Back to Details

generation-prompt-builder

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

Generation Prompt Builder

生成提示词构建器

Purpose

用途

Use this skill to turn user material into clear, production-ready prompts for AI generation tools. The output is prompt text only. Do not call image, video, audio, or CLI tools from this skill.
使用此 skill 将用户素材转化为清晰、可直接投入生产的 AI 生成工具提示词。输出仅限提示词文本。不要从此 skill 调用图像、视频、音频或 CLI 工具。

When To Use

何时使用

Use when the user asks to:
  • generate prompts from ad copy, product information, scripts, briefs, rough ideas, or reference-image descriptions
  • optimize, polish, professionalize, translate, or restructure an existing prompt
  • create prompt variants for image, video, audio, product visuals, posters, UI mockups, character art, or creative assets
  • preserve an existing concept while making it more specific and controllable
Boundary: If the user wants GPT Image execution, image editing, storyboard planning, shot lists, multi-shot video breakdowns, or pre-writing advice only, route to the corresponding dedicated skill instead.
当用户要求：
  • 根据广告文案、产品信息、脚本、需求简报、粗略想法或参考图描述生成提示词
  • 优化、打磨、专业化、翻译或重组现有提示词
  • 为图像、视频、音频、产品视觉、海报、UI 模型、角色艺术或创意素材创建提示词变体
  • 保留现有概念，同时使其更具体、更可控
边界：如果用户希望执行 GPT Image、编辑图像、规划 storyboard、编写镜头清单、拆分多镜头视频，或仅需生成前的写作建议，应交由对应的专用 skill 处理。

Core Rules

核心规则

  1. Preserve the user's intent: keep the same subject, product, story beats, emotional tone, brand constraints, and ending unless the user asks for variants.
  2. Do not invent unsupported product facts, specs, prices, brand claims, slogans, certifications, or safety claims.
  3. If a key detail is missing but not blocking, make the smallest reasonable assumption and label it in one short line.
  4. If a missing detail would change the result significantly, ask 1-3 short questions before finalizing.
  5. Keep prompts directly usable. Avoid long lectures, vague style filler, and unnecessary model jargon.
  6. Match the user's requested output language. If no language is specified, use the conversation language for explanation and English for prompt blocks when that improves model compatibility.
  1. 保留用户意图：除非用户要求生成变体，否则应保持相同的主体、产品、故事节拍、情绪基调、品牌约束和结尾不变。
  2. 不要编造缺乏依据的产品事实、规格、价格、品牌声明、口号、认证或安全声明。
  3. 如果缺少关键细节但并不妨碍继续，可以做出最小且合理的假设，并用一行简短说明进行标注。
  4. 如果缺失的细节会显著改变结果，请在定稿前提出 1-3 个简短问题。
  5. 确保提示词可直接使用。避免冗长说教、模糊的风格填充词和不必要的模型术语。
  6. 匹配用户要求的输出语言。如果未指定语言，说明部分使用对话语言；当有助于提升模型兼容性时，提示词块使用英语。

Workflow

工作流

  1. Identify modality: image, video, audio, mixed, or unspecified.
  2. Extract anchors: subject, goal, audience, style, mood, scene, composition, motion, text, constraints, and reference material.
  3. Decide whether the request is generation from scratch, optimization, translation, or variants.
  4. Write the prompt in a structured order the target model can follow.
  5. Add negative constraints only when useful for control.
  6. Briefly explain meaningful changes.
  1. 确定模态：图像、视频、音频、混合或未指定。
  2. 提取关键锚点：主体、目标、受众、风格、情绪、场景、构图、运动、文本、约束和参考素材。
  3. 判断任务类型：从零生成、优化、翻译或创建变体。
  4. 按照目标模型易于理解的顺序编写提示词。
  5. 仅在有助于控制结果时添加负面约束。
  6. 简要说明有意义的改动。

Input Extraction

输入提取

Extract different information from different input types:
  • Copy / ad text: core selling point, audience, brand voice, required text, visual metaphor, forbidden claims.
  • Product information: appearance, material, color, scale, use case, benefit, brand elements, and fact boundaries.
  • Script / story: characters, scene, action, emotional turn, time order, dialogue or captions, ending.
  • Reference-image description: identity, composition, style, color, material, pose, scene relationship, and allowed changes.
  • Existing prompt: original hook, ambiguity, contradictions, overloaded details, missing visual controls, removable generic words.
If the input is thin but sufficient, make a first pass and label assumptions. Ask questions only when product facts, exact text, reference use, or target modality is unclear enough to change the result.
根据不同输入类型提取相应信息：
  • 文案/广告文本：核心卖点、受众、品牌语气、必需文本、视觉隐喻、禁止使用的声明。
  • 产品信息：外观、材质、颜色、尺寸、使用场景、优势、品牌元素和事实边界。
  • 脚本/故事：角色、场景、动作、情绪转折、时间顺序、对话或字幕、结尾。
  • 参考图描述：身份特征、构图、风格、颜色、材质、姿势、场景关系和允许进行的改动。
  • 现有提示词：原始切入点、歧义、矛盾、过度堆叠的细节、缺失的视觉控制项，以及可删除的泛化词语。
如果输入内容较少但已足够完成初步处理，可以先执行并标注所做的假设。只有当产品事实、准确文本、参考素材的使用方式或目标模态不够明确，且会显著改变结果时，才需要提问。

Optimization Dimensions

优化维度

Strengthen only the dimensions that matter for the task:
DimensionWhat to sharpen
Story / beatsCause and effect, scene goal, emotional turn, who wants what, obstacle, result
ContinuityTime, place, wardrobe, props, character identity, product appearance
Shot languageShot scale, angle, focal-length feel, camera movement, establishing / insert / reaction coverage
Visual specificityLighting, color temperature, palette, materials, weather, era, spatial layers
Motion / videoPacing, action order, duration hints, transitions, subtitles or on-screen text rules
Image compositionFocal subject, background hierarchy, negative space, symmetry, rule of thirds, aspect ratio
AudioTimbre, pace, language, sound-effect layers, music role, ambience and spatial feel
仅强化与当前任务有关的维度：
维度强化重点
故事/节拍因果关系、场景目标、情绪转折、角色诉求、阻碍和结果
连续性时间、地点、服装、道具、角色身份和产品外观
镜头语言景别、角度、焦段感、摄像机运动，以及定场镜头、插入镜头和反应镜头的覆盖
视觉具体性光线、色温、配色、材质、天气、时代和空间层次
运动/视频节奏、动作顺序、时长提示、转场、字幕或屏幕文字规则
图像构图视觉焦点、背景层级、负空间、对称性、三分法和宽高比
音频音色、速度、语言、音效层次、音乐作用、环境声和空间感

Prompt Patterns

提示词模式

For image prompts, prioritize:
  • subject and exact visual identity
  • environment and background hierarchy
  • composition, camera angle, focal-length feel, and framing
  • lighting, color palette, material details, and texture
  • style, medium, and rendering quality
  • exact visible text if any
  • aspect ratio or output format when stated
For video prompts, prioritize:
  • subject and scene goal
  • action sequence and emotional progression
  • shot scale, camera movement, pace, and transition
  • lighting, environment, materials, and continuity
  • duration hints only when the user gave or implied them
  • audio, subtitles, or on-screen text only when relevant
For audio prompts, prioritize:
  • voice, mood, rhythm, language, and pacing
  • music role, instruments, ambience, and sound-effect layers
  • delivery style and production feel
对于图像提示词，优先考虑：
  • 主体及其准确的视觉身份
  • 环境和背景层级
  • 构图、摄像机角度、焦段感和取景
  • 光线、配色、材质细节和纹理
  • 风格、媒介和渲染质量
  • 需要呈现的任何准确可见文本
  • 用户已指定的宽高比或输出格式
对于视频提示词，优先考虑：
  • 主体和场景目标
  • 动作顺序和情绪递进
  • 景别、摄像机运动、节奏和转场
  • 光线、环境、材质和连续性
  • 仅在用户提供或暗示时添加时长提示
  • 仅在相关时加入音频、字幕或屏幕文字
对于音频提示词，优先考虑：
  • 声音、情绪、节奏、语言和速度
  • 音乐作用、乐器、环境声和音效层次
  • 演绎方式和制作质感

Ready-To-Use Templates

即用模板

Product visual:
text
Create a high-quality commercial visual for [product]. Preserve the true appearance: [material, color, shape, logo placement]. Show [use case / core benefit] through [scene or action]. Composition: [centered hero / rule of thirds / flat lay / three-quarter angle]. Lighting: [soft studio / natural daylight / high-contrast rim light]. Background: [clean solid color / lifestyle environment / brand-color setting]. Do not add unsupported certifications, slogans, or packaging copy.
Poster / typography:
text
Create a poster for [topic/product/event]. Exact visible text: "[text]". Clear hierarchy between headline, supporting information, and visual subject. Typography: [type direction], readable, balanced spacing. Visual style: [style]. Avoid misspelled text, extra words, crowded layout, and illegible small type.
Character image:
text
Create [character type], age/body/facial features: [description], hair: [description], outfit: [material, color, style], accessories: [description]. Pose: [pose]. Expression: [expression]. Style: [style]. Background: [background]. Keep a single character, consistent identity, no watermark, no extra text.
Video segment:
text
[Subject] performs [action sequence] in [scene]. The shot begins with [starting shot scale / angle] and follows with [camera movement]. Pacing: [slow / fast / escalating]. Lighting and color: [description]. Visual change: [key change]. Overall style: [style]. Avoid multiple location jumps, contradictory actions, and complex hard-to-render text.
UI / product interface:
text
Create a realistic UI mockup for [product/app]. Screen type: [desktop/mobile/dashboard]. Include [core components]. Clear information hierarchy, consistent spacing, readable labels, explicit interaction states. The result should feel like a real shippable product interface, not a decorative marketing poster.
产品视觉：
text
为 [产品] 创建高质量商业视觉图。保持真实外观：[材质、颜色、形状、Logo 位置]。通过 [场景或动作] 展现 [使用场景/核心优势]。构图：[居中主视觉/三分法/平铺/四分之三视角]。光线：[柔和棚拍/自然日光/高对比度轮廓光]。背景：[纯净纯色/生活方式环境/品牌色场景]。不要添加缺乏依据的认证、口号或包装文案。
海报/排版：
text
为 [主题/产品/活动] 创建海报。准确可见文本："[文本]"。在标题、补充信息和视觉主体之间建立清晰的层级关系。排版：[字体设计方向]，清晰易读、间距均衡。视觉风格：[风格]。避免文字拼写错误、多余词语、拥挤布局和难以辨认的小字号文字。
角色图像：
text
创建 [角色类型]，年龄/体型/面部特征：[描述]，头发：[描述]，服装：[材质、颜色、风格]，配饰：[描述]。姿势：[姿势]。表情：[表情]。风格：[风格]。背景：[背景]。保持单一角色和身份一致，不要水印，不要额外文字。
视频片段：
text
[主体] 在 [场景] 中执行 [动作顺序]。镜头从 [起始景别/角度] 开始，并采用 [摄像机运动] 跟随。节奏：[缓慢/快速/逐渐增强]。光线与色彩：[描述]。视觉变化：[关键变化]。整体风格：[风格]。避免场景跳切过多、动作矛盾和难以渲染的复杂文字。
UI/产品界面：
text
为 [产品/应用] 创建逼真的 UI 模型。屏幕类型：[桌面端/移动端/仪表盘]。包含 [核心组件]。建立清晰的信息层级，保持间距一致、标签易读，并明确展示交互状态。成品应像真正可交付的产品界面，而不是装饰性的营销海报。

Output Format

输出格式

Use this format unless the user requested another one:
text
Final Prompt
[ready-to-copy prompt]

Negative Prompt / Constraints
[only if useful]

Change Notes
- [short note]
- [short note]

Optional Variants
1. [optional variant, if useful]
2. [optional variant, if useful]
For new prompt creation, rename 
Final Prompt
to 
Generated Prompt
.
除非用户另有要求，否则使用以下格式：
text
最终提示词
[可直接复制的提示词]

负面提示词/约束
[仅在有用时填写]

修改说明
- [简短说明]
- [简短说明]

可选变体
1. [可选变体，如有用]
2. [可选变体，如有用]
对于新创建的提示词，将 
最终提示词
重命名为 
生成的提示词
。

Quality Checks

质量检查

  • The prompt preserves the user's subject, intent, and constraints.
  • Any assumptions are small, labeled, and necessary.
  • Product and brand claims are not invented.
  • The output is ready to copy into a generation tool.
  • Platform-specific flags are included only when the user requested a specific platform.
  • 提示词保留了用户的主体、意图和约束。
  • 所有假设都较小、经过标注且确有必要。
  • 没有编造产品和品牌声明。
  • 输出可直接复制到生成工具中使用。
  • 仅在用户指定特定平台时加入平台专属参数。

Anti-Patterns

反模式

  • Do not rewrite a simple product prompt into an unrelated story.
  • Do not replace concrete user details with generic words like "cinematic" or "premium" without actionable visual specifics.
  • Do not add platform-specific flags such as seeds, CFG, sampler, model IDs, or resolution suffixes unless the user requests a specific platform.
  • Do not produce a full storyboard unless the user asks for multi-shot planning.
  • 不要将简单的产品提示词改写成无关的故事。
  • 不要在缺少可执行视觉细节时，用“电影感”或“高端”等泛化词语替代用户的具体描述。
  • 除非用户指定特定平台，不要添加 seed、CFG、sampler、model ID 或分辨率后缀等平台专属参数。
  • 除非用户要求进行多镜头规划，否则不要制作完整的 storyboard。