AI Prompt Engineering · 课时

图像生成提示词的构成

Q: 「图像生成提示词的构成」这节课中我会学到什么？

主体、风格、媒介、光照、色彩搭配和构图元素。 你通过在浏览器中直接运行的动手代码来练习 AI Prompt Engineering，全天候 AI 导师会在你学习这节课的过程中回答你的问题。

主体、风格、媒介、光照、色彩搭配和构图元素。

图像生成提示词的构成 是 CoddyKit 上的免费 AI Prompt Engineering 课时。这是第 1 节课，共 4 节。你可以在下方免费阅读本课时的完整内容 — 然后在浏览器中使用内置代码编辑器和全天候 AI 导师进行实践。这是 AI Prompt Engineering 学习路径的一部分，你的进度在网页和 CoddyKit 应用中同步。 AI Prompt Engineering 课程共包含 4 节课。

作为配方的图像提示词

图像生成提示词是视觉效果的配方。这就像烹饪一样，素材的顺序和配比十分重要。缺少元素会产生普通的结果；加入错误元素则会导致出人意料的失败。掌握提示词的组成结构，才能获得可预测的创作控制力。

六个核心组成部分

每个优秀的图像提示词都包含六个组成部分：主体（是什么）、风格（呈现方式）、媒介（艺术形式）、光照（照明）、色彩调色板（色调）和构图（取景）。省略任何一个部分都会让模型自行猜测，通常会采用普通的默认设置。

# Anatomy of an image prompt
prompt_components = {
    'subject':    'A lone lighthouse on a rocky coastline',
    'style':      'dramatic, moody, cinematic',
    'medium':     'oil painting',
    'lighting':   'stormy overcast sky, waves crashing, dramatic side-lighting',
    'color_palette': 'desaturated blues and grays with warm amber light from the lighthouse',
    'composition': 'wide establishing shot, rule of thirds, lighthouse at left third'
}

# Assemble into a prompt string
full_prompt = (
    '{subject}, {style}, {medium}, {lighting}, '
    '{color_palette}, {composition}'
).format(**prompt_components)

print(full_prompt)
# A lone lighthouse on a rocky coastline, dramatic, moody, cinematic,
# oil painting, stormy overcast sky, waves crashing, dramatic side-lighting,
# desaturated blues and grays with warm amber light from the lighthouse,
# wide establishing shot, rule of thirds, lighthouse at left third

主体：是什么

主体是最重要的组成部分，它告诉模型要描绘什么。请尽可能具体：包括物种、年龄、情绪、动作、环境和空间关系。

# Subject specificity comparison

# Weak subject:
weak = 'a person in a city'

# Strong subject (same concept, much more specific):
strong = (
    'a young woman in her 30s, wearing a vintage 1960s trench coat, '
    'standing at a rain-soaked street corner in Tokyo at night, '
    'looking up at neon signs reflected in the puddles, '
    'holding a dripping umbrella, expression of quiet wonder'
)

# The strong subject answers:
# WHO: young woman, 30s
# WHAT WEARING: 1960s trench coat
# WHERE: Tokyo street corner
# WHEN: night, raining
# WHAT DOING: standing, looking up
# EXPRESSION: quiet wonder
# KEY DETAIL: neon reflections in puddles, dripping umbrella

print('Weak:', weak)
print('Strong:', strong[:100], '...')

风格：如何呈现

风格描述词告诉模型图像应采用的视觉语言和审美格调。它们可以参考艺术流派、特定艺术家、视觉媒介或抽象的审美特质。

style_examples = [
    # Art movements
    'impressionist', 'art nouveau', 'bauhaus', 'minimalist', 'surrealist',

    # Artist references
    'in the style of Monet', 'reminiscent of Hopper', 'inspired by Klimt',

    # Visual media
    'film noir', 'vaporwave aesthetic', 'cottagecore', 'brutalist',

    # Quality descriptors
    'highly detailed', 'cinematic', 'editorial photography style',
    'concept art', 'matte painting', 'character design sheet',

    # Mood
    'ethereal', 'gritty', 'whimsical', 'melancholic', 'vibrant', 'serene'
]

# Combining styles creates unique aesthetics
combined_style = 'cyberpunk aesthetic meets art nouveau, highly detailed, dark ethereal'
print('Combined style:', combined_style)

# Warning: too many style directives create incoherence
too_many = 'impressionist, minimalist, surrealist, photorealistic, anime, baroque'
print('Too many (incoherent):', too_many)

媒介：艺术形式

媒介定义实体或数字艺术形式。它会从根本上改变生成图像的纹理、线条质感和色调。常见媒介及其产生的效果如下：

medium_guide = {
    # Traditional media
    'oil painting': 'Rich, textured, classic look with visible brushwork',
    'watercolor': 'Soft edges, translucent washes, paper texture visible',
    'pencil sketch': 'Line art, cross-hatching, grayscale, raw feel',
    'charcoal drawing': 'Soft, smudgy, high contrast, dramatic shadows',
    'ink illustration': 'Bold lines, flat colors or crosshatching',

    # Photography
    'photorealistic': 'Looks like a real photograph',
    'film photography': 'Grain, color shift, analog feel',
    'macro photography': 'Extreme close-up, shallow depth of field',
    'long exposure photography': 'Motion blur, light trails',

    # Digital / 3D
    '3D render': 'CGI quality, precise geometry',
    'octane render': 'Photorealistic 3D with ray-tracing quality',
    'Blender 3D': 'CGI aesthetic, often used with subdivision modeling',
    'pixel art': 'Retro 8-bit or 16-bit style, visible pixels',
    'vector illustration': 'Clean, flat, scalable design style'
}

for medium, description in list(medium_guide.items())[:5]:
    print(f'{medium}: {description}')

光照：塑造氛围

光照是视觉艺术中最有力的氛围塑造因素。同一主体在黄金时刻的阳光下，与在刺眼的审讯室灯光下，会呈现完全不同的效果。请掌握光照相关词汇。

lighting_vocabulary = {
    # Time of day
    'golden hour': 'Warm orange-yellow, long shadows, magic hour feel',
    'blue hour': 'Cool blue twilight, soft diffused light',
    'harsh midday': 'Hard shadows, washed-out, unflattering (usually avoided)',
    'overcast': 'Soft even light, no shadows, good for portraits',

    # Studio / artificial
    'studio lighting': 'Controlled, professional, even illumination',
    'Rembrandt lighting': 'Triangle of light on cheek, dramatic portrait technique',
    'neon lighting': 'Colorful, urban, cyberpunk feel',
    'candlelight': 'Warm, flickering, intimate',

    # Dramatic
    'chiaroscuro': 'Extreme light/dark contrast, Baroque dramatic style',
    'volumetric lighting': 'God rays, light shafts through fog or dust',
    'backlit / rim light': 'Subject outlined by light from behind, halo effect',
    'bioluminescent': 'Glowing from within, alien or underwater feel'
}

example = 'volumetric lighting, golden hour, warm glow filtering through forest canopy'
print('Lighting example:', example)

色彩调色板：情绪基调

色彩调色板控制图像的情绪基调和视觉统一性。您可以用描述性语言指定调色板，也可以参考艺术流派、电影或自然景观。

color_palette_examples = [
    # Temperature-based
    'warm earth tones: rust, ochre, sienna, cream',
    'cool blues and silvers, icy palette',
    'neutral gray monochrome with single red accent',

    # Mood-based
    'muted, desaturated, melancholic color grading',
    'vibrant saturated colors, tropical energy',
    'pastel soft colors, dreamlike softness',

    # Reference-based
    'Wes Anderson color palette: pastel pinks and greens',
    'film noir: high contrast black and white with amber shadows',
    'synthwave neon: pink, purple, cyan on dark backgrounds',

    # Nature-based
    'autumn forest: burnt orange, golden yellow, deep brown',
    'arctic palette: white, pale blue, grey with deep navy accents',
]

# Color palettes can clash with lighting — ensure they work together
clash = 'vibrant tropical colors + film noir lighting'  # incoherent
harmony = 'warm amber and ochre tones + golden hour lighting'  # coherent

print('Harmonious combo:', harmony)

构图：取景与焦点

构图指令告诉模型如何取景——应包含什么、应强调什么，以及应将各元素放置在哪里。借用摄影和电影摄影中的词汇通常很有效。

composition_vocabulary = {
    # Camera distance
    'extreme close-up': 'fills frame with a single detail (eye, hand, texture)',
    'close-up': 'face or object fills most of frame',
    'medium shot': 'waist to head, character-focused',
    'wide shot': 'full body in environment context',
    'establishing shot': 'landscape/environment, tiny or no character',
    'aerial / bird\'s eye view': 'looking straight down',
    'worm\'s eye view': 'looking straight up from below',

    # Composition rules
    'rule of thirds': 'subject at intersection of 1/3 lines',
    'centered composition': 'symmetrical, formal, powerful',
    'leading lines': 'lines guide eye toward subject',
    'negative space': 'large empty area emphasizes subject',
    'frame within frame': 'archway, window, or shape frames subject',

    # Depth
    'shallow depth of field': 'sharp subject, blurred background (bokeh)',
    'deep focus': 'everything in sharp focus front to back'
}

print('Composition example: close-up portrait, rule of thirds, shallow depth of field, bokeh background')

完整组装提示词

按照合理顺序组装全部六个组成部分，可以持续生成高质量的图像提示词。推荐顺序为：主体 → 媒介 → 风格 → 光照 → 色彩调色板 → 构图。

def build_image_prompt(
    subject, medium, style, lighting, color_palette, composition,
    quality_boost='highly detailed, 8K resolution'
):
    parts = [
        subject,
        medium,
        style,
        lighting,
        color_palette,
        composition,
        quality_boost
    ]
    # Filter out None values and join
    return ', '.join(p for p in parts if p)

# Example: Portrait
portrait = build_image_prompt(
    subject='elderly Japanese fisherman mending nets at sunrise, weathered hands, peaceful expression',
    medium='oil painting',
    style='impressionist, highly detailed, classical technique',
    lighting='golden hour sunrise, warm light from the left, soft shadows',
    color_palette='warm ochres, golds, deep navy sea in background',
    composition='medium shot, rule of thirds, subject right third, ocean left'
)
print(portrait)

为多张图像保持一致性

生成一系列图像（角色设定表、分镜图）时，请使用基础提示词模板，并为不同图像中会变化的元素替换变量，以保持一致性。

BASE_CHARACTER_TEMPLATE = (
    '{character_description}, '
    '{action_description}, '
    'oil painting, concept art style, '
    'dramatic studio lighting, '
    'muted jewel tones with gold accents, '
    'close-up portrait, centered composition, '
    'highly detailed, cinematic'
)

character = (
    'a tall warrior woman with dark braided hair, emerald eyes, '
    'wearing ornate silver plate armor with dragon motifs'
)

action_variants = [
    'standing at attention, stoic expression, arms crossed',
    'battle-ready, sword raised, fierce expression, mid-action',
    'resting against a stone wall, exhausted but resolute, soft smile',
    'close-up portrait, neutral expression, slight three-quarter view'
]

for action in action_variants:
    prompt = BASE_CHARACTER_TEMPLATE.format(
        character_description=character,
        action_description=action
    )
    print(f'Generating: {action[:40]}...')
    # call image API with prompt

质量标记词

大多数图像模型都会响应质量标记词——这些术语用于表明所需的输出保真度。请有策略地使用它们来提升细节水平，同时不改变主体或风格。

quality_tokens = {
    'high_detail': [
        'highly detailed', 'intricate detail', 'ultra-detailed',
        '8K resolution', '4K wallpaper quality'
    ],
    'photorealism': [
        'photorealistic', 'hyperrealistic', 'photographed by',
        'DSLR photo', 'shot on 35mm film'
    ],
    'professional_quality': [
        'award-winning', 'professional photography', 'editorial quality',
        'museum quality', 'masterpiece'
    ],
    'rendering_quality': [
        'octane render', 'unreal engine 5', 'ray tracing',
        'subsurface scattering', 'global illumination'
    ]
}

# Best practice: add 2-3 quality tokens at the end of the prompt
quality_suffix = 'highly detailed, cinematic, 8K resolution'
print('Add to end of any prompt:', quality_suffix)

# Warning: do not overload with quality tokens — they dilute each other
too_many_quality = 'masterpiece, best quality, ultra-detailed, perfect, amazing, award-winning'
print('Over-specified (less effective):', too_many_quality[:60])

快速检查

图像生成提示词中的哪个组成部分最直接地控制图像的情绪氛围？

图像提示词结构总结

构建良好的图像提示词包含六个相互协调的组成部分：

主体：WHO 和 WHAT——最重要的组成部分；应描述得非常具体
风格：审美格调、艺术流派或艺术家参考
媒介：艺术形式（油画、写实照片、三维渲染、像素艺术）
光照：塑造氛围的照明（黄金时刻、明暗对照法、霓虹灯）
色彩调色板：情绪温度（温暖的大地色、冷峻的低饱和色、霓虹色）
构图：取景（特写、三分法、建立镜头）

最后添加 2—3 个质量标记词。使用模板来保持多图系列的一致性。

常见问题解答

「图像生成提示词的构成」课时是免费的吗？

是的 — 「图像生成提示词的构成」的完整文本可在网页上免费阅读。要进行交互式练习（内置代码编辑器和全天候 AI 导师）并解锁 AI Prompt Engineering 课程的其余内容，请升级到 CoddyKit PRO。 AI Prompt Engineering 课程共包含 4 节课。

「图像生成提示词的构成」这节课中我会学到什么？

主体、风格、媒介、光照、色彩搭配和构图元素。你通过在浏览器中直接运行的动手代码来练习 AI Prompt Engineering，全天候 AI 导师会在你学习这节课的过程中回答你的问题。

学习 AI Prompt Engineering 需要有经验吗？

无需任何先前经验。CoddyKit 上的 AI Prompt Engineering 课程适合初学者到高级学习者，你可以从这里开始或从头开始，按照自己的节奏学习。这是第 1 节课，共 4 节。

「图像生成提示词的构成」课时需要多长时间？

大多数 CoddyKit 课程大约需要 5–10 分钟。每节课都很精短且互动，所以你能稳步进步，并在网页和应用中从离开的地方继续。

我能在这节 AI Prompt Engineering 课中编写并运行代码吗？

能。每节 AI Prompt Engineering 课都包含内置代码编辑器，你可以在浏览器中直接编写并运行真实代码，并获得即时 AI 反馈 — 无需本地设置。

此课程中的所有课时

← 返回 AI Prompt Engineering