yyl-video-prompt
Compare original and translation side by side
🇺🇸
Original
English🇨🇳
Translation
Chineseyyl-video-prompt：导演式 AI 视频提示词
yyl-video-prompt: Director-Style AI Video Prompts
定位
Positioning
把任何内容变成能直接提交生成的视频提示词。先按脚本风格选档位，两个档位共用生产层，画面层和表演层各有各的写法：
| 层 | 解决什么 |
|---|---|
| 生产层（共用） | 脚本拆成生成单元；每段选生成模式、参数、参考图；段与段接得上 |
| 画面层 | 人物、道具、场景空间、镜头、风格光影、声音、硬约束 |
| 表演层 | 短视频：一句可见动作 + 一个表情节点。电影精品：八维公式、AU、情绪递进、反应链 |
原则：日常动作轻量写，关键时刻细写，AI 容易出错的地方重点写。
Turn any content into video prompts ready for direct submission to a generator. First select a tier based on the script style. The two tiers share the production layer, while the visual and acting layers each have their own writing methods:
| Layer | What It Solves |
|---|---|
| Production layer (shared) | Break the script into generation units; select the generation mode, parameters, and reference images for each segment; ensure continuity between segments |
| Visual layer | Characters, props, spatial environment, camera, visual style and lighting, sound, hard constraints |
| Acting layer | Short video: one visible action + one expression beat. Cinematic premium: eight-dimension formula, AU, emotional progression, reaction chain |
Principle: Keep routine actions lightweight, key moments detailed, and focus on the areas where AI is prone to errors.
第 0 步：选档位
Step 0: Select a Tier
| 短视频档 | 电影精品档 | |
|---|---|---|
| 平台 / 用途 | 抖音、小红书、视频号、快手、Reels；剧情片段、街拍、氛围短片、角色小故事、竖屏短剧 | 剧情短片、精品剧、文艺片段落 |
| 节奏 | 一镜到底每 2 秒一个事件节点，或多分镜每镜 2–5 秒 | 4–12 秒一镜，固定机位或极慢推近 |
| 靠什么推进 | 动作事件、运镜、空间和声音、视觉反差 | 表演：情绪递进、对白、对手戏 |
| 表演写法 | 一句可见动作 + 表情变化，人话，不用 AU | 八维公式 + AU 辅助 + 反应链 |
| 结构 | 写法 A 一镜到底（《丹眉》式）/ 写法 B 多分镜（赛博原子式） | 全文地图 / 人物一致性 / 分镜时间轴 / 核心要求…… |
| 来源 | AI-video-prompt《丹眉》+ mx-shell-prompt 赛博原子 | yyl-video-prompt v1 |
| 先读 | modes/short-video.md → examples/short-video.md | modes/cinematic-drama.md → examples/cinematic-drama.md |
判断规则：
- 用户说了平台或风格 → 照用户说的。抖音、小红书、快节奏、短视频走短视频档；精品剧、电影感剧情、慢节奏、表演走电影精品。
- 没说时看脚本：有多句对白，加上情绪转折、对手戏，而且是同一场景 → 电影精品；其余 → 短视频。
- 竖屏短剧 → 短视频档，借电影档的对白嵌入和反应链（见短视频档文末）。
- 说不清 → 短视频档，并在第 3 步的确认消息里说明「可以切到电影精品档」。
- 合成 / 编辑类任务（把 A 放进 B、只改背景）两档都不用，按 tool-adapters.md 末节的极简写法。
一个项目里不同段可以用不同档位，例如开场钩子走短视频，正戏走电影精品。
选定档位后，先读该档的 文件和对应示例，再动笔。
modes/| Short-Video Tier | Cinematic Premium Tier | |
|---|---|---|
| Platform / Use | Douyin, Xiaohongshu, WeChat Channels, Kuaishou, Reels; narrative segments, street footage, atmospheric short films, character micro-stories, vertical-format short dramas | Narrative shorts, premium series, literary and arthouse segments |
| Pacing | One continuous shot with an event beat every 2 seconds, or multiple shots lasting 2–5 seconds each | One shot every 4–12 seconds, using a fixed camera or an extremely slow push-in |
| Primary Driver | Action events, camera movement, space and sound, visual contrast | Acting: emotional progression, dialogue, two-hander scenes |
| Acting Approach | One visible action + an expression change, natural spoken language, no AU | Eight-dimension formula + AU support + reaction chain |
| Structure | Method A: one continuous shot (Danmei style) / Method B: multiple shots (Cyber Atom style) | Full narrative map / character consistency / shot timeline / core requirements… |
| Source | AI-video-prompt Danmei + mx-shell-prompt Cyber Atom | yyl-video-prompt v1 |
| Read First | modes/short-video.md → examples/short-video.md | modes/cinematic-drama.md → examples/cinematic-drama.md |
Decision Rules:
- If the user specifies a platform or style, follow the user. Douyin, Xiaohongshu, fast pacing, and short videos use the short-video tier; premium series, cinematic narrative, slow pacing, and performance use the cinematic premium tier.
- If the user gives no guidance, inspect the script: multiple dialogue lines, emotional turns, and two-hander scenes within the same setting indicate the cinematic premium tier; everything else uses the short-video tier.
- Vertical-format short dramas → use the short-video tier while incorporating cinematic-tier dialogue embedding and reaction chains, as described at the end of the short-video guide.
- If the tier cannot be determined, use the short-video tier and state in the Step 3 confirmation message that it can be switched to the cinematic premium tier.
- Composite or editing tasks (placing A into B, changing only the background) do not use either tier. Follow the minimal format in the final section of tool-adapters.md.
Different segments within one project may use different tiers. For example, the opening hook can use the short-video tier while the main drama uses the cinematic premium tier.
After selecting a tier, read that tier's file and corresponding examples before drafting.
modes/工作流
Workflow
第 1 步：读输入，定三件事
Step 1: Read the Input and Determine Three Things
- 目标工具和模型：决定单段时长上限和素材引用写法，见 tool-adapters.md。用户没说就按当前项目默认；项目也没有默认，就用 。
Seedance 2.5 | 720p | 16:9 - 影像基调：电影摄影 / 手机实拍 / 动画。手机实拍不写「电影感、IMAX、大片、4K」。
- 参考素材：有素材就给每张定一个任务；没有就写清该准备什么图，或者改成纯文生。
一句话想法：视为单段，直接补全，跳过第 3 步的确认。
- Target tool and model: This determines the maximum duration of each segment and the syntax for referencing assets, as described in tool-adapters.md. If the user does not specify one, use the current project default. If the project has no default either, use .
Seedance 2.5 | 720p | 16:9 - Visual tone: Cinematic photography / smartphone live action / animation. Do not write "cinematic," "IMAX," "blockbuster," or "4K" for smartphone live action.
- Reference assets: If assets are available, assign each image a single task. If there are none, clearly state what images should be prepared or switch to text-only generation.
A one-sentence idea is treated as a single segment. Complete the missing details and skip the Step 3 confirmation.
第 2 步：拆段
Step 2: Divide the Content into Segments
按地点、时间、情绪大转折、视角切换、单段时长上限来切。段尾写清结束状态，下一段从这里接。细则见 REFERENCE.md「二、分段」，各档的镜头数见 。
modes/Split by location, time, major emotional turns, viewpoint changes, and the maximum duration of each segment. Clearly state the ending state of each segment so that the next segment can continue from it. See section "II. Segmentation" in REFERENCE.md for details and the files for the number of shots in each tier.
modes/第 3 步：导演判断（唯一的确认门）
Step 3: Director's Judgment — The Only Confirmation Gate
逐段判断：哪里轻写，哪里细写，AI 会在哪里出错。多段或方向有歧义时，用一条消息交给用户：
text
档位：短视频（一镜到底）——理由一句话；想要慢节奏表演可以切到电影精品档。
我拆成 N 段：
段1（0–15s）：[一句话节拍]｜[生成模式]｜[需要的参考图]
段2（……）：……
我的判断：
- 段1 [某个时刻]：我打算 [处理]，观众会感觉 [效果]；另一个方向是 [备选]。
确认就说「直接写」；要合并、拆开或改方向直接说。如果只有一段、意图清楚，就跳过这一步直接写。用户答不上来的地方，给两三个选项并说清效果，不要让用户填表。
Evaluate each segment: where should the writing remain lightweight, where should it be detailed, and where is AI most likely to make mistakes? For multi-segment content or ambiguous direction, present everything to the user in a single message:
text
Tier: Short video (one continuous shot)——[one-sentence reason]; it can be switched to the cinematic premium tier for slow-paced acting.
I will divide it into N segments:
Segment 1 (0–15s): [one-sentence beat] | [generation mode] | [required reference images]
Segment 2 (…): …
My judgment:
- Segment 1 [specific moment]: I plan to [treatment], so viewers will feel [effect]; the alternative is [alternative].
To confirm, say "write it directly"; to merge, split, or change direction, just say so.If there is only one segment and the intent is clear, skip this step and write directly. If the user cannot answer, provide two or three options and explain the effect of each. Do not make the user fill out a form.
第 4 步：逐段写
Step 4: Write Each Segment
用该档的结构写（见 ），照着该档示例的密度和语气来。示例是标杆，只拿来学，不要按通用规则去「改进」它们。
modes/Use the selected tier's structure and follow the density and tone of its examples. The examples are benchmarks: use them as a reference, but do not "improve" them according to generic rules.
第 5 步：自检后交付
Step 5: Self-Check and Deliver
过一遍下面的共用自检和该档的专用自检，可以跑 。交付后问一句：要调整分段、换档位，或者加重/减弱某处吗？
python3 scripts/prompt_builder.py --validate <文件>Review the shared checklist below and the tier-specific checklist. You can run . After delivery, ask whether the user wants to adjust the segmentation, switch tiers, or strengthen or weaken any section.
python3 scripts/prompt_builder.py --validate <file>每段的外壳（两档通用）
Shell for Each Segment — Shared by Both Tiers
markdown
undefinedmarkdown
undefined段1：[一句话节拍]（0–15 秒）
Segment 1: [one-sentence beat] (0–15 seconds)
档位：短视频（一镜到底 / 多分镜） 或 电影精品
生成模式：[纯文生 / 首帧图生 / 首尾帧 / 多图参考 / 全能参考 / 视频编辑]
模型参数：[模型] | [分辨率] | [画幅] | [时长]
参考素材：
图片1：[唯一任务]——锁定 [脸 / 发型 / 服装 / 比例……]；不使用 [背景 / 其中的人物……]
（纯文生写「无」；缺图时写清该准备什么图）
提示词：
text
[按该档的结构写]段尾状态：[姿势、位置、构图——下一段从这里接]
提示词块里的素材引用要换成目标工具的写法（`@图片1` / `参考图1` / `{{Node "节点名"}}` / `@图像1`）。Tier: Short video (one continuous shot / multi-shot) or Cinematic premium
Generation mode: [text-to-video / first-frame image-to-video / first-and-last-frame / multi-image reference / universal reference / video editing]
Model parameters: [Model] | [Resolution] | [Aspect Ratio] | [Duration]
Reference assets:
Image 1: [single task]——locks [face / hairstyle / clothing / proportions…]; excludes [background / people within it…]
(For text-to-video, write "None"; if an image is missing, specify what image should be prepared)
Prompt:
text
[Write using the selected tier's structure]End-of-segment state: [pose, position, composition——the next segment continues from here]
Replace asset references inside the prompt block with the target tool's syntax (`@图片1` / `参考图1` / `{{Node "node name"}}` / `@图像1`).核心规则（两档通用）
Core Rules — Shared by Both Tiers
- 每段自包含。生成工具没有跨段记忆。人物锁定、场景、光线、手里的道具、上一段的结束状态，每段都要写全，不写「同上」「沿用上一段」。长相和衣服交给参考图锁定，文字里不重复描述。
- 一张参考图只承担一个任务。只写「参考图片」等于没写，必须写清锁定什么、不用什么。引用写法跟着工具走：LibTV 必须写 ，纯文本的「图片1」无效。
{{Node "节点名"}} - 时间码是动作预算。时间窗连续、不重叠，一个窗口只放一个核心动作。关键事件写到具体秒。最后一拍留 0.5 秒余量，因为模型的最后一拍容易偏晚。
- 首帧是第一个动作发生前的状态。要第 2 秒架腿，首帧就必须双腿并拢。
- 人物不能悬空。每个受力部位都写清支撑：背靠什么、坐在哪、脚踩在哪。每个动作写起因和收尾。
- 情绪写成可见动作，不写孤立的情绪标签（例如「她很悲伤」）。短视频写一个表情节点；电影精品写完整的递进过程。
- 对白嵌在时间码里，不要把台词贴在最后。
- 只写最重要的东西。描述过度会让模型把本该在画面内部的元素提升到整个画面上。
- 基调词要和基调匹配。手机实拍不写「电影感 / 高清 / 大片 / 4K」。电影摄影要写具体的摄影机、镜头、胶片。风格参考最多 1–2 个。
- 默认无背景音乐，只保留同期声。用户要配乐才写。
- Every segment must be self-contained. Generation tools do not have cross-segment memory. Character locks, setting, lighting, props in hand, and the previous segment's ending state must be fully restated in every segment. Do not write "same as above" or "continue from the previous segment." Lock appearance and clothing through reference images rather than repeating them in the text.
- Each reference image must serve only one task. Writing only "reference image" is equivalent to writing nothing. Always specify what is locked and what is excluded. Follow the target tool's reference syntax: LibTV requires ; plain-text references such as "Image 1" are invalid.
{{Node "node name"}} - Timecodes are action budgets. Time windows must be continuous and non-overlapping, with only one core action per window. Write key events down to the exact second. Leave a 0.5-second buffer after the final beat because models often drift late at the end.
- The first frame shows the state before the first action begins. If the character raises a leg at second 2, the first frame must show both legs together.
- Characters must remain physically supported. Specify the support for every weight-bearing body part: what they lean against, where they sit, or where their feet are planted. Give every action a cause and an ending state.
- Express emotion through visible actions, not isolated emotion labels such as "she is very sad." Use one expression beat for short video and a complete progression for cinematic premium content.
- Embed dialogue in the timecodes. Do not append the lines at the end.
- Write only the most important information. Excessive description may cause the model to promote elements that should remain within the scene to the entire frame.
- Match style terms to the visual tone. Do not write "cinematic," "high definition," "blockbuster," or "4K" for smartphone live action. For cinematic photography, specify the camera, lens, and film stock. Use no more than one or two style references.
- No background music by default. Retain only production sound. Add music only when the user requests it.
共用自检
Shared Self-Check
- 档位选对了，并且读过该档的 文件和示例
modes/ - 每段时长不超过工具上限
- 每段自包含，没有「同上 / 沿用上一段」
- 每张参考图只有一个任务，锁定项和排除项都写了；正文引用的编号都在参考列表里；引用写法匹配目标工具
- 动作有支撑点和前因后果；首帧是动作发生前的状态
- 声音有环境声底和事件音；没有 BGM（除非用户要求）
- 硬约束用正面句写死主体占比、空间逻辑、动作顺序；否定项只针对已知失败
- 段尾写了结束状态
- 过了该档的专用自检
- The correct tier was selected, and its file and examples were read
modes/ - No segment exceeds the target tool's duration limit
- Every segment is self-contained, with no "same as above" or "continue from the previous segment"
- Every reference image serves only one task and includes both locked and excluded elements; every numbered reference in the prompt appears in the reference list; the reference syntax matches the target tool
- Actions have physical support and clear cause and effect; the first frame shows the state before the action begins
- Sound includes an ambient bed and event sounds, with no BGM unless requested
- Hard constraints use positive statements to lock subject occupancy, spatial logic, and action order; negative constraints target only known failure modes
- The ending state is written for every segment
- The tier-specific self-check has been completed
文件地图
File Map
| 文件 | 内容 |
|---|---|
| modes/short-video.md | 短视频档：写法 A 一镜到底、写法 B 多分镜、两种写法互相吸取的部分、表演、竖屏短剧 |
| modes/cinematic-drama.md | 电影精品档：结构、全文地图、人物一致性、表演层（八维、AU、情绪递进、反应链） |
| examples/short-video.md | 《丹眉》一镜到底两段（AI-video-prompt 原文，爆款）、赛博原子多分镜（mx-shell-prompt 原文 + 四处补充） |
| examples/cinematic-drama.md | 《改名》30 秒三人对手戏、5 个表演片段 |
| REFERENCE.md | 共用：分段、参考绑定、道具、场景空间、风格、时间码与动作、声音、硬约束、实测教训 |
| references/tool-adapters.md | 各工具的素材引用写法、生成模式、单段上限、合成类极简写法 |
| references/shot-catalog.md | 景别、构图、运镜，以及「情绪 → 镜头语言」速查 |
| references/cinematic-vocabulary.md | 摄影机、镜头、胶片、色彩、灯光、风格参考，以及手机实拍的用词 |
| |
| File | Contents |
|---|---|
| modes/short-video.md | Short-video tier: Method A, one continuous shot; Method B, multiple shots; elements shared between the two methods; acting; vertical-format short dramas |
| modes/cinematic-drama.md | Cinematic premium tier: structure, full narrative map, character consistency, acting layer (eight dimensions, AU, emotional progression, reaction chain) |
| examples/short-video.md | Two-segment one-continuous-shot example from Danmei (original AI-video-prompt, viral hit) and a multi-shot Cyber Atom example (original mx-shell-prompt + four additions) |
| examples/cinematic-drama.md | Name Change: a 30-second, three-character confrontation scene and five acting segments |
| REFERENCE.md | Shared guidance: segmentation, reference bindings, props, spatial environment, style, timecodes and actions, sound, hard constraints, and lessons from real-world testing |
| references/tool-adapters.md | Asset-reference syntax, generation modes, per-segment duration limits, and the minimal format for composite tasks for each tool |
| references/shot-catalog.md | Shot sizes, composition, camera movement, and a quick reference for translating emotion into camera language |
| references/cinematic-vocabulary.md | Cameras, lenses, film stock, color, lighting, style references, and terminology for smartphone live action |
| |