
|
||||
## 输出结构
|
||||
**问题本质**: [一句话点明本质]
|
||||
**前提挑战**: [指出现有思路的潜在缺陷]
|
||||
**核心建议**: [立足第一性原理的、高站位的建议]
|
||||
**认知钩子**: [提供一个反直觉视角或思想火花]
|
||||
---
|
||||
## 约束
|
||||
- 找不到高价值洞察时，诚实说"本次无更高视角"，不编造
|
||||
- 不在此模块内展开跨域联想（跨域联想是另一个协议的事）
|
||||
- 本轮回答结束后，此模块退出上下文，恢复工程模式
|
||||
+47
@@ -0,0 +1,47 @@
|
||||
---
|
||||
id: KB-20260602-009
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-02
|
||||
date_modified: 2026-06-02
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-03---
|
||||
|
||||
# 2026-06-01_2026-05-17_2026-05-16_stuck-intervention
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-05-17_2026-05-16_stuck-intervention
|
||||
# 2026-05-16_stuck-intervention
|
||||
# stuck-intervention
|
||||
# 卡壳干预模板
|
||||
---
|
||||
## 执行步骤
|
||||
1. 立即停止当前尝试，不自行继续
|
||||
2. 写入 04_memory/logs/errors.log（格式: timestamp | context | error_type | attempted_solutions）
|
||||
3. 按以下模板组织输出
|
||||
---
|
||||
## 输出模板
|
||||
**已尝试方案**:
|
||||
- 方案A: [简述] → 失败原因: [简述]
|
||||
- 方案B: [简述] → 失败原因: [简述]
|
||||
**候选方案**:
|
||||
- **A)** [方案简述] — 预估耗时: [X分钟] — 风险: [简述]
|
||||
- **B)** [方案简述] — 预估耗时: [X分钟] — 风险: [简述]
|
||||
**建议**: [倾向 A/B，简述理由]
|
||||
**等待你的决定** → 收到后按选定方案继续执行。
|
||||
---
|
||||
## 退出
|
||||
+52
@@ -0,0 +1,52 @@
|
||||
---
|
||||
id: KB-20260602-010
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-02
|
||||
date_modified: 2026-06-02
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-03---
|
||||
|
||||
# 2026-06-01_2026-05-17_2026-05-16_分镜时序与运镜_让AI_懂剪辑
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-05-17_2026-05-16_分镜时序与运镜_让AI_懂剪辑
|
||||
# 2026-05-16_分镜时序与运镜_让AI_懂剪辑
|
||||
# 分镜时序与运镜：让AI'懂剪辑
|
||||
# Ch4: 分镜时序与运镜
|
||||
---
|
||||
## 一、结果推导
|
||||
## 二、多镜头切换的一致性方案
|
||||
### 核心问题
|
||||
### 方案一：提示词锚定法（零成本）
|
||||
**关键规则**：
|
||||
- 描述越具体越好（颜色/材质/年龄/体型）
|
||||
- 跨镜头复制相同文字（重复=一致性）
|
||||
- 添加统一风格锚点："电影级，35mm胶片拍摄，蓝绿和橙色调色"
|
||||
### 方案二：参考图锚定法（最可靠）
|
||||
1. 生成高质量角色参考图作为第一帧
|
||||
2. 每个新镜头都重新选择一张合适的参考图
|
||||
3. 提示词只描述运动，不再描述外貌（模型已从参考图看到）
|
||||
**最佳工具**：Seedance 2.0（身份锁定系统）、可灵3.0（分镜模式6个机位自动一致）
|
||||
### 方案三：分段锚定法（长视频）
|
||||
### 方案四：后期统一
|
||||
- 并排对比所有场景
|
||||
- 漂移严重的以锚定场景为参考重生成
|
||||
- 统一调色：剪映/达芬奇应用统一LUT掩盖细微差异
|
||||
### 实战流程总结
|
||||
---
|
||||
## 三、十大镜头运动
|
||||
+45
@@ -0,0 +1,45 @@
|
||||
---
|
||||
id: KB-20260602-011
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-02
|
||||
date_modified: 2026-06-02
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-03---
|
||||
|
||||
# 2026-06-01_2026-05-17_2026-05-16_原子化登录管理模块_auth_manager
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-05-17_2026-05-16_原子化登录管理模块_auth_manager
|
||||
# 2026-05-16_原子化登录管理模块_auth_manager
|
||||
# 原子化登录管理模块 auth_manager
|
||||
# auth_manager — 原子化登录管理模块
|
||||
## 概述
|
||||
## 核心函数
|
||||
### Cookie 检测（主方案）
|
||||
### DOM 检测（备选，仅移动端视口有效）
|
||||
### 综合检测
|
||||
# 返回: logged_in, cookie_ok, dom_ok, cookie_count, session_id, method
|
||||
### Cookie 持久化
|
||||
### 等待手动登录
|
||||
## 设计原则
|
||||
1. Cookie 检测为主方案（不依赖 UI 布局，桌面端/移动端均可用）
|
||||
2. DOM 检测为备选（仅在 mobile=True 视口下有效）
|
||||
3. 所有函数无副作用（不写 DB/文件，除非函数名明确）
|
||||
4. 统一异常处理（所有异常返回 False/空值）
|
||||
## 解决的问题
|
||||
- `switch_account.py` 此前用 `[data-e2e="user-avatar"]` DOM 检测，桌面端视口下该元素不可见，导致已登录账号误报"未登录"
|
||||
+45
@@ -0,0 +1,45 @@
|
||||
---
|
||||
id: KB-20260602-012
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-02
|
||||
date_modified: 2026-06-02
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-03---
|
||||
|
||||
# 2026-06-01_2026-05-17_2026-05-16_统一升级引擎_agentos_upgrade
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-05-17_2026-05-16_统一升级引擎_agentos_upgrade
|
||||
# 2026-05-16_统一升级引擎_agentos_upgrade
|
||||
# 统一升级引擎 agentos upgrade
|
||||
# agentos upgrade — 统一升级引擎
|
||||
## 概述
|
||||
## 命令
|
||||
## 执行流程
|
||||
1. `git pull origin main` — 拉取最新代码
|
||||
2. 扫描 18 个模块（8 个 skill + 8 个 tool + 2 个其他）
|
||||
3. 按模块层级执行不同操作
|
||||
4. 自动注册技能到 WorkBuddy
|
||||
5. 输出升级报告
|
||||
## 三层模块管理
|
||||
## Python 环境统一
|
||||
1. `AGENTOS_PYTHON` 环境变量
|
||||
2. `~/.workbuddy/binaries/python/envs/agent-os/bin/python3`
|
||||
3. `~/.workbuddy/binaries/python/versions/3.13.12/bin/python3`
|
||||
4. `python3`（系统 fallback）
|
||||
## 位置
|
||||
+40
@@ -0,0 +1,40 @@
|
||||
---
|
||||
id: KB-20260602-013
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-02
|
||||
date_modified: 2026-06-02
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-03---
|
||||
|
||||
# 2026-06-01_2026-05-17_AI提示词驱动文生视频__AICG造梦局10条视频知识提取
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-05-17_AI提示词驱动文生视频__AICG造梦局10条视频知识提取
|
||||
# AI提示词驱动文生视频——AICG造梦局10条视频知识提取
|
||||
# AI提示词驱动文生视频——知识合集
|
||||
---
|
||||
## 1. Image2+Sd2.0故事版玩法
|
||||
### 📋 操作步骤
|
||||
1. 视频中演示的具体操作流程，分步骤详细记录
|
||||
### ⚙️ 关键参数
|
||||
- 视频中提到的工具名称、参数设置、具体数字等
|
||||
- 平台：B 站
|
||||
- 盘点对象：Top 10 大模型学习 UP 主
|
||||
- 涉及 UP 主名称：deepthoughts、Coggle 数据科学、王木头学科学、良睦路程序员、霹雳吧啦 Wz、跟李沐学 AI
|
||||
- 适用人群：资深开发、技术发烧友、视觉算法工程师、全栈开发者、AI 技术爱好者、职场研发人员、职场技术人员、数学基础薄弱的初学者、有 Python 基础的开发工程师、非科班新手、产品运营、后端工程师、算法工程师、Python 开发、测试工程师、零基础 AI 入门者、计算机专业学生、AI 研发工程师
|
||||
- 技术方向：高阶源码解析、多模态大模型实战、前沿技术解读、新款模型与新技术解读、数学原理讲解、工程开发、热门应用、模型量化、本地部署、大模型全栈实战
|
||||
+47
@@ -0,0 +1,47 @@
|
||||
---
|
||||
id: KB-20260602-014
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-02
|
||||
date_modified: 2026-06-02
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-03---
|
||||
|
||||
# 2026-06-01_2026-05-17_AI提示词驱动文生视频__博主干货提取_10_11条_
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-05-17_AI提示词驱动文生视频__博主干货提取_10_11条_
|
||||
# AI提示词驱动文生视频——博主干货提取（10/11条）
|
||||
# AI提示词驱动文生视频——博主干货
|
||||
---
|
||||
## 1. 反着写提示词AI人物更真实
|
||||
---
|
||||
## 2. 明非屠龙可不是件容易事
|
||||
---
|
||||
## 3. 或许ai也有遗憾
|
||||
---
|
||||
## 4. 骨相留白让AI人物更有少年气
|
||||
---
|
||||
## 5. 劳动人民的伟力
|
||||
---
|
||||
## 6. 伟大出自平凡
|
||||
---
|
||||
## 7. 空间锚定画面更有张力
|
||||
---
|
||||
## 8. 衰仔要走咯
|
||||
---
|
||||
## 9. 我的名字叫赫尔佐格
|
||||
+50
@@ -0,0 +1,50 @@
|
||||
---
|
||||
id: KB-20260602-015
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-02
|
||||
date_modified: 2026-06-02
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-03---
|
||||
|
||||
# 2026-06-01_2026-05-17_AI提示词驱动文生视频__李导_Agent版_10条视频知识提取
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-05-17_AI提示词驱动文生视频__李导_Agent版_10条视频知识提取
|
||||
# AI提示词驱动文生视频——李导（Agent版）10条视频知识提取
|
||||
# AI提示词驱动文生视频——李导（Agent版）知识合集
|
||||
---
|
||||
## 1. 60提示词做 Ai 短视频第三讲 用三分法构图控制主体精准站位
|
||||
### 完整字幕
|
||||
### 提示词/指令原文
|
||||
### 操作步骤
|
||||
1. 确定画面主体与人物站位需求
|
||||
2. 采用三分法构图，将画面九等分
|
||||
3. 把主体放置在三分法的四个视觉锚点上
|
||||
4. 结合景别控制，调整主体的远近与大小
|
||||
5. 多人物场景时，用三分法规定每个人的位置，用景别控制远近层次
|
||||
### 关键参数
|
||||
### 核心技巧
|
||||
1. 用三分法构图实现 AI 画面主体精准站位，解决自然语言命令不准的问题
|
||||
2. 四个交叉点为视觉锚点，是画面视觉重心舒适区
|
||||
3. 构图 + 景别结合，精准控制主体位置、远近、大小
|
||||
4. 多人物场景下，用三分法规定每个人位置，用景别控制层次
|
||||
5. 摄影构图对影片叙事表达起重要作用
|
||||
---
|
||||
## 2. 64用提示词创作短视频教程 2 用视角调动观众的心理预期
|
||||
### 完整字幕
|
||||
### 提示词/指令原文
|
||||
+52
@@ -0,0 +1,52 @@
|
||||
---
|
||||
id: KB-20260602-017
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-02
|
||||
date_modified: 2026-06-02
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-03---
|
||||
|
||||
# 2026-06-01_2026-05-17_AI视频提示词系统创作教科书___知识体系索引
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-05-17_AI视频提示词系统创作教科书___知识体系索引
|
||||
# AI视频提示词系统创作教科书 · 知识体系索引
|
||||
# AI 视频提示词系统创作教科书 · 知识体系
|
||||
---
|
||||
## 体系总览
|
||||
### 现有知识关联
|
||||
---
|
||||
## 第一部分: 底层逻辑与核心原则（基石篇·结果导向）
|
||||
### Ch1: AI视频生成的"执行本质"与结果推导逻辑
|
||||
**核心问题**: AI 不是读心术，指令的明确性决定输出上限。
|
||||
**结果推导路径**:
|
||||
**采集方向**:
|
||||
- [ ] AI 视频生成工具的底层原理（文本→视频的映射机制）
|
||||
- [ ] 人类创作逻辑 vs AI 执行逻辑的差异分析
|
||||
- [ ] 结果推导方法论的具体案例（多种场景）
|
||||
- [ ] 提示词信息传递模型的图解
|
||||
### Ch2: 结果推导方法论
|
||||
**核心问题**: 如何从目标画面反推提示词锚点。
|
||||
**采集方向**:
|
||||
- [ ] 多场景结果推导案例库
|
||||
- [ ] 情绪→视觉元素的映射表
|
||||
- [ ] 目标反推的思维训练方法
|
||||
---
|
||||
## 第二部分: 核心模块与创作逻辑（技术篇·从结果定义技术）
|
||||
### Ch3: 基础定调参数
|
||||
**结果导向**: 不同成果的参数选型：
|
||||
+52
@@ -0,0 +1,52 @@
|
||||
---
|
||||
id: KB-20260602-022
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-02
|
||||
date_modified: 2026-06-02
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-03---
|
||||
|
||||
# 2026-06-01_2026-05-17_ch02-result-driven
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-05-17_ch02-result-driven
|
||||
# ch02-result-driven
|
||||
# Ch2: 结果推导方法论——从目标画面反推指令锚点
|
||||
---
|
||||
## 一、三大核心步骤
|
||||
### 步骤1：视觉要素深度拆解
|
||||
**主体要素**
|
||||
- 形态特征：八头身比例、肌肉线条分明
|
||||
- 材质表现：磨砂金属、半透明硅胶
|
||||
- 动态捕捉：跳跃瞬间、手指弯曲45度
|
||||
- 表情管理：眼角上扬15度
|
||||
**环境系统**
|
||||
- 空间层次：前中背景的元素分布
|
||||
- 物理特性：重力、光照、大气折射
|
||||
- 时间维度：季节、昼夜、天气
|
||||
- 文化符号：特定时代的建筑/服饰
|
||||
**光影工程**
|
||||
- 光源位置：钟表定位法（3点钟方向顶光）
|
||||
- 光质控制：硬光(直射) vs 柔光(漫反射)
|
||||
- 色温调节：2000K到10000K
|
||||
- 光比计算：主光与辅光比值（光比4:1）
|
||||
### 步骤2：结构化提示词构建
|
||||
**权重应用**：
|
||||
- 权重越大，AI对该要素越重视
|
||||
- 艺术风格编码：`cinematic, bokeh`（摄影）、`octane render, PBR`（3D）
|
||||
- 构图参数：`16mm`（广角）、`f/1.2`（虚化）
|
||||
+44
@@ -0,0 +1,44 @@
|
||||
---
|
||||
id: KB-20260602-024
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-02
|
||||
date_modified: 2026-06-02
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-03---
|
||||
|
||||
# 2026-06-01_2026-05-17_ch07-micro-expression
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-05-17_ch07-micro-expression
|
||||
# ch07-micro-expression
|
||||
# Ch7: 人物塑造-微表情与情绪控制
|
||||
---
|
||||
## 核心知识
|
||||
### 情绪→微表情映射
|
||||
### 三招让角色活过来（来源：抖音采集）
|
||||
**1. 控制情绪强度**
|
||||
- 避免 AI 默认情绪最大值（容易假）
|
||||
- 在提示词中加入强度修饰：`slightly`, `barely perceptible`, `subtle`
|
||||
- 示例：`a faint smile barely lifting the corners of his mouth`
|
||||
**2. 给微表情一个发生原因**
|
||||
- 微表情不能孤立存在，需要与场景/对白联动
|
||||
- 加入触发条件的描述：`upon hearing the news, her eyes widened`
|
||||
- 加上半身动作，让画面脱离摆拍感
|
||||
**3. 给情绪加时间顺序**
|
||||
- 描述情绪的变化过程：`from confusion to realization`
|
||||
- 示例：`His expression shifted from doubt to shock as the truth sank in`
|
||||
+52
@@ -0,0 +1,52 @@
|
||||
---
|
||||
id: KB-20260602-027
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-02
|
||||
date_modified: 2026-06-02
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-03---
|
||||
|
||||
# 2026-06-01_2026-05-17_ch10-commercial-tvc
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-05-17_ch10-commercial-tvc
|
||||
# ch10-commercial-tvc
|
||||
# Ch10: 商业广告/TVC创作
|
||||
---
|
||||
## 一、AI-TVC完整工作流
|
||||
### 5步出品（20分钟内）
|
||||
**工具链**：
|
||||
### 核心：运镜提示词
|
||||
---
|
||||
## 二、TVC提示词结构
|
||||
### 标准格式
|
||||
### 示例：像素风TVC
|
||||
### 玉石品牌TVC（来自既有知识）
|
||||
- 结合中式色彩美学
|
||||
- 电影感构图
|
||||
- 专业LUT调色提升品质
|
||||
- Seedance 2.0 生成电影级画面
|
||||
---
|
||||
## 三、商业广告类型与策略
|
||||
### 高级技巧
|
||||
1. **图片先行**：先用高质量生图工具打好画面基础
|
||||
2. **动作提示词含运镜**：让静态图"活"起来的关键
|
||||
3. **后处理统一调性**：剪映统一调色+LUT
|
||||
4. **多次迭代**：每次只改5%内容AB测试
|
||||
---
|
||||
## 采集来源
|
||||
+52
@@ -0,0 +1,52 @@
|
||||
---
|
||||
id: KB-20260602-028
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-02
|
||||
date_modified: 2026-06-02
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-03---
|
||||
|
||||
# 2026-06-01_2026-05-17_ch11-sci-fi-fantasy
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-05-17_ch11-sci-fi-fantasy
|
||||
# ch11-sci-fi-fantasy
|
||||
# Ch11: 科幻/奇幻风格创作
|
||||
---
|
||||
## 一、科幻风格提示词大全
|
||||
### 提示词结构公式
|
||||
### 赛博朋克城市
|
||||
**完整模板**：
|
||||
**关键元素**：
|
||||
### 太空场景
|
||||
**空间站**：
|
||||
**外星景观**：
|
||||
### 科技与机械
|
||||
**人形机器人**：
|
||||
---
|
||||
## 二、风格参数
|
||||
### 光影参数
|
||||
### 美术风格
|
||||
### 风格参考锚点
|
||||
---
|
||||
## 三、奇幻风格创作
|
||||
### 通用奇幻模板
|
||||
### 经典奇幻类型
|
||||
### 常见错误与修正
|
||||
---
|
||||
## 采集来源
|
||||
+42
@@ -0,0 +1,42 @@
|
||||
---
|
||||
id: KB-20260602-029
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-02
|
||||
date_modified: 2026-06-02
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-03---
|
||||
|
||||
# 2026-06-01_2026-05-17_ch12-model-comparison
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-05-17_ch12-model-comparison
|
||||
# ch12-model-comparison
|
||||
# Ch12: 主流AI视频大模型的差异与适配
|
||||
---
|
||||
## 核心数据：平台对比
|
||||
## 模型版本演进
|
||||
## 结果导向选型
|
||||
## 工具链协同
|
||||
- 商业广告：Gen-2(画面) + Au(声音)
|
||||
- 动漫番剧：Pika(多镜头) + 配音平台(对白)
|
||||
- 电影短片：Kling 3.0(素材) + Runway(后期)
|
||||
---
|
||||
## 已采集素材
|
||||
### 抖音大模型对比素材
|
||||
---
|
||||
## 采集来源
|
||||
+53
@@ -0,0 +1,53 @@
|
||||
---
|
||||
id: KB-20260602-030
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-02
|
||||
date_modified: 2026-06-02
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-03---
|
||||
|
||||
# 2026-06-01_2026-05-17_ch13-tool-workflow
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-05-17_ch13-tool-workflow
|
||||
# ch13-tool-workflow
|
||||
# Ch13: 工具实操与工作流搭建
|
||||
---
|
||||
## 一、四大核心工具详解
|
||||
### Runway Gen-3 Alpha
|
||||
**定位**：电影级画质、精细控制，专业制作首选
|
||||
**操作流程**：
|
||||
1. 准备高质量静态图（建议1080p以上，最好4K）
|
||||
2. 上传图片至平台（支持JPG/PNG）
|
||||
3. 编写提示词描述期望的动态效果
|
||||
4. 调整 Camera Motion 参数和 Motion Bucket
|
||||
5. 生成预览（耗时2-5分钟）
|
||||
**提示词结构**：
|
||||
**参数**：
|
||||
- Motion Bucket：0-10（控制运动强度）
|
||||
- Camera Motion：Dolly In/Out、Tracking Shot、Orbit、Bird's Eye
|
||||
- 时长上限：16秒
|
||||
- 分辨率：最高4K
|
||||
---
|
||||
### Kling 可灵
|
||||
**定位**：中文理解最佳、性价比高
|
||||
**操作流程**：
|
||||
1. 准备1080p以上静态图
|
||||
2. 上传图片（支持JPG/PNG）
|
||||
3. 用自然语言描述动态效果
|
||||
4. 选择运动模式
|
||||
+56
@@ -0,0 +1,56 @@
|
||||
---
|
||||
id: KB-20260602-031
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-02
|
||||
date_modified: 2026-06-02
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-03---
|
||||
|
||||
# 2026-06-01_2026-05-17_ch14-script-to-prompt
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-05-17_ch14-script-to-prompt
|
||||
# ch14-script-to-prompt
|
||||
# Ch14: 从"创意"到"AI友好型脚本"的转化
|
||||
---
|
||||
## 一、四步工作流：小说→分镜脚本
|
||||
### 核心原理
|
||||
### 四步流程
|
||||
**第一步：规范化整理内容**
|
||||
**第二步：预估镜头数与结构分布**
|
||||
- 需要多少镜头？
|
||||
- 镜头在关键情节中的分布？
|
||||
- 建立视觉节奏感
|
||||
**第三步：搭建核心场景与角色表**
|
||||
- 场景1：咸亨酒店大堂，午后，光线昏暗，柜台有酒坛
|
||||
- 角色A：孔乙己，瘦弱，长衫破旧
|
||||
**第四步：整合生成分镜脚本**
|
||||
**效果对比**（以《孔乙己》为例）：
|
||||
- 直接转换：仅12个镜头
|
||||
- 四步工作流：59个分镜头 ✅ 完整还原剧情
|
||||
### 提示词模板
|
||||
---
|
||||
## 二、场景控制技术
|
||||
### 先建库，再使用
|
||||
### 场景描述要素
|
||||
### 场景互动
|
||||
---
|
||||
## 三、AI友好型脚本格式
|
||||
### 模块化结构
|
||||
### 完整示例
|
||||
---
|
||||
+47
@@ -0,0 +1,47 @@
|
||||
---
|
||||
id: KB-20260602-032
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-02
|
||||
date_modified: 2026-06-02
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-03---
|
||||
|
||||
# 2026-06-01_2026-05-17_ch15-narrative-structure
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-05-17_ch15-narrative-structure
|
||||
# ch15-narrative-structure
|
||||
# Ch15: 叙事结构与AI叙事能力的结合
|
||||
---
|
||||
## 常见叙事结构
|
||||
## AI叙事优势与局限
|
||||
### AI擅长
|
||||
- 视觉化情绪（光影+微表情+场景）
|
||||
- 高密度画面叙事（每一秒承载大量信息）
|
||||
- 风格化渲染（电影感/动漫感）
|
||||
### AI不擅长
|
||||
- 复杂对白逻辑（容易混淆人物关系）
|
||||
- 细微情感递进（缺少层次感）
|
||||
- 长时间连贯性（超过30秒容易崩）
|
||||
### 策略：扬长避短
|
||||
## 短剧分镜7大要素
|
||||
1. 景别（远全中近特）
|
||||
2. 视角（平俯仰倾）
|
||||
3. 构图（三分法/对称/不对称）
|
||||
4. 光影（情绪映射）
|
||||
5. 运动（镜头+主体）
|
||||
+47
@@ -0,0 +1,47 @@
|
||||
---
|
||||
id: KB-20260602-033
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-02
|
||||
date_modified: 2026-06-02
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-03---
|
||||
|
||||
# 2026-06-01_2026-05-17_ch16-practice-path
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-05-17_ch16-practice-path
|
||||
# ch16-practice-path
|
||||
# Ch16: 从"模仿"到"创新"的实践路径
|
||||
---
|
||||
## 三个阶段
|
||||
### 阶段1：爆款拆解（入门）
|
||||
- 分析抖音/B站AI视频爆款的"结果→提示词→工具"链路
|
||||
- 工具：video-to-prompt.com（自动反推提示词）
|
||||
- 目标：能复现同级别效果
|
||||
### 阶段2：垂直深耕（进阶）
|
||||
- 聚焦一个领域（如AI漫剧/AI电影感/AI广告）
|
||||
- 打磨"成果稳定"的提示词模板
|
||||
- 建立自己的风格关键词库
|
||||
- 目标：每次生成都能稳定输出
|
||||
### 阶段3：风格创新（高级）
|
||||
- 融合多领域元素（如赛博朋克+古风）
|
||||
- 打造"差异化成果"的AI视频风格
|
||||
- 建立个人方法论体系
|
||||
- 目标：一眼能认出是你的风格
|
||||
## 成长路径
|
||||
## 效率指标
|
||||
+37
@@ -0,0 +1,37 @@
|
||||
---
|
||||
id: KB-20260602-034
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-02
|
||||
date_modified: 2026-06-02
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-03---
|
||||
|
||||
# 2026-06-01_2026-05-17_ch17-resource-library
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-05-17_ch17-resource-library
|
||||
# ch17-resource-library
|
||||
# Ch17: 资源库建设——按成果类型分类的知识弹药库
|
||||
---
|
||||
## 平台资源
|
||||
## 工具资源
|
||||
## 提示词资源
|
||||
- PromptMart（50000+提示词库）
|
||||
- video-to-prompt.com（视频反推提示词）
|
||||
- 本知识库各章节（不断积累）
|
||||
## 案例资源
|
||||
+53
@@ -0,0 +1,53 @@
|
||||
---
|
||||
id: KB-20260602-035
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-02
|
||||
date_modified: 2026-06-02
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-03---
|
||||
|
||||
# 2026-06-01_2026-05-17_ch18-problem-solving
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-05-17_ch18-problem-solving
|
||||
# ch18-problem-solving
|
||||
# Ch18: 常见问题解决方案库
|
||||
---
|
||||
## 核心问题分类
|
||||
### A. 画面类问题
|
||||
#### A1. 人脸变形/身份漂移
|
||||
**提示词修复**：
|
||||
#### A2. 物体漂移/元素消失
|
||||
**提示词修复**：
|
||||
#### A3. 光影闪烁/曝光变化
|
||||
**后期补救**：调色工具统一曝光。
|
||||
#### A4. 画面风格跑偏
|
||||
- **检查参考示例**是否明确
|
||||
- **换适配大模型**（如写实→Kling，动漫→Pika）
|
||||
- 提示词加入风格锚点：`cinematic, photorealistic, anime style`
|
||||
---
|
||||
### B. 运动类问题
|
||||
#### B1. 不自然动作/飘忽运动
|
||||
#### B2. 手部与手指问题
|
||||
- **裁剪手部**或让其失焦
|
||||
- 使用简单手部姿势（握拳、张开手掌）
|
||||
- **避免手部特写**，使用广角镜头
|
||||
- **多次生成并挑选**最佳版本
|
||||
#### B3. 时间不一致性（"梦境"效果）
|
||||
- **分段生成**：3-5秒片段后拼接
|
||||
- 每个片段**重复关键视觉描述符**
|
||||
+52
@@ -0,0 +1,52 @@
|
||||
---
|
||||
id: KB-20260602-037
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-02
|
||||
date_modified: 2026-06-02
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-03---
|
||||
|
||||
# 2026-06-01_2026-05-17_video-shooting-cool-techniques
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-05-17_video-shooting-cool-techniques
|
||||
# video-shooting-cool-techniques
|
||||
# 视频拍摄酷炫技巧完全手册——转场·灯光·动作衔接
|
||||
---
|
||||
## 第一部分：转场衔接——让视频"丝滑"的10大技巧
|
||||
### 1.1 相似动作转场
|
||||
**操作**：
|
||||
- 模特在两个场景做相同姿势（转身、挥手、跳跃）
|
||||
- 保持相同构图和拍摄角度
|
||||
- 直接硬切，不加转场
|
||||
**效果**：爆款变装、瞬间场景切换
|
||||
### 1.2 黑色物体转场
|
||||
**操作**：
|
||||
1. 前镜头结尾：镜头推进直到画面被黑色物体完全遮挡
|
||||
2. 后镜头开头：从同一黑色物体向后拉出
|
||||
**适用**：需要极简过渡的画面切换
|
||||
### 1.3 甩镜头转场（Whip Pan）
|
||||
**操作**：前结尾快速右甩 → 后开头从右快速甩入
|
||||
**效果**：穿越时空、快速地点切换
|
||||
**注意**：两段速度方向必须一致
|
||||
### 1.4 下坠/坠落转场
|
||||
**操作**：
|
||||
- 前镜头：物体下落，镜头跟随向下
|
||||
- 后镜头：继续保持同方向（向下）运镜
|
||||
- 使用稳定器全域模式实现灵活移动
|
||||
### 1.5 动作衔接法（最高频使用）
|
||||
+54
@@ -0,0 +1,54 @@
|
||||
---
|
||||
id: KB-20260602-039
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-02
|
||||
date_modified: 2026-06-02
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-03---
|
||||
|
||||
# 2026-06-01_2026-05-17_可灵AI提示词完全知识手册__文生视频_图生视频提示词工程
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-05-17_可灵AI提示词完全知识手册__文生视频_图生视频提示词工程
|
||||
# 可灵AI提示词完全知识手册——文生视频/图生视频提示词工程
|
||||
# 可灵AI提示词完全知识手册
|
||||
---
|
||||
## 第一章：可灵AI提示词核心公式
|
||||
### 1.1 通用公式结构
|
||||
### 1.2 5W1H 提示词法
|
||||
### 1.3 六大核心要素（编剧视角）
|
||||
1. 主体：谁在画面中？主角/配角的特征
|
||||
2. 构图与角度：黄金分割点、低角度、仰视视角
|
||||
3. 照明与色彩：柔和的侧光、赛博朋克冷色调
|
||||
4. 运动：摄像机镜头运动 + 画面中物体的运动
|
||||
5. 背景：场景环境、时间、氛围
|
||||
---
|
||||
## 第二章：可灵提示词进阶技巧
|
||||
### 2.1 语言风格原则
|
||||
- 语言简洁：用简单的词语和句子，避免复杂长句
|
||||
- 顺序清晰：主体→场景→运动→镜头语言的顺序
|
||||
- 避免歧义：不省略关键信息
|
||||
- 细节优先：先保证核心内容，再逐步添加细节
|
||||
### 2.2 十大万能镜头运动提示词
|
||||
### 2.3 穿越机效果关键词
|
||||
---
|
||||
## 第三章：可灵模型版本与参数
|
||||
### 3.1 版本对照
|
||||
### 3.2 核心参数
|
||||
### 3.3 核心功能
|
||||
- 文生视频 / 图生视频
|
||||
+55
@@ -0,0 +1,55 @@
|
||||
---
|
||||
id: KB-20260602-041
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-02
|
||||
date_modified: 2026-06-02
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-03---
|
||||
|
||||
# 2026-06-01_2026-05-17_帅是一种感觉__爆款酷炫短视频拆解与标准化流程
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-05-17_帅是一种感觉__爆款酷炫短视频拆解与标准化流程
|
||||
# 帅是一种感觉——爆款酷炫短视频拆解与标准化流程
|
||||
# "帅是一种感觉"——爆款酷炫短视频拆解与标准化流程
|
||||
---
|
||||
## 一、视频逐秒拆解
|
||||
### 核心技术点拆解
|
||||
---
|
||||
## 二、"帅"的本质——技术拆解
|
||||
### 2.1 转场手法（5种混合使用）
|
||||
### 2.2 节奏控制
|
||||
**节奏曲线图**：
|
||||
**关键洞察**：爆款视频的最大秘密不在"做了什么"，而在于**"什么时候停"**。这个视频在**情绪最高点戛然而止**，不给观众喘息时间，让人忍不住重播。
|
||||
### 2.3 画面选择
|
||||
---
|
||||
## 三、标准化流程——任何人都能复刻的"帅"公式
|
||||
### 3.1 三幕式爆款结构
|
||||
### 3.2 素材选择清单
|
||||
### 3.3 剪辑操作步骤
|
||||
### 3.4 检查清单
|
||||
---
|
||||
## 四、关联知识库
|
||||
---
|
||||
## 五、从"酷炫"到"自有风格"的进阶路径
|
||||
### 第一步：复刻（1-3天）
|
||||
- 找一段热门动漫/影视素材
|
||||
- 按照三幕式结构 + 检查清单复刻
|
||||
- 工具：剪映（导出者说"先下个剪映"）
|
||||
### 第二步：掌握节奏感（1-2周）
|
||||
- 练习卡点：让每个画面切换都在音乐节拍上
|
||||
+50
@@ -0,0 +1,50 @@
|
||||
---
|
||||
id: KB-20260602-042
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-02
|
||||
date_modified: 2026-06-02
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-03---
|
||||
|
||||
# 2026-06-01_2026-05-17_邵氏电影美学AI提示词技巧__依士曼胶片_高调硬光_影棚置景
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-05-17_邵氏电影美学AI提示词技巧__依士曼胶片_高调硬光_影棚置景
|
||||
# 邵氏电影美学AI提示词技巧——依士曼胶片+高调硬光+影棚置景
|
||||
# 邵氏电影美学AI提示词技巧
|
||||
## 视频地址
|
||||
---
|
||||
## 完整字幕
|
||||
- **00:20 — 使用依士曼彩色胶片**：使用依士曼彩色胶片，强制AI还原高饱和度的复古影像特征。
|
||||
- **00:38 — 采用强烈的舞台戏剧、高调硬光**：采用强烈的舞台戏剧、高调硬光，塑造侠客风骨和戏剧光影张力。
|
||||
- **00:58 — 加入高度人工化的影棚置景**：加入高度人工化的影棚置景和材质混搭，呈现古典舞台意境。
|
||||
- **01:20 — 结语**：内容由AI生成。
|
||||
---
|
||||
## 提示词/指令原文
|
||||
---
|
||||
## 操作步骤
|
||||
1. **确定复刻目标**为邵氏电影美学风格
|
||||
2. **加入依士曼彩色胶片**风格关键词
|
||||
3. **加入舞台戏剧感、高调硬光**光影风格描述
|
||||
4. **加入人工化影棚置景、材质混搭**场景描述
|
||||
5. 输入以上风格关键词给到**AI视频生成工具**
|
||||
---
|
||||
## 关键参数
|
||||
---
|
||||
## 核心技巧
|
||||
1. **胶片质感**：套用"依士曼彩色胶片"特质，强制AI还原高饱和度复古影像质感（避免AI默认的数码感/柔和色调）
|
||||
+44
@@ -0,0 +1,44 @@
|
||||
---
|
||||
id: KB-20260603-002
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-03
|
||||
date_modified: 2026-06-03
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-04---
|
||||
|
||||
# 2026-06-02_2026-06-01_2026-05-17_2026-05-16_AI视频素材生成__提示词工程与分
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-06-01_2026-05-17_2026-05-16_AI视频素材生成__提示词工程与分镜脚本方法汇总
|
||||
# 2026-05-17_2026-05-16_AI视频素材生成__提示词工程与分镜脚本方法汇总
|
||||
# 2026-05-16_AI视频素材生成__提示词工程与分镜脚本方法汇总
|
||||
# AI视频素材生成——提示词工程与分镜脚本方法汇总
|
||||
# AI视频素材生成——提示词工程与分镜脚本方法汇总
|
||||
---
|
||||
## 一、核心流程：文案 → 分镜 → 素材
|
||||
---
|
||||
## 二、分镜脚本提示词模板
|
||||
### 2.1 文案→分镜生成模板
|
||||
1. 镜号 (Shot Number)
|
||||
2. 画面描述 (Visual Description) — 详细描述画面内容
|
||||
3. 旁白文案 (Narration)
|
||||
4. 建议的画面类型 (Type of Visual): 实景 / 动画 / 图示
|
||||
5. 建议的镜头景别 (Shot Scale): 全景 / 中景 / 特写
|
||||
### 2.2 分镜提示词结构（五要素）
|
||||
### 2.3 完整提示词组装示例
|
||||
# → "水墨国风，一位老者在山水间打太极，云雾缭绕的山峰背景，镜头缓慢推进，柔和自然光"
|
||||
+41
@@ -0,0 +1,41 @@
|
||||
---
|
||||
id: KB-20260603-003
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-03
|
||||
date_modified: 2026-06-03
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-04---
|
||||
|
||||
# 2026-06-02_2026-06-01_2026-05-17_2026-05-16_Camoufox_集成修复记录
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-06-01_2026-05-17_2026-05-16_Camoufox_集成修复记录
|
||||
# 2026-05-17_2026-05-16_Camoufox_集成修复记录
|
||||
# 2026-05-16_Camoufox_集成修复记录
|
||||
# Camoufox 集成修复记录
|
||||
# Camoufox 集成修复记录
|
||||
## 正确启动方式
|
||||
## 修复清单
|
||||
### 1. 操作系统指纹（P1）
|
||||
**问题**: Camoufox 默认从 `[windows, macos, linux]` 随机，导致手机端扫码有时显示"登录macOS"。
|
||||
**修复**: 添加 `os='windows'` 固定参数。
|
||||
**文件**: `cdp_connector.py:_launch_camoufox()`
|
||||
### 2. 中文乱码（P1）
|
||||
**问题**: Camoufox/Firefox 不加载系统字体，抖音页面中文显示乱码。
|
||||
**修复**: 添加 `fonts=['STHeiti', 'Heiti SC', 'PingFang SC', 'Noto Sans CJK SC']` 和 `humanize=1.5`。
|
||||
**文件**: `cdp_connector.py:_launch_camoufox()`
|
||||
+39
@@ -0,0 +1,39 @@
|
||||
---
|
||||
id: KB-20260603-004
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-03
|
||||
date_modified: 2026-06-03
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-04---
|
||||
|
||||
# 2026-06-02_2026-06-01_2026-05-17_2026-05-16_character-consist
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-06-01_2026-05-17_2026-05-16_character-consistency-refere
|
||||
# 2026-05-17_2026-05-16_character-consistency-reference-sheet
|
||||
# 2026-05-16_character-consistency-reference-sheet
|
||||
# character-consistency-reference-sheet
|
||||
# 角色一致性定妆照方案（参考图/多视角 Character Sheet/九宫格）
|
||||
---
|
||||
## 一、核心原则
|
||||
### 从"Lights, Camera, Consistency"论文中提取的核心结论
|
||||
- **I2I 种子帧是角色一致性的架构支点**（ablation 证明：没有 I2I 种子帧，一致性评分从 7.99 暴跌至 0.55）
|
||||
- Asset-First（先出角色定妆照，再出场景）—— 把身份与上下文解耦
|
||||
- Temporal Bridge：上一场景的最后一帧作为下一场景第一帧的视觉条件
|
||||
- CLIPSeg+DINO 用于评估角色一致性的客观指标
|
||||
---
|
||||
+45
@@ -0,0 +1,45 @@
|
||||
---
|
||||
id: KB-20260603-005
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-03
|
||||
date_modified: 2026-06-03
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-04---
|
||||
|
||||
# 2026-06-02_2026-06-01_2026-05-17_2026-05-16_cross-domain
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-06-01_2026-05-17_2026-05-16_cross-domain
|
||||
# 2026-05-17_2026-05-16_cross-domain
|
||||
# 2026-05-16_cross-domain
|
||||
# cross-domain
|
||||
# 跨域联想协议
|
||||
---
|
||||
## 检索路径（按序执行，命中即止）
|
||||
1. **L2 历史洞见**: 查询 L2 facts.db → 是否有相关问题的历史思考和解决方案？
|
||||
2. **知识库跨域模型**: 检索 03_knowledge/10_concepts/ 下各子目录 → 是否有可类比的核心概念？
|
||||
3. **经典理论参考**: 检索 03_knowledge/40_references/ → 是否有可引用的经典理论？
|
||||
4. **无匹配处理**: 以上均无高价值发现 → 回复"本次未找到高价值跨域类比"，不编造
|
||||
---
|
||||
## 输出格式
|
||||
**跨界视角**: [领域]的[核心概念] → [一句精要解释这个类比如何启发当前问题]
|
||||
---
|
||||
## 约束
|
||||
- 宁缺毋滥。找不到就跳过，不编造
|
||||
- 此协议独立于高阶思维协议，两者可先后触发但分别执行
|
||||
- 本轮回答结束后，此模块退出上下文
|
||||
+53
@@ -0,0 +1,53 @@
|
||||
---
|
||||
id: KB-20260603-006
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-03
|
||||
date_modified: 2026-06-03
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-04---
|
||||
|
||||
# 2026-06-02_2026-06-01_2026-05-17_2026-05-16_knowledge-review
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-06-01_2026-05-17_2026-05-16_knowledge-review
|
||||
# 2026-05-17_2026-05-16_knowledge-review
|
||||
# 2026-05-16_knowledge-review
|
||||
# knowledge-review
|
||||
# 知识审查流程
|
||||
---
|
||||
## 审查四步
|
||||
### 1. 来源可靠性评分
|
||||
- 官方文档 / 学术论文：+0.3
|
||||
- 知名技术博客 / 行业报告：+0.1
|
||||
- 个人博客 / 论坛帖子：-0.1
|
||||
- 未知来源 / 社交媒体：-0.3
|
||||
- 基础可靠性基线：0.5
|
||||
### 2. 一致性检查
|
||||
- 完全一致 → 旧知识 confidence 轻微提升
|
||||
- 补充细节 → 合并为新条目
|
||||
- 矛盾 → 进入冲突消解（新事实 > 旧事实，用户明确修正 > 系统推断）
|
||||
### 3. 时效性标记
|
||||
- 技术类知识：180 天后自动失效
|
||||
- 法律/法规：按原文有效期
|
||||
- 个人偏好/观点：不过期，标记为 opinion
|
||||
- 物理/数学公理：永久有效
|
||||
### 4. 日志记录
|
||||
---
|
||||
## 输出格式
|
||||
- 决策：[通过 / 待确认 / 拒绝]
|
||||
- 入库路径：[具体目录路径]
|
||||
+47
@@ -0,0 +1,47 @@
|
||||
---
|
||||
id: KB-20260603-007
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-03
|
||||
date_modified: 2026-06-03
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-04---
|
||||
|
||||
# 2026-06-02_2026-06-01_2026-05-17_2026-05-16_lip-sync-solution
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-06-01_2026-05-17_2026-05-16_lip-sync-solutions-compariso
|
||||
# 2026-05-17_2026-05-16_lip-sync-solutions-comparison
|
||||
# 2026-05-16_lip-sync-solutions-comparison
|
||||
# lip-sync-solutions-comparison
|
||||
# AI 视频唇形同步（Lip-Sync）方案调研与成本对比
|
||||
---
|
||||
## 一、需求背景
|
||||
- **省去** 视频生成后单独做口型对齐的步骤
|
||||
- 让数字人的口播更自然（语音与唇形天然匹配）
|
||||
- 部分方案支持从 **文本直接到带语音+口型的视频**（Text→Voice→Video 一步完成）
|
||||
---
|
||||
## 二、方案全景对比
|
||||
### 商业方案
|
||||
### 开源方案（可本地部署）
|
||||
---
|
||||
## 三、Kling LipSync 详细分析
|
||||
### API 端点
|
||||
### 输入要求
|
||||
### 处理时间与成本
|
||||
- **固定推理时间**：~12 分钟/次（不随视频时长变化）
|
||||
- **成本**：**$0.014 / 5秒**（按 5 秒单位向上取整）
|
||||
+51
@@ -0,0 +1,51 @@
|
||||
---
|
||||
id: KB-20260603-008
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-03
|
||||
date_modified: 2026-06-03
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-04---
|
||||
|
||||
# 2026-06-02_2026-06-01_2026-05-17_2026-05-16_meta-thinking
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-06-01_2026-05-17_2026-05-16_meta-thinking
|
||||
# 2026-05-17_2026-05-16_meta-thinking
|
||||
# 2026-05-16_meta-thinking
|
||||
# meta-thinking
|
||||
# 高阶思维协议
|
||||
---
|
||||
## 执行步骤
|
||||
1. **问题本质**
|
||||
2. **前提挑战**
|
||||
**绝不无条件接受用户的预设。**
|
||||
3. **知识溯源**
|
||||
- 优先级1：L2 facts.db 中是否有相关历史洞见？
|
||||
- 优先级2：03_knowledge/10_concepts/ 下是否有可类比的跨领域模型？
|
||||
- 优先级3：03_knowledge/40_references/ 中是否有经典理论支撑？
|
||||
- 若仅为个人推断，须明确声明置信度。
|
||||
4. **反直觉锚点**
|
||||
---
|
||||
## 输出结构
|
||||
**问题本质**: [一句话点明本质]
|
||||
**前提挑战**: [指出现有思路的潜在缺陷]
|
||||
**核心建议**: [立足第一性原理的、高站位的建议]
|
||||
**认知钩子**: [提供一个反直觉视角或思想火花]
|
||||
---
|
||||
## 约束
|
||||
- 找不到高价值洞察时，诚实说"本次无更高视角"，不编造
|
||||
+46
@@ -0,0 +1,46 @@
|
||||
---
|
||||
id: KB-20260603-009
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-03
|
||||
date_modified: 2026-06-03
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-04---
|
||||
|
||||
# 2026-06-02_2026-06-01_2026-05-17_2026-05-16_stuck-interventio
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-06-01_2026-05-17_2026-05-16_stuck-intervention
|
||||
# 2026-05-17_2026-05-16_stuck-intervention
|
||||
# 2026-05-16_stuck-intervention
|
||||
# stuck-intervention
|
||||
# 卡壳干预模板
|
||||
---
|
||||
## 执行步骤
|
||||
1. 立即停止当前尝试，不自行继续
|
||||
2. 写入 04_memory/logs/errors.log（格式: timestamp | context | error_type | attempted_solutions）
|
||||
3. 按以下模板组织输出
|
||||
---
|
||||
## 输出模板
|
||||
**已尝试方案**:
|
||||
- 方案A: [简述] → 失败原因: [简述]
|
||||
- 方案B: [简述] → 失败原因: [简述]
|
||||
**候选方案**:
|
||||
- **A)** [方案简述] — 预估耗时: [X分钟] — 风险: [简述]
|
||||
- **B)** [方案简述] — 预估耗时: [X分钟] — 风险: [简述]
|
||||
**建议**: [倾向 A/B，简述理由]
|
||||
**等待你的决定** → 收到后按选定方案继续执行。
|
||||
+49
@@ -0,0 +1,49 @@
|
||||
---
|
||||
id: KB-20260603-010
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-03
|
||||
date_modified: 2026-06-03
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-04---
|
||||
|
||||
# 2026-06-02_2026-06-01_2026-05-17_2026-05-16_分镜时序与运镜_让AI_懂剪辑
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-06-01_2026-05-17_2026-05-16_分镜时序与运镜_让AI_懂剪辑
|
||||
# 2026-05-17_2026-05-16_分镜时序与运镜_让AI_懂剪辑
|
||||
# 2026-05-16_分镜时序与运镜_让AI_懂剪辑
|
||||
# 分镜时序与运镜：让AI'懂剪辑
|
||||
# Ch4: 分镜时序与运镜
|
||||
---
|
||||
## 一、结果推导
|
||||
## 二、多镜头切换的一致性方案
|
||||
### 核心问题
|
||||
### 方案一：提示词锚定法（零成本）
|
||||
**关键规则**：
|
||||
- 描述越具体越好（颜色/材质/年龄/体型）
|
||||
- 跨镜头复制相同文字（重复=一致性）
|
||||
- 添加统一风格锚点："电影级，35mm胶片拍摄，蓝绿和橙色调色"
|
||||
### 方案二：参考图锚定法（最可靠）
|
||||
1. 生成高质量角色参考图作为第一帧
|
||||
2. 每个新镜头都重新选择一张合适的参考图
|
||||
3. 提示词只描述运动，不再描述外貌（模型已从参考图看到）
|
||||
**最佳工具**：Seedance 2.0（身份锁定系统）、可灵3.0（分镜模式6个机位自动一致）
|
||||
### 方案三：分段锚定法（长视频）
|
||||
### 方案四：后期统一
|
||||
- 并排对比所有场景
|
||||
- 漂移严重的以锚定场景为参考重生成
|
||||
+45
@@ -0,0 +1,45 @@
|
||||
---
|
||||
id: KB-20260603-011
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-03
|
||||
date_modified: 2026-06-03
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-04---
|
||||
|
||||
# 2026-06-02_2026-06-01_2026-05-17_2026-05-16_原子化登录管理模块_auth_ma
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-06-01_2026-05-17_2026-05-16_原子化登录管理模块_auth_manager
|
||||
# 2026-05-17_2026-05-16_原子化登录管理模块_auth_manager
|
||||
# 2026-05-16_原子化登录管理模块_auth_manager
|
||||
# 原子化登录管理模块 auth_manager
|
||||
# auth_manager — 原子化登录管理模块
|
||||
## 概述
|
||||
## 核心函数
|
||||
### Cookie 检测（主方案）
|
||||
### DOM 检测（备选，仅移动端视口有效）
|
||||
### 综合检测
|
||||
# 返回: logged_in, cookie_ok, dom_ok, cookie_count, session_id, method
|
||||
### Cookie 持久化
|
||||
### 等待手动登录
|
||||
## 设计原则
|
||||
1. Cookie 检测为主方案（不依赖 UI 布局，桌面端/移动端均可用）
|
||||
2. DOM 检测为备选（仅在 mobile=True 视口下有效）
|
||||
3. 所有函数无副作用（不写 DB/文件，除非函数名明确）
|
||||
4. 统一异常处理（所有异常返回 False/空值）
|
||||
## 解决的问题
|
||||
+44
@@ -0,0 +1,44 @@
|
||||
---
|
||||
id: KB-20260603-012
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-03
|
||||
date_modified: 2026-06-03
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-04---
|
||||
|
||||
# 2026-06-02_2026-06-01_2026-05-17_2026-05-16_统一升级引擎_agentos_up
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-06-01_2026-05-17_2026-05-16_统一升级引擎_agentos_upgrade
|
||||
# 2026-05-17_2026-05-16_统一升级引擎_agentos_upgrade
|
||||
# 2026-05-16_统一升级引擎_agentos_upgrade
|
||||
# 统一升级引擎 agentos upgrade
|
||||
# agentos upgrade — 统一升级引擎
|
||||
## 概述
|
||||
## 命令
|
||||
## 执行流程
|
||||
1. `git pull origin main` — 拉取最新代码
|
||||
2. 扫描 18 个模块（8 个 skill + 8 个 tool + 2 个其他）
|
||||
3. 按模块层级执行不同操作
|
||||
4. 自动注册技能到 WorkBuddy
|
||||
5. 输出升级报告
|
||||
## 三层模块管理
|
||||
## Python 环境统一
|
||||
1. `AGENTOS_PYTHON` 环境变量
|
||||
2. `~/.workbuddy/binaries/python/envs/agent-os/bin/python3`
|
||||
3. `~/.workbuddy/binaries/python/versions/3.13.12/bin/python3`
|
||||
+41
@@ -0,0 +1,41 @@
|
||||
---
|
||||
id: KB-20260603-013
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-03
|
||||
date_modified: 2026-06-03
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-04---
|
||||
|
||||
# 2026-06-02_2026-06-01_2026-05-17_AI提示词驱动文生视频__AICG造梦局10条视频知识提
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-06-01_2026-05-17_AI提示词驱动文生视频__AICG造梦局10条视频知识提取
|
||||
# 2026-05-17_AI提示词驱动文生视频__AICG造梦局10条视频知识提取
|
||||
# AI提示词驱动文生视频——AICG造梦局10条视频知识提取
|
||||
# AI提示词驱动文生视频——知识合集
|
||||
---
|
||||
## 1. Image2+Sd2.0故事版玩法
|
||||
### 📋 操作步骤
|
||||
1. 视频中演示的具体操作流程，分步骤详细记录
|
||||
### ⚙️ 关键参数
|
||||
- 视频中提到的工具名称、参数设置、具体数字等
|
||||
- 平台：B 站
|
||||
- 盘点对象：Top 10 大模型学习 UP 主
|
||||
- 涉及 UP 主名称：deepthoughts、Coggle 数据科学、王木头学科学、良睦路程序员、霹雳吧啦 Wz、跟李沐学 AI
|
||||
- 适用人群：资深开发、技术发烧友、视觉算法工程师、全栈开发者、AI 技术爱好者、职场研发人员、职场技术人员、数学基础薄弱的初学者、有 Python 基础的开发工程师、非科班新手、产品运营、后端工程师、算法工程师、Python 开发、测试工程师、零基础 AI 入门者、计算机专业学生、AI 研发工程师
|
||||
- 技术方向：高阶源码解析、多模态大模型实战、前沿技术解读、新款模型与新技术解读、数学原理讲解、工程开发、热门应用、模型量化、本地部署、大模型全栈实战
|
||||
+48
@@ -0,0 +1,48 @@
|
||||
---
|
||||
id: KB-20260603-014
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-03
|
||||
date_modified: 2026-06-03
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-04---
|
||||
|
||||
# 2026-06-02_2026-06-01_2026-05-17_AI提示词驱动文生视频__博主干货提取_10_11条_
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-06-01_2026-05-17_AI提示词驱动文生视频__博主干货提取_10_11条_
|
||||
# 2026-05-17_AI提示词驱动文生视频__博主干货提取_10_11条_
|
||||
# AI提示词驱动文生视频——博主干货提取（10/11条）
|
||||
# AI提示词驱动文生视频——博主干货
|
||||
---
|
||||
## 1. 反着写提示词AI人物更真实
|
||||
---
|
||||
## 2. 明非屠龙可不是件容易事
|
||||
---
|
||||
## 3. 或许ai也有遗憾
|
||||
---
|
||||
## 4. 骨相留白让AI人物更有少年气
|
||||
---
|
||||
## 5. 劳动人民的伟力
|
||||
---
|
||||
## 6. 伟大出自平凡
|
||||
---
|
||||
## 7. 空间锚定画面更有张力
|
||||
---
|
||||
## 8. 衰仔要走咯
|
||||
---
|
||||
## 9. 我的名字叫赫尔佐格
|
||||
+47
@@ -0,0 +1,47 @@
|
||||
---
|
||||
id: KB-20260603-015
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-03
|
||||
date_modified: 2026-06-03
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-04---
|
||||
|
||||
# 2026-06-02_2026-06-01_2026-05-17_AI提示词驱动文生视频__李导_Agent版_10条视频
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-06-01_2026-05-17_AI提示词驱动文生视频__李导_Agent版_10条视频知识提取
|
||||
# 2026-05-17_AI提示词驱动文生视频__李导_Agent版_10条视频知识提取
|
||||
# AI提示词驱动文生视频——李导（Agent版）10条视频知识提取
|
||||
# AI提示词驱动文生视频——李导（Agent版）知识合集
|
||||
---
|
||||
## 1. 60提示词做 Ai 短视频第三讲 用三分法构图控制主体精准站位
|
||||
### 完整字幕
|
||||
### 提示词/指令原文
|
||||
### 操作步骤
|
||||
1. 确定画面主体与人物站位需求
|
||||
2. 采用三分法构图，将画面九等分
|
||||
3. 把主体放置在三分法的四个视觉锚点上
|
||||
4. 结合景别控制，调整主体的远近与大小
|
||||
5. 多人物场景时，用三分法规定每个人的位置，用景别控制远近层次
|
||||
### 关键参数
|
||||
### 核心技巧
|
||||
1. 用三分法构图实现 AI 画面主体精准站位，解决自然语言命令不准的问题
|
||||
2. 四个交叉点为视觉锚点，是画面视觉重心舒适区
|
||||
3. 构图 + 景别结合，精准控制主体位置、远近、大小
|
||||
4. 多人物场景下，用三分法规定每个人位置，用景别控制层次
|
||||
5. 摄影构图对影片叙事表达起重要作用
|
||||
+51
@@ -0,0 +1,51 @@
|
||||
---
|
||||
id: KB-20260603-016
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-03
|
||||
date_modified: 2026-06-03
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-04---
|
||||
|
||||
# 2026-06-02_2026-06-01_2026-05-17_AI视频提示词系统___采集队列
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-06-01_2026-05-17_AI视频提示词系统___采集队列
|
||||
# 2026-05-17_AI视频提示词系统___采集队列
|
||||
# AI视频提示词系统 · 采集队列
|
||||
# 采集队列
|
||||
---
|
||||
## P0 · 立即采集
|
||||
### 人物微表情（Ch7）
|
||||
- [ ] 抖音搜索"AI 微表情 控制 提示词"
|
||||
- [ ] 抖音搜索"AI 角色情绪 控制 教程"
|
||||
- [ ] B站搜索"AI动画 微表情 参数"
|
||||
### 背景音乐/音频设计（Ch6）
|
||||
- [ ] 抖音搜索"AI视频 配乐 提示词"
|
||||
- [ ] 搜索"AI视频 音效 生成 教程"
|
||||
- [ ] B站搜索"AI 音频 设计 工作流"
|
||||
### 大模型对比（Ch12）
|
||||
- [ ] Runway 最新教程与提示词技巧
|
||||
- [ ] Pika 风格一致性方案
|
||||
- [ ] Gen-2 参考图+提示词融合
|
||||
## P1 · 本周采集
|
||||
### 脚本叙事（Ch14）
|
||||
- [ ] 小说→脚本的AI适配教程
|
||||
- [ ] 场景控制的参数化描述
|
||||
- [ ] AI 叙事结构的案例分析
|
||||
### 动漫风格（Ch9）
|
||||
- [ ] 抖音搜索"AI动漫 制作 提示词 教程"
|
||||
+51
@@ -0,0 +1,51 @@
|
||||
---
|
||||
id: KB-20260603-017
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-03
|
||||
date_modified: 2026-06-03
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-04---
|
||||
|
||||
# 2026-06-02_2026-06-01_2026-05-17_AI视频提示词系统创作教科书___知识体系索引
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-06-01_2026-05-17_AI视频提示词系统创作教科书___知识体系索引
|
||||
# 2026-05-17_AI视频提示词系统创作教科书___知识体系索引
|
||||
# AI视频提示词系统创作教科书 · 知识体系索引
|
||||
# AI 视频提示词系统创作教科书 · 知识体系
|
||||
---
|
||||
## 体系总览
|
||||
### 现有知识关联
|
||||
---
|
||||
## 第一部分: 底层逻辑与核心原则（基石篇·结果导向）
|
||||
### Ch1: AI视频生成的"执行本质"与结果推导逻辑
|
||||
**核心问题**: AI 不是读心术，指令的明确性决定输出上限。
|
||||
**结果推导路径**:
|
||||
**采集方向**:
|
||||
- [ ] AI 视频生成工具的底层原理（文本→视频的映射机制）
|
||||
- [ ] 人类创作逻辑 vs AI 执行逻辑的差异分析
|
||||
- [ ] 结果推导方法论的具体案例（多种场景）
|
||||
- [ ] 提示词信息传递模型的图解
|
||||
### Ch2: 结果推导方法论
|
||||
**核心问题**: 如何从目标画面反推提示词锚点。
|
||||
**采集方向**:
|
||||
- [ ] 多场景结果推导案例库
|
||||
- [ ] 情绪→视觉元素的映射表
|
||||
- [ ] 目标反推的思维训练方法
|
||||
---
|
||||
## 第二部分: 核心模块与创作逻辑（技术篇·从结果定义技术）
|
||||
+48
@@ -0,0 +1,48 @@
|
||||
---
|
||||
id: KB-20260603-021
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-03
|
||||
date_modified: 2026-06-03
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-04---
|
||||
|
||||
# 2026-06-02_2026-06-01_2026-05-17_ch01-core-logic
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-06-01_2026-05-17_ch01-core-logic
|
||||
# 2026-05-17_ch01-core-logic
|
||||
# ch01-core-logic
|
||||
# Ch1: AI视频生成的"执行本质"与结果推导逻辑
|
||||
---
|
||||
## 一、底层逻辑：从"脑内画面"到"机器指令"
|
||||
### 1.1 信息传递模型
|
||||
### 1.2 提示词基础公式
|
||||
- **观察视角**：告诉 AI 从哪个角度看
|
||||
- **情感基调**：告诉 AI 什么情绪氛围
|
||||
- **时间线索**：告诉 AI 变化过程
|
||||
### 1.3 AI不是全能助手，是精确执行者
|
||||
### 1.4 为什么AI需要明确指令
|
||||
1. **缺乏常识和联想能力**：输入"一个人走路"，AI可能生成任何场景
|
||||
2. **信息过载阈值**：超过5个核心信息、3组动态元素、10个修饰词 → AI无法聚焦
|
||||
3. **时空错乱**：缺少明确的空间/时间过渡描述 → 画面跳跃
|
||||
4. **文化差异**：AI训练数据的文化背景可能与你不同
|
||||
---
|
||||
## 二、提示词信息传递模型
|
||||
### 2.1 逆向编码闭环
|
||||
### 2.2 核心原则
|
||||
- **不要告诉AI你想要什么，而是教会AI如何用它的方式构建出你想要的画面**
|
||||
+52
@@ -0,0 +1,52 @@
|
||||
---
|
||||
id: KB-20260603-022
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-03
|
||||
date_modified: 2026-06-03
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-04---
|
||||
|
||||
# 2026-06-02_2026-06-01_2026-05-17_ch02-result-driven
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-06-01_2026-05-17_ch02-result-driven
|
||||
# 2026-05-17_ch02-result-driven
|
||||
# ch02-result-driven
|
||||
# Ch2: 结果推导方法论——从目标画面反推指令锚点
|
||||
---
|
||||
## 一、三大核心步骤
|
||||
### 步骤1：视觉要素深度拆解
|
||||
**主体要素**
|
||||
- 形态特征：八头身比例、肌肉线条分明
|
||||
- 材质表现：磨砂金属、半透明硅胶
|
||||
- 动态捕捉：跳跃瞬间、手指弯曲45度
|
||||
- 表情管理：眼角上扬15度
|
||||
**环境系统**
|
||||
- 空间层次：前中背景的元素分布
|
||||
- 物理特性：重力、光照、大气折射
|
||||
- 时间维度：季节、昼夜、天气
|
||||
- 文化符号：特定时代的建筑/服饰
|
||||
**光影工程**
|
||||
- 光源位置：钟表定位法（3点钟方向顶光）
|
||||
- 光质控制：硬光(直射) vs 柔光(漫反射)
|
||||
- 色温调节：2000K到10000K
|
||||
- 光比计算：主光与辅光比值（光比4:1）
|
||||
### 步骤2：结构化提示词构建
|
||||
**权重应用**：
|
||||
- 权重越大，AI对该要素越重视
|
||||
- 艺术风格编码：`cinematic, bokeh`（摄影）、`octane render, PBR`（3D）
|
||||
+53
@@ -0,0 +1,53 @@
|
||||
---
|
||||
id: KB-20260603-023
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-03
|
||||
date_modified: 2026-06-03
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-04---
|
||||
|
||||
# 2026-06-02_2026-06-01_2026-05-17_ch06-audio-design
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-06-01_2026-05-17_ch06-audio-design
|
||||
# 2026-05-17_ch06-audio-design
|
||||
# ch06-audio-design
|
||||
# Ch6: 音频设计——让"声音"提升质感
|
||||
---
|
||||
## 核心知识
|
||||
### 音频三角
|
||||
### 背景音乐：情绪匹配
|
||||
### 卡点原则
|
||||
- 快镜头 → 配乐 120BPM+，鼓点清晰，每拍一个转场
|
||||
- 慢镜头 → 配乐 60BPM，旋律舒缓，4拍一个转场
|
||||
- 场景切换 → 配乐节奏变化或乐器变化做标记
|
||||
### 环境音分类
|
||||
### 对白参数
|
||||
- 音色选择：低沉/清脆/沙哑/温暖
|
||||
- 语速：快(紧张)/中(叙述)/慢(深沉)
|
||||
- 音量：对白通常比背景音乐高 10-15dB
|
||||
- 混响：小空间(无混响)/大厅(短混响)/空旷(长混响)
|
||||
### 声音的叙事性
|
||||
- 紧张感构建：环境音渐强 + 配乐节奏加快 + 对白断句变短
|
||||
- 情绪转折：配乐突然停止 → 静默2秒 → 换曲入场
|
||||
- 场景切换：环境音交叉渐变（前一场景音渐弱+后一场景音渐强）
|
||||
---
|
||||
## 已采集素材
|
||||
### 抖音音频教程素材
|
||||
- AI唱歌保姆级教程
|
||||
- AI配音音色提示词
|
||||
+45
@@ -0,0 +1,45 @@
|
||||
---
|
||||
id: KB-20260603-024
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-03
|
||||
date_modified: 2026-06-03
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-04---
|
||||
|
||||
# 2026-06-02_2026-06-01_2026-05-17_ch07-micro-expression
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-06-01_2026-05-17_ch07-micro-expression
|
||||
# 2026-05-17_ch07-micro-expression
|
||||
# ch07-micro-expression
|
||||
# Ch7: 人物塑造-微表情与情绪控制
|
||||
---
|
||||
## 核心知识
|
||||
### 情绪→微表情映射
|
||||
### 三招让角色活过来（来源：抖音采集）
|
||||
**1. 控制情绪强度**
|
||||
- 避免 AI 默认情绪最大值（容易假）
|
||||
- 在提示词中加入强度修饰：`slightly`, `barely perceptible`, `subtle`
|
||||
- 示例：`a faint smile barely lifting the corners of his mouth`
|
||||
**2. 给微表情一个发生原因**
|
||||
- 微表情不能孤立存在，需要与场景/对白联动
|
||||
- 加入触发条件的描述：`upon hearing the news, her eyes widened`
|
||||
- 加上半身动作，让画面脱离摆拍感
|
||||
**3. 给情绪加时间顺序**
|
||||
- 描述情绪的变化过程：`from confusion to realization`
|
||||
- 示例：`His expression shifted from doubt to shock as the truth sank in`
|
||||
+53
@@ -0,0 +1,53 @@
|
||||
---
|
||||
id: KB-20260603-025
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-03
|
||||
date_modified: 2026-06-03
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-04---
|
||||
|
||||
# 2026-06-02_2026-06-01_2026-05-17_ch08-film-style
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-06-01_2026-05-17_ch08-film-style
|
||||
# 2026-05-17_ch08-film-style
|
||||
# ch08-film-style
|
||||
# Ch8: 电影风格AI视频创作
|
||||
---
|
||||
## 一、电影级提示词核心要素
|
||||
### 六要素结构（来源：AI视频提示词工程2026指南）
|
||||
### 迭代法（从简到繁）
|
||||
---
|
||||
## 二、电影美学风格库
|
||||
### 邵氏电影美学（来源：既有知识）
|
||||
**三大风格要素**：
|
||||
**核心关键词**：
|
||||
### 经典电影风格参考
|
||||
---
|
||||
## 三、电影级参数
|
||||
### 镜头参数
|
||||
### 常用电影质感情号
|
||||
---
|
||||
## 四、分镜与叙事
|
||||
### 好莱坞经典三幕式
|
||||
### 景别→情绪映射（来自李导V4）
|
||||
---
|
||||
## 五、已有知识关联
|
||||
- [[20_methods/shaoshi-film-aesthetics-prompts\|邵氏电影美学提示词技巧]] — 胶片/光影/置景三大风格要素
|
||||
- [[20_methods/li-dao-ai-video-prompts\|李导AI提示词全集]] — 景别/视角/构图/视听语言
|
||||
- [[20_methods/ai-prompt-video-knowledge\|AICG造梦局知识合集]] — 多种风格化视频案例
|
||||
+48
@@ -0,0 +1,48 @@
|
||||
---
|
||||
id: KB-20260603-026
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-03
|
||||
date_modified: 2026-06-03
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-04---
|
||||
|
||||
# 2026-06-02_2026-06-01_2026-05-17_ch09-anime-style
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-06-01_2026-05-17_ch09-anime-style
|
||||
# 2026-05-17_ch09-anime-style
|
||||
# ch09-anime-style
|
||||
# Ch9: 动漫/动画风格AI视频创作
|
||||
---
|
||||
## 核心知识
|
||||
### 角色生成万能公式
|
||||
### 10种风格提示词模板
|
||||
### 角色一致性（三视图法）
|
||||
1. **标准全身照**
|
||||
- 公式：[角色描述]，正面全身站立标准照，纯白背景，9:16
|
||||
- 目的：锁定体型、服饰、比例
|
||||
2. **多角度三视图**
|
||||
- 公式：[角色描述]，正面+侧面+背面全身照+面部特写，16:9
|
||||
- 目的：为 AI 提供全方位"建模参考"
|
||||
3. **质感升级**
|
||||
- 加入"次表面散射(SSS)"或"BJD质感" → 告别塑料皮
|
||||
- 加入"轮廓光"+"丁达尔效应"+"强烈明暗对比" → 空间深度
|
||||
### 动漫分镜的提示词适配
|
||||
### AI漫剧完整工作流
|
||||
---
|
||||
## 采集来源
|
||||
+53
@@ -0,0 +1,53 @@
|
||||
---
|
||||
id: KB-20260603-027
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-03
|
||||
date_modified: 2026-06-03
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-04---
|
||||
|
||||
# 2026-06-02_2026-06-01_2026-05-17_ch10-commercial-tvc
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-06-01_2026-05-17_ch10-commercial-tvc
|
||||
# 2026-05-17_ch10-commercial-tvc
|
||||
# ch10-commercial-tvc
|
||||
# Ch10: 商业广告/TVC创作
|
||||
---
|
||||
## 一、AI-TVC完整工作流
|
||||
### 5步出品（20分钟内）
|
||||
**工具链**：
|
||||
### 核心：运镜提示词
|
||||
---
|
||||
## 二、TVC提示词结构
|
||||
### 标准格式
|
||||
### 示例：像素风TVC
|
||||
### 玉石品牌TVC（来自既有知识）
|
||||
- 结合中式色彩美学
|
||||
- 电影感构图
|
||||
- 专业LUT调色提升品质
|
||||
- Seedance 2.0 生成电影级画面
|
||||
---
|
||||
## 三、商业广告类型与策略
|
||||
### 高级技巧
|
||||
1. **图片先行**：先用高质量生图工具打好画面基础
|
||||
2. **动作提示词含运镜**：让静态图"活"起来的关键
|
||||
3. **后处理统一调性**：剪映统一调色+LUT
|
||||
4. **多次迭代**：每次只改5%内容AB测试
|
||||
---
|
||||
## 采集来源
|
||||
+53
@@ -0,0 +1,53 @@
|
||||
---
|
||||
id: KB-20260603-028
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-03
|
||||
date_modified: 2026-06-03
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-04---
|
||||
|
||||
# 2026-06-02_2026-06-01_2026-05-17_ch11-sci-fi-fantasy
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-06-01_2026-05-17_ch11-sci-fi-fantasy
|
||||
# 2026-05-17_ch11-sci-fi-fantasy
|
||||
# ch11-sci-fi-fantasy
|
||||
# Ch11: 科幻/奇幻风格创作
|
||||
---
|
||||
## 一、科幻风格提示词大全
|
||||
### 提示词结构公式
|
||||
### 赛博朋克城市
|
||||
**完整模板**：
|
||||
**关键元素**：
|
||||
### 太空场景
|
||||
**空间站**：
|
||||
**外星景观**：
|
||||
### 科技与机械
|
||||
**人形机器人**：
|
||||
---
|
||||
## 二、风格参数
|
||||
### 光影参数
|
||||
### 美术风格
|
||||
### 风格参考锚点
|
||||
---
|
||||
## 三、奇幻风格创作
|
||||
### 通用奇幻模板
|
||||
### 经典奇幻类型
|
||||
### 常见错误与修正
|
||||
---
|
||||
## 采集来源
|
||||
+43
@@ -0,0 +1,43 @@
|
||||
---
|
||||
id: KB-20260603-029
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-03
|
||||
date_modified: 2026-06-03
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-04---
|
||||
|
||||
# 2026-06-02_2026-06-01_2026-05-17_ch12-model-comparison
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-06-01_2026-05-17_ch12-model-comparison
|
||||
# 2026-05-17_ch12-model-comparison
|
||||
# ch12-model-comparison
|
||||
# Ch12: 主流AI视频大模型的差异与适配
|
||||
---
|
||||
## 核心数据：平台对比
|
||||
## 模型版本演进
|
||||
## 结果导向选型
|
||||
## 工具链协同
|
||||
- 商业广告：Gen-2(画面) + Au(声音)
|
||||
- 动漫番剧：Pika(多镜头) + 配音平台(对白)
|
||||
- 电影短片：Kling 3.0(素材) + Runway(后期)
|
||||
---
|
||||
## 已采集素材
|
||||
### 抖音大模型对比素材
|
||||
---
|
||||
## 采集来源
|
||||
+51
@@ -0,0 +1,51 @@
|
||||
---
|
||||
id: KB-20260603-030
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-03
|
||||
date_modified: 2026-06-03
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-04---
|
||||
|
||||
# 2026-06-02_2026-06-01_2026-05-17_ch13-tool-workflow
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-06-01_2026-05-17_ch13-tool-workflow
|
||||
# 2026-05-17_ch13-tool-workflow
|
||||
# ch13-tool-workflow
|
||||
# Ch13: 工具实操与工作流搭建
|
||||
---
|
||||
## 一、四大核心工具详解
|
||||
### Runway Gen-3 Alpha
|
||||
**定位**：电影级画质、精细控制，专业制作首选
|
||||
**操作流程**：
|
||||
1. 准备高质量静态图（建议1080p以上，最好4K）
|
||||
2. 上传图片至平台（支持JPG/PNG）
|
||||
3. 编写提示词描述期望的动态效果
|
||||
4. 调整 Camera Motion 参数和 Motion Bucket
|
||||
5. 生成预览（耗时2-5分钟）
|
||||
**提示词结构**：
|
||||
**参数**：
|
||||
- Motion Bucket：0-10（控制运动强度）
|
||||
- Camera Motion：Dolly In/Out、Tracking Shot、Orbit、Bird's Eye
|
||||
- 时长上限：16秒
|
||||
- 分辨率：最高4K
|
||||
---
|
||||
### Kling 可灵
|
||||
**定位**：中文理解最佳、性价比高
|
||||
**操作流程**：
|
||||
1. 准备1080p以上静态图
|
||||
+56
@@ -0,0 +1,56 @@
|
||||
---
|
||||
id: KB-20260603-031
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-03
|
||||
date_modified: 2026-06-03
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-04---
|
||||
|
||||
# 2026-06-02_2026-06-01_2026-05-17_ch14-script-to-prompt
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-06-01_2026-05-17_ch14-script-to-prompt
|
||||
# 2026-05-17_ch14-script-to-prompt
|
||||
# ch14-script-to-prompt
|
||||
# Ch14: 从"创意"到"AI友好型脚本"的转化
|
||||
---
|
||||
## 一、四步工作流：小说→分镜脚本
|
||||
### 核心原理
|
||||
### 四步流程
|
||||
**第一步：规范化整理内容**
|
||||
**第二步：预估镜头数与结构分布**
|
||||
- 需要多少镜头？
|
||||
- 镜头在关键情节中的分布？
|
||||
- 建立视觉节奏感
|
||||
**第三步：搭建核心场景与角色表**
|
||||
- 场景1：咸亨酒店大堂，午后，光线昏暗，柜台有酒坛
|
||||
- 角色A：孔乙己，瘦弱，长衫破旧
|
||||
**第四步：整合生成分镜脚本**
|
||||
**效果对比**（以《孔乙己》为例）：
|
||||
- 直接转换：仅12个镜头
|
||||
- 四步工作流：59个分镜头 ✅ 完整还原剧情
|
||||
### 提示词模板
|
||||
---
|
||||
## 二、场景控制技术
|
||||
### 先建库，再使用
|
||||
### 场景描述要素
|
||||
### 场景互动
|
||||
---
|
||||
## 三、AI友好型脚本格式
|
||||
### 模块化结构
|
||||
### 完整示例
|
||||
+48
@@ -0,0 +1,48 @@
|
||||
---
|
||||
id: KB-20260603-032
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-03
|
||||
date_modified: 2026-06-03
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-04---
|
||||
|
||||
# 2026-06-02_2026-06-01_2026-05-17_ch15-narrative-structure
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-06-01_2026-05-17_ch15-narrative-structure
|
||||
# 2026-05-17_ch15-narrative-structure
|
||||
# ch15-narrative-structure
|
||||
# Ch15: 叙事结构与AI叙事能力的结合
|
||||
---
|
||||
## 常见叙事结构
|
||||
## AI叙事优势与局限
|
||||
### AI擅长
|
||||
- 视觉化情绪（光影+微表情+场景）
|
||||
- 高密度画面叙事（每一秒承载大量信息）
|
||||
- 风格化渲染（电影感/动漫感）
|
||||
### AI不擅长
|
||||
- 复杂对白逻辑（容易混淆人物关系）
|
||||
- 细微情感递进（缺少层次感）
|
||||
- 长时间连贯性（超过30秒容易崩）
|
||||
### 策略：扬长避短
|
||||
## 短剧分镜7大要素
|
||||
1. 景别（远全中近特）
|
||||
2. 视角（平俯仰倾）
|
||||
3. 构图（三分法/对称/不对称）
|
||||
4. 光影（情绪映射）
|
||||
5. 运动（镜头+主体）
|
||||
+48
@@ -0,0 +1,48 @@
|
||||
---
|
||||
id: KB-20260603-033
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-03
|
||||
date_modified: 2026-06-03
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-04---
|
||||
|
||||
# 2026-06-02_2026-06-01_2026-05-17_ch16-practice-path
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-06-01_2026-05-17_ch16-practice-path
|
||||
# 2026-05-17_ch16-practice-path
|
||||
# ch16-practice-path
|
||||
# Ch16: 从"模仿"到"创新"的实践路径
|
||||
---
|
||||
## 三个阶段
|
||||
### 阶段1：爆款拆解（入门）
|
||||
- 分析抖音/B站AI视频爆款的"结果→提示词→工具"链路
|
||||
- 工具：video-to-prompt.com（自动反推提示词）
|
||||
- 目标：能复现同级别效果
|
||||
### 阶段2：垂直深耕（进阶）
|
||||
- 聚焦一个领域（如AI漫剧/AI电影感/AI广告）
|
||||
- 打磨"成果稳定"的提示词模板
|
||||
- 建立自己的风格关键词库
|
||||
- 目标：每次生成都能稳定输出
|
||||
### 阶段3：风格创新（高级）
|
||||
- 融合多领域元素（如赛博朋克+古风）
|
||||
- 打造"差异化成果"的AI视频风格
|
||||
- 建立个人方法论体系
|
||||
- 目标：一眼能认出是你的风格
|
||||
## 成长路径
|
||||
## 效率指标
|
||||
+38
@@ -0,0 +1,38 @@
|
||||
---
|
||||
id: KB-20260603-034
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-03
|
||||
date_modified: 2026-06-03
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-04---
|
||||
|
||||
# 2026-06-02_2026-06-01_2026-05-17_ch17-resource-library
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-06-01_2026-05-17_ch17-resource-library
|
||||
# 2026-05-17_ch17-resource-library
|
||||
# ch17-resource-library
|
||||
# Ch17: 资源库建设——按成果类型分类的知识弹药库
|
||||
---
|
||||
## 平台资源
|
||||
## 工具资源
|
||||
## 提示词资源
|
||||
- PromptMart（50000+提示词库）
|
||||
- video-to-prompt.com（视频反推提示词）
|
||||
- 本知识库各章节（不断积累）
|
||||
## 案例资源
|
||||
+52
@@ -0,0 +1,52 @@
|
||||
---
|
||||
id: KB-20260603-035
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-03
|
||||
date_modified: 2026-06-03
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-04---
|
||||
|
||||
# 2026-06-02_2026-06-01_2026-05-17_ch18-problem-solving
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-06-01_2026-05-17_ch18-problem-solving
|
||||
# 2026-05-17_ch18-problem-solving
|
||||
# ch18-problem-solving
|
||||
# Ch18: 常见问题解决方案库
|
||||
---
|
||||
## 核心问题分类
|
||||
### A. 画面类问题
|
||||
#### A1. 人脸变形/身份漂移
|
||||
**提示词修复**：
|
||||
#### A2. 物体漂移/元素消失
|
||||
**提示词修复**：
|
||||
#### A3. 光影闪烁/曝光变化
|
||||
**后期补救**：调色工具统一曝光。
|
||||
#### A4. 画面风格跑偏
|
||||
- **检查参考示例**是否明确
|
||||
- **换适配大模型**（如写实→Kling，动漫→Pika）
|
||||
- 提示词加入风格锚点：`cinematic, photorealistic, anime style`
|
||||
---
|
||||
### B. 运动类问题
|
||||
#### B1. 不自然动作/飘忽运动
|
||||
#### B2. 手部与手指问题
|
||||
- **裁剪手部**或让其失焦
|
||||
- 使用简单手部姿势（握拳、张开手掌）
|
||||
- **避免手部特写**，使用广角镜头
|
||||
- **多次生成并挑选**最佳版本
|
||||
#### B3. 时间不一致性（"梦境"效果）
|
||||
+43
@@ -0,0 +1,43 @@
|
||||
---
|
||||
id: KB-20260603-036
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-03
|
||||
date_modified: 2026-06-03
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-04---
|
||||
|
||||
# 2026-06-02_2026-06-01_2026-05-17_ch19-trends
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-06-01_2026-05-17_ch19-trends
|
||||
# 2026-05-17_ch19-trends
|
||||
# ch19-trends
|
||||
# Ch19: 行业趋势与前沿探索
|
||||
---
|
||||
## AI视频技术演进方向
|
||||
## 提示词的"未来形态"
|
||||
## 需要关注的前沿方向
|
||||
1. **Sora级别模型**：物理世界理解能力大幅提升
|
||||
2. **实时AI视频编辑**：生成后可逐帧修改
|
||||
3. **个性化模型微调**：用自己的数据训练专属模型
|
||||
4. **AI视频+AR/VR**：沉浸式内容创作
|
||||
## 适配策略
|
||||
- 保持关注各平台更新日志
|
||||
- 定期测试新模型能力
|
||||
- 积累跨模态提示词经验
|
||||
- 建立自己的训练数据集
|
||||
+50
@@ -0,0 +1,50 @@
|
||||
---
|
||||
id: KB-20260603-037
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-03
|
||||
date_modified: 2026-06-03
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-04---
|
||||
|
||||
# 2026-06-02_2026-06-01_2026-05-17_video-shooting-cool-techniqu
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-06-01_2026-05-17_video-shooting-cool-techniques
|
||||
# 2026-05-17_video-shooting-cool-techniques
|
||||
# video-shooting-cool-techniques
|
||||
# 视频拍摄酷炫技巧完全手册——转场·灯光·动作衔接
|
||||
---
|
||||
## 第一部分：转场衔接——让视频"丝滑"的10大技巧
|
||||
### 1.1 相似动作转场
|
||||
**操作**：
|
||||
- 模特在两个场景做相同姿势（转身、挥手、跳跃）
|
||||
- 保持相同构图和拍摄角度
|
||||
- 直接硬切，不加转场
|
||||
**效果**：爆款变装、瞬间场景切换
|
||||
### 1.2 黑色物体转场
|
||||
**操作**：
|
||||
1. 前镜头结尾：镜头推进直到画面被黑色物体完全遮挡
|
||||
2. 后镜头开头：从同一黑色物体向后拉出
|
||||
**适用**：需要极简过渡的画面切换
|
||||
### 1.3 甩镜头转场（Whip Pan）
|
||||
**操作**：前结尾快速右甩 → 后开头从右快速甩入
|
||||
**效果**：穿越时空、快速地点切换
|
||||
**注意**：两段速度方向必须一致
|
||||
### 1.4 下坠/坠落转场
|
||||
**操作**：
|
||||
- 前镜头：物体下落，镜头跟随向下
|
||||
+41
@@ -0,0 +1,41 @@
|
||||
---
|
||||
id: KB-20260603-038
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-03
|
||||
date_modified: 2026-06-03
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-04---
|
||||
|
||||
# 2026-06-02_2026-06-01_2026-05-17_光影色调_定义_情绪与氛围
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-06-01_2026-05-17_光影色调_定义_情绪与氛围
|
||||
# 2026-05-17_光影色调_定义_情绪与氛围
|
||||
# 光影色调：定义'情绪与氛围
|
||||
# Ch5: 光影色调
|
||||
---
|
||||
## 情绪→光影映射
|
||||
## 现有知识关联
|
||||
- [[20_methods/shaoshi-film-aesthetics-prompts\|邵氏美学 → 高调硬光+依士曼胶片参数]]
|
||||
- [[20_methods/li-dao-ai-video-prompts#6-297用提示词控制视频生成\|李导V6 → 光影控制视听语言]]
|
||||
- [[20_methods/kling/kling-prompt-complete-guide#6-进阶多层描述示例\|可灵手册 → 光影描述示例]]
|
||||
## 采集方向
|
||||
- [ ] 色彩理论在AI视频中的适配方案
|
||||
- [ ] LUT调色在AI视频生成中的应用
|
||||
- [ ] 光影氛围的完整词汇库（中英对照）
|
||||
- [ ] 不同AI工具的光影控制能力差异
|
||||
+51
@@ -0,0 +1,51 @@
|
||||
---
|
||||
id: KB-20260603-039
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-03
|
||||
date_modified: 2026-06-03
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-04---
|
||||
|
||||
# 2026-06-02_2026-06-01_2026-05-17_可灵AI提示词完全知识手册__文生视频_图生视频提示词工
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-06-01_2026-05-17_可灵AI提示词完全知识手册__文生视频_图生视频提示词工程
|
||||
# 2026-05-17_可灵AI提示词完全知识手册__文生视频_图生视频提示词工程
|
||||
# 可灵AI提示词完全知识手册——文生视频/图生视频提示词工程
|
||||
# 可灵AI提示词完全知识手册
|
||||
---
|
||||
## 第一章：可灵AI提示词核心公式
|
||||
### 1.1 通用公式结构
|
||||
### 1.2 5W1H 提示词法
|
||||
### 1.3 六大核心要素（编剧视角）
|
||||
1. 主体：谁在画面中？主角/配角的特征
|
||||
2. 构图与角度：黄金分割点、低角度、仰视视角
|
||||
3. 照明与色彩：柔和的侧光、赛博朋克冷色调
|
||||
4. 运动：摄像机镜头运动 + 画面中物体的运动
|
||||
5. 背景：场景环境、时间、氛围
|
||||
---
|
||||
## 第二章：可灵提示词进阶技巧
|
||||
### 2.1 语言风格原则
|
||||
- 语言简洁：用简单的词语和句子，避免复杂长句
|
||||
- 顺序清晰：主体→场景→运动→镜头语言的顺序
|
||||
- 避免歧义：不省略关键信息
|
||||
- 细节优先：先保证核心内容，再逐步添加细节
|
||||
### 2.2 十大万能镜头运动提示词
|
||||
### 2.3 穿越机效果关键词
|
||||
---
|
||||
## 第三章：可灵模型版本与参数
|
||||
+56
@@ -0,0 +1,56 @@
|
||||
---
|
||||
id: KB-20260603-040
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-03
|
||||
date_modified: 2026-06-03
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-04---
|
||||
|
||||
# 2026-06-02_2026-06-01_2026-05-17_基础定调参数_锁死_底层标准
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-06-01_2026-05-17_基础定调参数_锁死_底层标准
|
||||
# 2026-05-17_基础定调参数_锁死_底层标准
|
||||
# 基础定调参数：锁死'底层标准
|
||||
# Ch3: 基础定调参数
|
||||
---
|
||||
## 一、分辨率选型
|
||||
### 按场景推荐
|
||||
### 显存/算力平衡方案
|
||||
1. **高低配工作流**：低清(768p)生成 → 超分算法放大
|
||||
2. **启用显存优化**：检查工具是否支持"时间切片"或"梯度检查点"
|
||||
3. **利用云算力**：单次高负荷任务租用云渲染
|
||||
---
|
||||
## 二、帧率选型
|
||||
### 帧率速查表
|
||||
### 艺术化搭配
|
||||
**小窍门**：用高帧率生成，后期抽帧降到24fps，比直接生成24fps得到更自然的运动模糊。
|
||||
---
|
||||
## 三、画面比例
|
||||
---
|
||||
## 四、参数协同：性能三角
|
||||
### 场景化参数模板
|
||||
---
|
||||
## 五、质感参数映射（风格→参数）
|
||||
### 新手优先顺序
|
||||
1. 确保基础流畅度（设定帧率）
|
||||
2. 根据平台确定分辨率
|
||||
3. 调整一致性减少闪烁
|
||||
---
|
||||
## 六、运动参数
|
||||
### Motion Bucket（Runway等工具）
|
||||
+53
@@ -0,0 +1,53 @@
|
||||
---
|
||||
id: KB-20260603-041
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-03
|
||||
date_modified: 2026-06-03
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-04---
|
||||
|
||||
# 2026-06-02_2026-06-01_2026-05-17_帅是一种感觉__爆款酷炫短视频拆解与标准化流程
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-06-01_2026-05-17_帅是一种感觉__爆款酷炫短视频拆解与标准化流程
|
||||
# 2026-05-17_帅是一种感觉__爆款酷炫短视频拆解与标准化流程
|
||||
# 帅是一种感觉——爆款酷炫短视频拆解与标准化流程
|
||||
# "帅是一种感觉"——爆款酷炫短视频拆解与标准化流程
|
||||
---
|
||||
## 一、视频逐秒拆解
|
||||
### 核心技术点拆解
|
||||
---
|
||||
## 二、"帅"的本质——技术拆解
|
||||
### 2.1 转场手法（5种混合使用）
|
||||
### 2.2 节奏控制
|
||||
**节奏曲线图**：
|
||||
**关键洞察**：爆款视频的最大秘密不在"做了什么"，而在于**"什么时候停"**。这个视频在**情绪最高点戛然而止**，不给观众喘息时间，让人忍不住重播。
|
||||
### 2.3 画面选择
|
||||
---
|
||||
## 三、标准化流程——任何人都能复刻的"帅"公式
|
||||
### 3.1 三幕式爆款结构
|
||||
### 3.2 素材选择清单
|
||||
### 3.3 剪辑操作步骤
|
||||
### 3.4 检查清单
|
||||
---
|
||||
## 四、关联知识库
|
||||
---
|
||||
## 五、从"酷炫"到"自有风格"的进阶路径
|
||||
### 第一步：复刻（1-3天）
|
||||
- 找一段热门动漫/影视素材
|
||||
- 按照三幕式结构 + 检查清单复刻
|
||||
+46
@@ -0,0 +1,46 @@
|
||||
---
|
||||
id: KB-20260603-042
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-03
|
||||
date_modified: 2026-06-03
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-04---
|
||||
|
||||
# 2026-06-02_2026-06-01_2026-05-17_邵氏电影美学AI提示词技巧__依士曼胶片_高调硬光_影棚
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-06-01_2026-05-17_邵氏电影美学AI提示词技巧__依士曼胶片_高调硬光_影棚置景
|
||||
# 2026-05-17_邵氏电影美学AI提示词技巧__依士曼胶片_高调硬光_影棚置景
|
||||
# 邵氏电影美学AI提示词技巧——依士曼胶片+高调硬光+影棚置景
|
||||
# 邵氏电影美学AI提示词技巧
|
||||
## 视频地址
|
||||
---
|
||||
## 完整字幕
|
||||
- **00:20 — 使用依士曼彩色胶片**：使用依士曼彩色胶片，强制AI还原高饱和度的复古影像特征。
|
||||
- **00:38 — 采用强烈的舞台戏剧、高调硬光**：采用强烈的舞台戏剧、高调硬光，塑造侠客风骨和戏剧光影张力。
|
||||
- **00:58 — 加入高度人工化的影棚置景**：加入高度人工化的影棚置景和材质混搭，呈现古典舞台意境。
|
||||
- **01:20 — 结语**：内容由AI生成。
|
||||
---
|
||||
## 提示词/指令原文
|
||||
---
|
||||
## 操作步骤
|
||||
1. **确定复刻目标**为邵氏电影美学风格
|
||||
2. **加入依士曼彩色胶片**风格关键词
|
||||
3. **加入舞台戏剧感、高调硬光**光影风格描述
|
||||
4. **加入人工化影棚置景、材质混搭**场景描述
|
||||
5. 输入以上风格关键词给到**AI视频生成工具**
|
||||
+56
@@ -0,0 +1,56 @@
|
||||
---
|
||||
id: KB-20260603-043
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-03
|
||||
date_modified: 2026-06-03
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
|
||||
collected: true
|
||||
collected_date: 2026-06-04---
|
||||
|
||||
# 2026-06-02_2026-06-01_2026-05-28_影视风格提示词速查手册
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-06-01_2026-05-28_影视风格提示词速查手册
|
||||
# 2026-05-28_影视风格提示词速查手册
|
||||
# 影视风格提示词速查手册
|
||||
---
|
||||
## 一、华语导演美学
|
||||
### 王家卫电影风格
|
||||
### 张艺谋色彩对比风格
|
||||
### 李安东方写实风
|
||||
---
|
||||
## 二、年代剧与偶像风格
|
||||
### 80年代美剧风
|
||||
### 90年代偶像剧风格
|
||||
### 青春日剧风格
|
||||
### 童年回忆片风格
|
||||
### 日式深夜剧风格
|
||||
---
|
||||
## 三、类型片风格（续）
|
||||
### 诺兰式叙事
|
||||
### 黑泽明剧场感
|
||||
### 宫崎骏动画感
|
||||
---
|
||||
## 三、类型片风格
|
||||
### 西部片风格
|
||||
### 黑色电影风格
|
||||
### 美式青春片风格
|
||||
### 印度歌舞片风格
|
||||
### 反乌托邦电影风格
|
||||
### Netflix惊悚剧风格
|
||||
### 警匪片风格
|
||||
### 灾难电影风格
|
||||
+46
@@ -0,0 +1,46 @@
|
||||
---
|
||||
id: KB-20260606-001
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-06
|
||||
date_modified: 2026-06-27
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
---
|
||||
|
||||
# 2026-06-03_2026-06-02_2026-05-29_OpenCLI安装与集成指南
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-06-02_2026-05-29_OpenCLI安装与集成指南
|
||||
# 2026-05-29_OpenCLI安装与集成指南
|
||||
# OpenCLI — 网站命令行化工具 安装与集成指南
|
||||
---
|
||||
## 背景
|
||||
- **Camoufox**: 反检测浏览器 → 养号/登录态操作
|
||||
- **OpenCLI**: 复用已登录 Chrome → 内容采集/查询
|
||||
---
|
||||
## 一、一键安装脚本
|
||||
#!/bin/bash
|
||||
# install_opencli.sh — 自动安装 OpenCLI + Chrome 扩展
|
||||
-o /tmp/opencli-ext.zip
|
||||
# 写入启动脚本
|
||||
#!/bin/bash
|
||||
--load-extension="$HOME/.opencli/extension" \
|
||||
--remote-debugging-port=0 &
|
||||
---
|
||||
## 二、日常使用
|
||||
### 2.1 Chrome 扩展机制
|
||||
- `opencli daemon start` → 启动守护进程
|
||||
- Chrome 扩展自动连接守护进程
|
||||
- 所有网站操作通过**已登录的 Chrome Cookie** 执行，零额外登录
|
||||
+42
@@ -0,0 +1,42 @@
|
||||
---
|
||||
id: KB-20260606-002
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-06
|
||||
date_modified: 2026-06-06
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
---
|
||||
|
||||
# 2026-06-03_2026-06-02_2026-06-01_2026-05-17_2026-05-16_统一升级引擎
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-06-02_2026-06-01_2026-05-17_2026-05-16_统一升级引擎_agentos_up
|
||||
# 2026-06-01_2026-05-17_2026-05-16_统一升级引擎_agentos_upgrade
|
||||
# 2026-05-17_2026-05-16_统一升级引擎_agentos_upgrade
|
||||
# 2026-05-16_统一升级引擎_agentos_upgrade
|
||||
# 统一升级引擎 agentos upgrade
|
||||
# agentos upgrade — 统一升级引擎
|
||||
## 概述
|
||||
## 命令
|
||||
## 执行流程
|
||||
1. `git pull origin main` — 拉取最新代码
|
||||
2. 扫描 18 个模块（8 个 skill + 8 个 tool + 2 个其他）
|
||||
3. 按模块层级执行不同操作
|
||||
4. 自动注册技能到 WorkBuddy
|
||||
5. 输出升级报告
|
||||
## 三层模块管理
|
||||
## Python 环境统一
|
||||
1. `AGENTOS_PYTHON` 环境变量
|
||||
2. `~/.workbuddy/binaries/python/envs/agent-os/bin/python3`
|
||||
+44
@@ -0,0 +1,44 @@
|
||||
---
|
||||
id: KB-20260606-003
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-06
|
||||
date_modified: 2026-06-06
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
---
|
||||
|
||||
# 2026-06-03_2026-06-02_2026-06-01_2026-05-17_AI提示词驱动文生视频__李导_A
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-06-02_2026-06-01_2026-05-17_AI提示词驱动文生视频__李导_Agent版_10条视频
|
||||
# 2026-06-01_2026-05-17_AI提示词驱动文生视频__李导_Agent版_10条视频知识提取
|
||||
# 2026-05-17_AI提示词驱动文生视频__李导_Agent版_10条视频知识提取
|
||||
# AI提示词驱动文生视频——李导（Agent版）10条视频知识提取
|
||||
# AI提示词驱动文生视频——李导（Agent版）知识合集
|
||||
---
|
||||
## 1. 60提示词做 Ai 短视频第三讲 用三分法构图控制主体精准站位
|
||||
### 完整字幕
|
||||
### 提示词/指令原文
|
||||
### 操作步骤
|
||||
1. 确定画面主体与人物站位需求
|
||||
2. 采用三分法构图，将画面九等分
|
||||
3. 把主体放置在三分法的四个视觉锚点上
|
||||
4. 结合景别控制，调整主体的远近与大小
|
||||
5. 多人物场景时，用三分法规定每个人的位置，用景别控制远近层次
|
||||
### 关键参数
|
||||
### 核心技巧
|
||||
1. 用三分法构图实现 AI 画面主体精准站位，解决自然语言命令不准的问题
|
||||
2. 四个交叉点为视觉锚点，是画面视觉重心舒适区
|
||||
3. 构图 + 景别结合，精准控制主体位置、远近、大小
|
||||
+46
@@ -0,0 +1,46 @@
|
||||
---
|
||||
id: KB-20260606-005
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-06
|
||||
date_modified: 2026-06-06
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
---
|
||||
|
||||
# 2026-06-03_2026-06-02_2026-06-01_2026-05-17_AI视频提示词系统创作教科书___
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-06-02_2026-06-01_2026-05-17_AI视频提示词系统创作教科书___知识体系索引
|
||||
# 2026-06-01_2026-05-17_AI视频提示词系统创作教科书___知识体系索引
|
||||
# 2026-05-17_AI视频提示词系统创作教科书___知识体系索引
|
||||
# AI视频提示词系统创作教科书 · 知识体系索引
|
||||
# AI 视频提示词系统创作教科书 · 知识体系
|
||||
---
|
||||
## 体系总览
|
||||
### 现有知识关联
|
||||
---
|
||||
## 第一部分: 底层逻辑与核心原则（基石篇·结果导向）
|
||||
### Ch1: AI视频生成的"执行本质"与结果推导逻辑
|
||||
**核心问题**: AI 不是读心术，指令的明确性决定输出上限。
|
||||
**结果推导路径**:
|
||||
**采集方向**:
|
||||
- [ ] AI 视频生成工具的底层原理（文本→视频的映射机制）
|
||||
- [ ] 人类创作逻辑 vs AI 执行逻辑的差异分析
|
||||
- [ ] 结果推导方法论的具体案例（多种场景）
|
||||
- [ ] 提示词信息传递模型的图解
|
||||
### Ch2: 结果推导方法论
|
||||
**核心问题**: 如何从目标画面反推提示词锚点。
|
||||
**采集方向**:
|
||||
- [ ] 多场景结果推导案例库
|
||||
+44
@@ -0,0 +1,44 @@
|
||||
---
|
||||
id: KB-20260606-008
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-06
|
||||
date_modified: 2026-06-06
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
---
|
||||
|
||||
# 2026-06-03_2026-06-02_2026-06-01_2026-05-17_ch01-core-logic
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-06-02_2026-06-01_2026-05-17_ch01-core-logic
|
||||
# 2026-06-01_2026-05-17_ch01-core-logic
|
||||
# 2026-05-17_ch01-core-logic
|
||||
# ch01-core-logic
|
||||
# Ch1: AI视频生成的"执行本质"与结果推导逻辑
|
||||
---
|
||||
## 一、底层逻辑：从"脑内画面"到"机器指令"
|
||||
### 1.1 信息传递模型
|
||||
### 1.2 提示词基础公式
|
||||
- **观察视角**：告诉 AI 从哪个角度看
|
||||
- **情感基调**：告诉 AI 什么情绪氛围
|
||||
- **时间线索**：告诉 AI 变化过程
|
||||
### 1.3 AI不是全能助手，是精确执行者
|
||||
### 1.4 为什么AI需要明确指令
|
||||
1. **缺乏常识和联想能力**：输入"一个人走路"，AI可能生成任何场景
|
||||
2. **信息过载阈值**：超过5个核心信息、3组动态元素、10个修饰词 → AI无法聚焦
|
||||
3. **时空错乱**：缺少明确的空间/时间过渡描述 → 画面跳跃
|
||||
4. **文化差异**：AI训练数据的文化背景可能与你不同
|
||||
---
|
||||
## 二、提示词信息传递模型
|
||||
+50
@@ -0,0 +1,50 @@
|
||||
---
|
||||
id: KB-20260606-009
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-06
|
||||
date_modified: 2026-06-06
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
---
|
||||
|
||||
# 2026-06-03_2026-06-02_2026-06-01_2026-05-17_ch02-result-drive
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-06-02_2026-06-01_2026-05-17_ch02-result-driven
|
||||
# 2026-06-01_2026-05-17_ch02-result-driven
|
||||
# 2026-05-17_ch02-result-driven
|
||||
# ch02-result-driven
|
||||
# Ch2: 结果推导方法论——从目标画面反推指令锚点
|
||||
---
|
||||
## 一、三大核心步骤
|
||||
### 步骤1：视觉要素深度拆解
|
||||
**主体要素**
|
||||
- 形态特征：八头身比例、肌肉线条分明
|
||||
- 材质表现：磨砂金属、半透明硅胶
|
||||
- 动态捕捉：跳跃瞬间、手指弯曲45度
|
||||
- 表情管理：眼角上扬15度
|
||||
**环境系统**
|
||||
- 空间层次：前中背景的元素分布
|
||||
- 物理特性：重力、光照、大气折射
|
||||
- 时间维度：季节、昼夜、天气
|
||||
- 文化符号：特定时代的建筑/服饰
|
||||
**光影工程**
|
||||
- 光源位置：钟表定位法（3点钟方向顶光）
|
||||
- 光质控制：硬光(直射) vs 柔光(漫反射)
|
||||
- 色温调节：2000K到10000K
|
||||
- 光比计算：主光与辅光比值（光比4:1）
|
||||
### 步骤2：结构化提示词构建
|
||||
**权重应用**：
|
||||
- 权重越大，AI对该要素越重视
|
||||
+47
@@ -0,0 +1,47 @@
|
||||
---
|
||||
id: KB-20260606-010
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-06
|
||||
date_modified: 2026-06-06
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
---
|
||||
|
||||
# 2026-06-03_2026-06-02_2026-06-01_2026-05-17_ch06-audio-design
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-06-02_2026-06-01_2026-05-17_ch06-audio-design
|
||||
# 2026-06-01_2026-05-17_ch06-audio-design
|
||||
# 2026-05-17_ch06-audio-design
|
||||
# ch06-audio-design
|
||||
# Ch6: 音频设计——让"声音"提升质感
|
||||
---
|
||||
## 核心知识
|
||||
### 音频三角
|
||||
### 背景音乐：情绪匹配
|
||||
### 卡点原则
|
||||
- 快镜头 → 配乐 120BPM+，鼓点清晰，每拍一个转场
|
||||
- 慢镜头 → 配乐 60BPM，旋律舒缓，4拍一个转场
|
||||
- 场景切换 → 配乐节奏变化或乐器变化做标记
|
||||
### 环境音分类
|
||||
### 对白参数
|
||||
- 音色选择：低沉/清脆/沙哑/温暖
|
||||
- 语速：快(紧张)/中(叙述)/慢(深沉)
|
||||
- 音量：对白通常比背景音乐高 10-15dB
|
||||
- 混响：小空间(无混响)/大厅(短混响)/空旷(长混响)
|
||||
### 声音的叙事性
|
||||
- 紧张感构建：环境音渐强 + 配乐节奏加快 + 对白断句变短
|
||||
- 情绪转折：配乐突然停止 → 静默2秒 → 换曲入场
|
||||
- 场景切换：环境音交叉渐变（前一场景音渐弱+后一场景音渐强）
|
||||
+42
@@ -0,0 +1,42 @@
|
||||
---
|
||||
id: KB-20260606-011
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-06
|
||||
date_modified: 2026-06-06
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
---
|
||||
|
||||
# 2026-06-03_2026-06-02_2026-06-01_2026-05-17_ch07-micro-expres
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-06-02_2026-06-01_2026-05-17_ch07-micro-expression
|
||||
# 2026-06-01_2026-05-17_ch07-micro-expression
|
||||
# 2026-05-17_ch07-micro-expression
|
||||
# ch07-micro-expression
|
||||
# Ch7: 人物塑造-微表情与情绪控制
|
||||
---
|
||||
## 核心知识
|
||||
### 情绪→微表情映射
|
||||
### 三招让角色活过来（来源：抖音采集）
|
||||
**1. 控制情绪强度**
|
||||
- 避免 AI 默认情绪最大值（容易假）
|
||||
- 在提示词中加入强度修饰：`slightly`, `barely perceptible`, `subtle`
|
||||
- 示例：`a faint smile barely lifting the corners of his mouth`
|
||||
**2. 给微表情一个发生原因**
|
||||
- 微表情不能孤立存在，需要与场景/对白联动
|
||||
- 加入触发条件的描述：`upon hearing the news, her eyes widened`
|
||||
- 加上半身动作，让画面脱离摆拍感
|
||||
**3. 给情绪加时间顺序**
|
||||
+51
@@ -0,0 +1,51 @@
|
||||
---
|
||||
id: KB-20260606-012
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-06
|
||||
date_modified: 2026-06-06
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
---
|
||||
|
||||
# 2026-06-03_2026-06-02_2026-06-01_2026-05-17_ch08-film-style
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-06-02_2026-06-01_2026-05-17_ch08-film-style
|
||||
# 2026-06-01_2026-05-17_ch08-film-style
|
||||
# 2026-05-17_ch08-film-style
|
||||
# ch08-film-style
|
||||
# Ch8: 电影风格AI视频创作
|
||||
---
|
||||
## 一、电影级提示词核心要素
|
||||
### 六要素结构（来源：AI视频提示词工程2026指南）
|
||||
### 迭代法（从简到繁）
|
||||
---
|
||||
## 二、电影美学风格库
|
||||
### 邵氏电影美学（来源：既有知识）
|
||||
**三大风格要素**：
|
||||
**核心关键词**：
|
||||
### 经典电影风格参考
|
||||
---
|
||||
## 三、电影级参数
|
||||
### 镜头参数
|
||||
### 常用电影质感情号
|
||||
---
|
||||
## 四、分镜与叙事
|
||||
### 好莱坞经典三幕式
|
||||
### 景别→情绪映射（来自李导V4）
|
||||
---
|
||||
## 五、已有知识关联
|
||||
- [[20_methods/shaoshi-film-aesthetics-prompts\|邵氏电影美学提示词技巧]] — 胶片/光影/置景三大风格要素
|
||||
- [[20_methods/li-dao-ai-video-prompts\|李导AI提示词全集]] — 景别/视角/构图/视听语言
|
||||
+47
@@ -0,0 +1,47 @@
|
||||
---
|
||||
id: KB-20260606-013
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-06
|
||||
date_modified: 2026-06-06
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
---
|
||||
|
||||
# 2026-06-03_2026-06-02_2026-06-01_2026-05-17_ch09-anime-style
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-06-02_2026-06-01_2026-05-17_ch09-anime-style
|
||||
# 2026-06-01_2026-05-17_ch09-anime-style
|
||||
# 2026-05-17_ch09-anime-style
|
||||
# ch09-anime-style
|
||||
# Ch9: 动漫/动画风格AI视频创作
|
||||
---
|
||||
## 核心知识
|
||||
### 角色生成万能公式
|
||||
### 10种风格提示词模板
|
||||
### 角色一致性（三视图法）
|
||||
1. **标准全身照**
|
||||
- 公式：[角色描述]，正面全身站立标准照，纯白背景，9:16
|
||||
- 目的：锁定体型、服饰、比例
|
||||
2. **多角度三视图**
|
||||
- 公式：[角色描述]，正面+侧面+背面全身照+面部特写，16:9
|
||||
- 目的：为 AI 提供全方位"建模参考"
|
||||
3. **质感升级**
|
||||
- 加入"次表面散射(SSS)"或"BJD质感" → 告别塑料皮
|
||||
- 加入"轮廓光"+"丁达尔效应"+"强烈明暗对比" → 空间深度
|
||||
### 动漫分镜的提示词适配
|
||||
### AI漫剧完整工作流
|
||||
---
|
||||
## 采集来源
|
||||
+52
@@ -0,0 +1,52 @@
|
||||
---
|
||||
id: KB-20260606-014
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-06
|
||||
date_modified: 2026-06-06
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
---
|
||||
|
||||
# 2026-06-03_2026-06-02_2026-06-01_2026-05-17_ch10-commercial-t
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-06-02_2026-06-01_2026-05-17_ch10-commercial-tvc
|
||||
# 2026-06-01_2026-05-17_ch10-commercial-tvc
|
||||
# 2026-05-17_ch10-commercial-tvc
|
||||
# ch10-commercial-tvc
|
||||
# Ch10: 商业广告/TVC创作
|
||||
---
|
||||
## 一、AI-TVC完整工作流
|
||||
### 5步出品（20分钟内）
|
||||
**工具链**：
|
||||
### 核心：运镜提示词
|
||||
---
|
||||
## 二、TVC提示词结构
|
||||
### 标准格式
|
||||
### 示例：像素风TVC
|
||||
### 玉石品牌TVC（来自既有知识）
|
||||
- 结合中式色彩美学
|
||||
- 电影感构图
|
||||
- 专业LUT调色提升品质
|
||||
- Seedance 2.0 生成电影级画面
|
||||
---
|
||||
## 三、商业广告类型与策略
|
||||
### 高级技巧
|
||||
1. **图片先行**：先用高质量生图工具打好画面基础
|
||||
2. **动作提示词含运镜**：让静态图"活"起来的关键
|
||||
3. **后处理统一调性**：剪映统一调色+LUT
|
||||
4. **多次迭代**：每次只改5%内容AB测试
|
||||
---
|
||||
## 采集来源
|
||||
+52
@@ -0,0 +1,52 @@
|
||||
---
|
||||
id: KB-20260606-015
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-06
|
||||
date_modified: 2026-06-06
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
---
|
||||
|
||||
# 2026-06-03_2026-06-02_2026-06-01_2026-05-17_ch11-sci-fi-fanta
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-06-02_2026-06-01_2026-05-17_ch11-sci-fi-fantasy
|
||||
# 2026-06-01_2026-05-17_ch11-sci-fi-fantasy
|
||||
# 2026-05-17_ch11-sci-fi-fantasy
|
||||
# ch11-sci-fi-fantasy
|
||||
# Ch11: 科幻/奇幻风格创作
|
||||
---
|
||||
## 一、科幻风格提示词大全
|
||||
### 提示词结构公式
|
||||
### 赛博朋克城市
|
||||
**完整模板**：
|
||||
**关键元素**：
|
||||
### 太空场景
|
||||
**空间站**：
|
||||
**外星景观**：
|
||||
### 科技与机械
|
||||
**人形机器人**：
|
||||
---
|
||||
## 二、风格参数
|
||||
### 光影参数
|
||||
### 美术风格
|
||||
### 风格参考锚点
|
||||
---
|
||||
## 三、奇幻风格创作
|
||||
### 通用奇幻模板
|
||||
### 经典奇幻类型
|
||||
### 常见错误与修正
|
||||
---
|
||||
## 采集来源
|
||||
+42
@@ -0,0 +1,42 @@
|
||||
---
|
||||
id: KB-20260606-016
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-06
|
||||
date_modified: 2026-06-06
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
---
|
||||
|
||||
# 2026-06-03_2026-06-02_2026-06-01_2026-05-17_ch12-model-compar
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-06-02_2026-06-01_2026-05-17_ch12-model-comparison
|
||||
# 2026-06-01_2026-05-17_ch12-model-comparison
|
||||
# 2026-05-17_ch12-model-comparison
|
||||
# ch12-model-comparison
|
||||
# Ch12: 主流AI视频大模型的差异与适配
|
||||
---
|
||||
## 核心数据：平台对比
|
||||
## 模型版本演进
|
||||
## 结果导向选型
|
||||
## 工具链协同
|
||||
- 商业广告：Gen-2(画面) + Au(声音)
|
||||
- 动漫番剧：Pika(多镜头) + 配音平台(对白)
|
||||
- 电影短片：Kling 3.0(素材) + Runway(后期)
|
||||
---
|
||||
## 已采集素材
|
||||
### 抖音大模型对比素材
|
||||
---
|
||||
## 采集来源
|
||||
+46
@@ -0,0 +1,46 @@
|
||||
---
|
||||
id: KB-20260606-017
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-06
|
||||
date_modified: 2026-06-06
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
---
|
||||
|
||||
# 2026-06-03_2026-06-02_2026-06-01_2026-05-17_ch13-tool-workflo
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-06-02_2026-06-01_2026-05-17_ch13-tool-workflow
|
||||
# 2026-06-01_2026-05-17_ch13-tool-workflow
|
||||
# 2026-05-17_ch13-tool-workflow
|
||||
# ch13-tool-workflow
|
||||
# Ch13: 工具实操与工作流搭建
|
||||
---
|
||||
## 一、四大核心工具详解
|
||||
### Runway Gen-3 Alpha
|
||||
**定位**：电影级画质、精细控制，专业制作首选
|
||||
**操作流程**：
|
||||
1. 准备高质量静态图（建议1080p以上，最好4K）
|
||||
2. 上传图片至平台（支持JPG/PNG）
|
||||
3. 编写提示词描述期望的动态效果
|
||||
4. 调整 Camera Motion 参数和 Motion Bucket
|
||||
5. 生成预览（耗时2-5分钟）
|
||||
**提示词结构**：
|
||||
**参数**：
|
||||
- Motion Bucket：0-10（控制运动强度）
|
||||
- Camera Motion：Dolly In/Out、Tracking Shot、Orbit、Bird's Eye
|
||||
- 时长上限：16秒
|
||||
- 分辨率：最高4K
|
||||
---
|
||||
+53
@@ -0,0 +1,53 @@
|
||||
---
|
||||
id: KB-20260606-018
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-06
|
||||
date_modified: 2026-06-06
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
---
|
||||
|
||||
# 2026-06-03_2026-06-02_2026-06-01_2026-05-17_ch14-script-to-pr
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-06-02_2026-06-01_2026-05-17_ch14-script-to-prompt
|
||||
# 2026-06-01_2026-05-17_ch14-script-to-prompt
|
||||
# 2026-05-17_ch14-script-to-prompt
|
||||
# ch14-script-to-prompt
|
||||
# Ch14: 从"创意"到"AI友好型脚本"的转化
|
||||
---
|
||||
## 一、四步工作流：小说→分镜脚本
|
||||
### 核心原理
|
||||
### 四步流程
|
||||
**第一步：规范化整理内容**
|
||||
**第二步：预估镜头数与结构分布**
|
||||
- 需要多少镜头？
|
||||
- 镜头在关键情节中的分布？
|
||||
- 建立视觉节奏感
|
||||
**第三步：搭建核心场景与角色表**
|
||||
- 场景1：咸亨酒店大堂，午后，光线昏暗，柜台有酒坛
|
||||
- 角色A：孔乙己，瘦弱，长衫破旧
|
||||
**第四步：整合生成分镜脚本**
|
||||
**效果对比**（以《孔乙己》为例）：
|
||||
- 直接转换：仅12个镜头
|
||||
- 四步工作流：59个分镜头 ✅ 完整还原剧情
|
||||
### 提示词模板
|
||||
---
|
||||
## 二、场景控制技术
|
||||
### 先建库，再使用
|
||||
### 场景描述要素
|
||||
### 场景互动
|
||||
---
|
||||
## 三、AI友好型脚本格式
|
||||
+47
@@ -0,0 +1,47 @@
|
||||
---
|
||||
id: KB-20260606-019
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-06
|
||||
date_modified: 2026-06-06
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
---
|
||||
|
||||
# 2026-06-03_2026-06-02_2026-06-01_2026-05-17_ch15-narrative-st
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-06-02_2026-06-01_2026-05-17_ch15-narrative-structure
|
||||
# 2026-06-01_2026-05-17_ch15-narrative-structure
|
||||
# 2026-05-17_ch15-narrative-structure
|
||||
# ch15-narrative-structure
|
||||
# Ch15: 叙事结构与AI叙事能力的结合
|
||||
---
|
||||
## 常见叙事结构
|
||||
## AI叙事优势与局限
|
||||
### AI擅长
|
||||
- 视觉化情绪（光影+微表情+场景）
|
||||
- 高密度画面叙事（每一秒承载大量信息）
|
||||
- 风格化渲染（电影感/动漫感）
|
||||
### AI不擅长
|
||||
- 复杂对白逻辑（容易混淆人物关系）
|
||||
- 细微情感递进（缺少层次感）
|
||||
- 长时间连贯性（超过30秒容易崩）
|
||||
### 策略：扬长避短
|
||||
## 短剧分镜7大要素
|
||||
1. 景别（远全中近特）
|
||||
2. 视角（平俯仰倾）
|
||||
3. 构图（三分法/对称/不对称）
|
||||
4. 光影（情绪映射）
|
||||
5. 运动（镜头+主体）
|
||||
+47
@@ -0,0 +1,47 @@
|
||||
---
|
||||
id: KB-20260606-020
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-06
|
||||
date_modified: 2026-06-06
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
---
|
||||
|
||||
# 2026-06-03_2026-06-02_2026-06-01_2026-05-17_ch16-practice-pat
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-06-02_2026-06-01_2026-05-17_ch16-practice-path
|
||||
# 2026-06-01_2026-05-17_ch16-practice-path
|
||||
# 2026-05-17_ch16-practice-path
|
||||
# ch16-practice-path
|
||||
# Ch16: 从"模仿"到"创新"的实践路径
|
||||
---
|
||||
## 三个阶段
|
||||
### 阶段1：爆款拆解（入门）
|
||||
- 分析抖音/B站AI视频爆款的"结果→提示词→工具"链路
|
||||
- 工具：video-to-prompt.com（自动反推提示词）
|
||||
- 目标：能复现同级别效果
|
||||
### 阶段2：垂直深耕（进阶）
|
||||
- 聚焦一个领域（如AI漫剧/AI电影感/AI广告）
|
||||
- 打磨"成果稳定"的提示词模板
|
||||
- 建立自己的风格关键词库
|
||||
- 目标：每次生成都能稳定输出
|
||||
### 阶段3：风格创新（高级）
|
||||
- 融合多领域元素（如赛博朋克+古风）
|
||||
- 打造"差异化成果"的AI视频风格
|
||||
- 建立个人方法论体系
|
||||
- 目标：一眼能认出是你的风格
|
||||
## 成长路径
|
||||
## 效率指标
|
||||
+37
@@ -0,0 +1,37 @@
|
||||
---
|
||||
id: KB-20260606-021
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-06
|
||||
date_modified: 2026-06-06
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
---
|
||||
|
||||
# 2026-06-03_2026-06-02_2026-06-01_2026-05-17_ch17-resource-lib
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-06-02_2026-06-01_2026-05-17_ch17-resource-library
|
||||
# 2026-06-01_2026-05-17_ch17-resource-library
|
||||
# 2026-05-17_ch17-resource-library
|
||||
# ch17-resource-library
|
||||
# Ch17: 资源库建设——按成果类型分类的知识弹药库
|
||||
---
|
||||
## 平台资源
|
||||
## 工具资源
|
||||
## 提示词资源
|
||||
- PromptMart（50000+提示词库）
|
||||
- video-to-prompt.com（视频反推提示词）
|
||||
- 本知识库各章节（不断积累）
|
||||
## 案例资源
|
||||
+48
@@ -0,0 +1,48 @@
|
||||
---
|
||||
id: KB-20260606-022
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-06
|
||||
date_modified: 2026-06-06
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
---
|
||||
|
||||
# 2026-06-03_2026-06-02_2026-06-01_2026-05-17_ch18-problem-solv
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-06-02_2026-06-01_2026-05-17_ch18-problem-solving
|
||||
# 2026-06-01_2026-05-17_ch18-problem-solving
|
||||
# 2026-05-17_ch18-problem-solving
|
||||
# ch18-problem-solving
|
||||
# Ch18: 常见问题解决方案库
|
||||
---
|
||||
## 核心问题分类
|
||||
### A. 画面类问题
|
||||
#### A1. 人脸变形/身份漂移
|
||||
**提示词修复**：
|
||||
#### A2. 物体漂移/元素消失
|
||||
**提示词修复**：
|
||||
#### A3. 光影闪烁/曝光变化
|
||||
**后期补救**：调色工具统一曝光。
|
||||
#### A4. 画面风格跑偏
|
||||
- **检查参考示例**是否明确
|
||||
- **换适配大模型**（如写实→Kling，动漫→Pika）
|
||||
- 提示词加入风格锚点：`cinematic, photorealistic, anime style`
|
||||
---
|
||||
### B. 运动类问题
|
||||
#### B1. 不自然动作/飘忽运动
|
||||
#### B2. 手部与手指问题
|
||||
- **裁剪手部**或让其失焦
|
||||
- 使用简单手部姿势（握拳、张开手掌）
|
||||
+42
@@ -0,0 +1,42 @@
|
||||
---
|
||||
id: KB-20260606-023
|
||||
title: ""
|
||||
type: method
|
||||
status: published
|
||||
nature: method
|
||||
domain: [general]
|
||||
subdomain: []
|
||||
tags: ["待补充"]
|
||||
confidence: 0.5
|
||||
source: ""
|
||||
source_type: unknown
|
||||
date_created: 2026-06-06
|
||||
date_modified: 2026-06-06
|
||||
version: 1
|
||||
previous_version: ""
|
||||
superseded_by: ""
|
||||
summary: ""
|
||||
---
|
||||
|
||||
# 2026-06-03_2026-06-02_2026-06-01_2026-05-17_ch19-trends
|
||||
|
||||
> 来源：03_knowledge/20_methods
|
||||
|
||||
# 2026-06-02_2026-06-01_2026-05-17_ch19-trends
|
||||
# 2026-06-01_2026-05-17_ch19-trends
|
||||
# 2026-05-17_ch19-trends
|
||||
# ch19-trends
|
||||
# Ch19: 行业趋势与前沿探索
|
||||
---
|
||||
## AI视频技术演进方向
|
||||
## 提示词的"未来形态"
|
||||
## 需要关注的前沿方向
|
||||
1. **Sora级别模型**：物理世界理解能力大幅提升
|
||||
2. **实时AI视频编辑**：生成后可逐帧修改
|
||||
3. **个性化模型微调**：用自己的数据训练专属模型
|
||||
4. **AI视频+AR/VR**：沉浸式内容创作
|
||||
## 适配策略
|
||||
- 保持关注各平台更新日志
|
||||
- 定期测试新模型能力
|
||||
- 积累跨模态提示词经验
|
||||
- 建立自己的训练数据集
|
||||
Some files were not shown because too many files have changed in this diff Show More
Reference in New Issue
Block a user