Which elements should an image-generation prompt include?

The chapter lists five core slots: subject (person / object / scene), style (photo / illustration / pixel / 3D), composition (angle, depth of field, subject position), lighting (warm or cool, soft or hard, time of day), colour palette (main plus accent), and finally detail constraints (materials, key elements). Drop any slot and the model improvises; fill them all and the output stays on-style.

When an image prompt drifts, what is the most efficient fix?

The chapter gives three iteration rules: coarse before fine (lock subject and style first, add detail later), positive before negative (say what you want before what you do not), and change one variable per iteration. The common mistake is rewriting five words at once, then not knowing which one moved the result. Keep every prompt version as a changelog so you can roll back.

Can I reuse the same prompt template across Midjourney, Stable Diffusion and DALL-E?

The skeleton transfers, the keywords do not. The chapter's `subject / style / composition / lighting / colour / detail` skeleton works on all three platforms, but weighting syntax (Midjourney's `::`, SD's `(word:1.3)`) and negative prompts (only SD has a real negative-prompt field) do not. Reuse the skeleton across platforms and tune the details against each tool's docs.

If the subject keeps coming out blurry or off-position, how do I fix the prompt?

Three moves: (1) add subject position and action — e.g. `centred composition, headline middle, whitespace below`; (2) specify camera angle and depth of field — e.g. `45-degree side overhead, shallow DoF`; (3) lock style with named references — e.g. `modern flat illustration with subtle paper texture`. The brand-poster example in the chapter shows exactly that fully-specified structure.

Image Generation

Q: For brand visuals, how do I keep multiple images on the same style?

Pin style, palette and composition as a locked block at the top of every prompt and only swap subject and detail below. Add a style-anchor phrase — e.g. `modern flat illustration, slight paper texture, warm palette: orange + cream + deep blue accent` — into every prompt. If the tool supports a reference image, feeding the same reference is the most stable way to enforce consistency.

Image generation prompts (overview)

This section collects prompts for exploring image generation / multimodal capabilities. The focus is on iterative prompt refinement to gradually steer the output toward what you want. The core of image tasks: clear visual goal + structured description + repeatable iteration strategy.

Common Scenarios

Product visual direction exploration (style, mood, color palette)
Quick visual asset drafts (posters, covers, illustrations)
Brand identity generation (characters, mascots, IP)
Scenario concept validation (interior/street/futuristic)

Key Elements (spell these out every time)

Subject: Person/object/scene
Style: Realistic/illustration/pixel/3D/paper craft
Composition: Angle, depth of field, subject position
Lighting & color tone: Warm/cool, soft/hard, time of day
Detail constraints: Materials, textures, key elements

Prompt Template (General)

Subject: {{SUBJECT}}
Style: {{STYLE}}
Composition: {{COMPOSITION}}
Lighting: {{LIGHTING}}
Color tone: {{COLOR_TONE}}
Details: {{DETAILS}}

Example 1: Brand Visual Direction

Subject: Hero poster for an AI learning brand
Style: Modern flat illustration, slight paper texture
Composition: Center-symmetric, main title centered, whitespace below
Lighting: Soft light, subtle gradient
Color tone: Warm palette, orange + off-white + dark blue accents
Details: Abstract circuit lines, learning elements (books, notes)

Example 2: Product Scene

Subject: Smart learning device on a desk
Style: Realistic photography
Composition: 45-degree overhead angle, shallow depth of field
Lighting: Afternoon natural light, entering from the left
Color tone: Neutral, slightly warm
Details: Wooden desk, coffee cup, sticky notes

Example 3: Character / IP

Subject: A cute AI assistant cat
Style: 3D cartoon
Composition: Front-facing half body, smiling
Lighting: Soft studio lighting
Color tone: Blue and white color scheme
Details: Wearing small headphones, simple logo on chest

Iteration Strategy

Rough first, details later: Lock down subject and style first, then add details
Positive first, then constraints: Describe "what you want" first, then add "what you don't want"
Change one variable at a time: Only tweak one element per iteration to isolate the effect
Log versions: Save each prompt for easy rollback and comparison

Common Problems & Fixes

Blurry subject: Add subject position and action descriptions
Style drift: Pin down style keywords + reference objects
Messy composition: Specify camera angle, depth of field, subject placement
Missing details: Add material and key element qualifiers

Index

/learn/prompt-master/prompt-image-generation-alphabet-person

📚 相关资源

❓ 常见问题

关于本章主题最常被搜索的问题，点击展开答案

图像生成 Prompt 应该写哪几个要素？

本章列了 5 个：主体（人物/物体/场景）、风格（写实/插画/像素/3D）、构图（视角/景深/主体位置）、光线（暖冷/柔硬/时间）、色调（主色 + 点缀色）、再加细节（材质、关键元素）。任何一个不写，模型就自由发挥；写齐了出图风格才稳定可控。

图像 Prompt 出图跑偏，最高效的修法是什么？

本章给了三条迭代铁律：先粗后细（先定主体+风格再补细节）、先正向再约束（先说要什么再说不要什么）、每次只改一个变量。常见错误是一次性改 5 个词，最后不知道是哪个起作用。建议保留每版 prompt 当 changelog，方便回滚。

Midjourney / Stable Diffusion / DALL-E 用同一套 Prompt 模板可以吗？

结构能复用，关键词不能。本章的「主体 / 风格 / 构图 / 光线 / 色调 / 细节」骨架对三家都成立，但权重语法（Midjourney 的 `::`、SD 的 `(word:1.3)`）和负面提示（SD 才支持 negative prompt 字段）不通用。换平台时复用骨架，细节再按各家文档调。

出图主体老是模糊或位置乱，Prompt 要怎么改？

三步：1) 增加主体的位置和动作描述（「中心对称，主标题居中，下方留白」）；2) 指定镜头与景深（「45 度侧俯视，浅景深」）；3) 用风格关键词锁定参考对象（「现代扁平插画，轻微纹理纸感」）。本章「品牌视觉海报」示例就是这种全要素结构。

做品牌视觉时，怎么用 Prompt 保证多张图风格一致？

把风格、色调、构图作为「锁定段」固定写在每条 Prompt 顶部，只换主体和细节。再加一个 style anchor 词组（例如「modern flat illustration, slight paper texture, warm palette: orange + cream + deep blue accent」），所有图都带这一段。能 reference image 的工具直接传同一张参考图最稳。