SuperAPI AI Directory
Partner中文AI newsFavorites
Skillsimage-to-prompt

image-to-prompt

Design & creation

View repository

Updated 10-16 10:21

About this skill

Image to Prompt - Image/Text to AI Painting Prompts

Convert user input (image or text) into high-quality, structured AI painting prompts.

Work Steps

Create a task checklist and perform the following two tasks:

  1. Process input
  2. Detailed analysis
  3. Generate output

Image input

Determine image source:

  1. Path/URL input(such as /path/to/image.png or https://example.com/image.jpg）

    • Use an image analysis tool (mcp__zai-mcp-server__analyze_image) to extract detailed visual information
    • Build prompts based on the analysis results

  2. Directly pasted input(the user pastes the image into the conversation)

    • Use the model's own visual capabilities to directly analyze the image in the conversation
    • Describe the image content in detail (no need to call external tools)

Analysis dimensions(flexibly adjust according to image content; complex images can be expanded to at most 4k analysis dimensions):

1. 核心视觉层 (Core Visual Layer)
Subject：
具象维度：识别人物/物体/生物的数量、姿态、动作、表情、服饰及物理特征。
抽象维度：若无具体主体，描述其形态（如流体、几何体）、运动态势（如旋转、爆炸、静止）及视觉聚合点。
Composition：
空间布局：分析画面布局（向心、三分法/Rules of thirds、黄金分割/Golden ratio）、焦点位置、对称性/非对称性。
视线引导：线条流向、层次感（前景/中景/背景）、画幅比例及其带来的视觉影响。
Color：
基础参数：主色调、配色方案（如互补色/Complementary、同类色/Analogous）、对比度、饱和度。
色彩心理：色彩传递的温度（冷/暖）、重量感及情感联想。
Light：
物理属性：光源方向（顶光/Top light、侧光/Side light、逆光/Backlight）、光线质感（柔光/Soft light、硬光/Hard light、漫反射/Diffused）。
氛围营造：明暗调性（High-key/Low-key）、阴影的形态与深浅、环境光感。
2. 风格与技法层 (Style & Technique Layer)
Style：
流派定位：艺术流派（如印象派/Impressionism、极简主义/Minimalism、赛博朋克/Cyberpunk）、时代感、文化特征。
视觉语言：摄影风格（如黑色电影/Film noir、纪实摄影/Documentary）或绘画技法（如厚涂/Impasto、水墨/Ink wash）。
Technical Details：
质感与纹理：材质表面（光滑/Glossy、粗糙/Rough）、颗粒感/噪点（Grain）、笔触痕迹。
特殊效果：光学瑕疵模拟（如色散/Chromatic aberration、暗角/Vignette）、动态模糊（Motion blur）、景深/虚化（Bokeh/Depth of field）、光晕（Bloom）。
3. 认知与叙事层 (Cognitive & Narrative Layer)
Micro-Details：
捕捉容易被忽略的微小元素（如背景中的隐藏物体、微小的文字信息、材质的细微破损）。
Emotion & Atmosphere：
整体基调：画面传递的第一直觉（如压抑、欢快、神秘、宁静）。
深层情感：分析画面元素如何引发特定的心理反应（如焦虑、孤独、希望）。
Semiotics & Metaphor：
叙事暗示：画面仿佛在讲述什么故事？捕捉的是哪个瞬间（发生前/发生后）？
象征意义：分析图像中的符号（如”镜子”代表内省，”风暴”代表混乱）及其文化互文性。
Synesthesia：
描述观看图像时引发的非视觉联想（如听觉上的噪音/寂静、触觉上的冷热/刺痛感）。

Analysis example(power pole in dense fog):

### **1. 核心视觉层 (Core Visual Layer)**

*   **Subject**：
    *   **具象维度**：画面中有三个主要的垂直结构——**木质电线杆**。前景的一根最为显著，略微向右倾斜，顶端带有横臂和变压器状物体，几根断裂或松弛的电线垂落下来。背景中还有两根较模糊的电线杆，呈线性排列延伸至远处。在画面右侧中景处，有一个极为渺小的**人形剪影**，似乎是一个穿着深色衣服的人孤独伫立。
    *   **抽象维度**：电线杆构成了强烈的**垂直线条**，与地面的水平线形成交错，但在浓雾的侵蚀下，这些线条显得脆弱且不完整。垂落的电线形成了无力的曲线，暗示着重力与废弃。

*   **Composition**：
    *   **空间布局**：采用**Deep depth of field**构图，利用透视原理将视线引向远方。前景巨大的倾斜电线杆占据了左侧主导地位，与右侧远处的微小人物形成**大小对比**和**视觉平衡**。
    *   **视线引导**：电线杆的排列形成了一条隐形的对角线引导线，将观众的视线从左前方向右后方牵引，最终消失在迷雾中。
    *   **画幅比例**：竖构图增强了画面的压抑感和纵深感，强调了天空（雾气）的沉重。

*   **Color**：
    *   **基础参数**：**极低饱和度**，接近单色画。主色调为灰绿色（Sage Green/Grey）和暗褐色（Sepia/Dark Brown）。
    *   **色彩心理**：这种暗淡、浑浊的色调传递出**寒冷、潮湿、陈旧**和**死寂**的感觉。缺乏暖色调，完全剥夺了画面的生机与希望感。

*   **Light**：
    *   **物理属性**：**极度柔和的漫反射光**（Soft/Diffused light）。由于浓雾的存在，光源方向不可辨认，光线平坦且均匀，没有强烈的阴影。
    *   **氛围营造**：属于**Low-key**摄影，整体亮度偏低。雾气作为一种介质，散射了光线，导致物体边缘模糊，营造出一种梦境般或末日后的朦胧感。

### **2. 风格与技法层 (Style & Technique Layer)**

*   **Style**：
    *   **流派定位**：
        *   **Southern Gothic**：荒凉的田野、废弃的工业设施，透露出一种衰败和不安。
        *   **Liminal Space**：这是一个过渡性的场所，既非起点也非终点，充满了不确定性和诡异的熟悉感。
        *   **Silent Hill Aesthetic**：浓雾、废弃设施和孤独的人影是该风格的标志性元素。
    *   **视觉语言**：具有强烈的**Cinematic**电影感，像是一部悬疑片或恐怖片的定场镜头。

*   **Technical Details**：
    *   **质感与纹理**：画面叠加了明显的**Film Grain**，增加了粗糙的质感，模拟了老旧照片或高感光度底片的效果。这种噪点不是瑕疵，而是增强了”真实感”和”纪实感”。
    *   **特殊效果**：**Atmospheric perspective**被运用到了极致，雾气随着距离增加而变浓，使得远处的物体逐渐溶解在背景中。

### **3. 认知与叙事层 (Cognitive & Narrative Layer)**

*   **Micro-Details**：
    *   **垂落的电线**：这不仅是物理细节，更是**功能失效**的标志。电线本应紧绷传递能量，这里的垂落意味着断联、废弃和文明的退场。
    *   **倾斜的角度**：前景电线杆不仅破旧，而且是歪斜的，暗示了地基的不稳或曾遭受过外力的撞击/风暴，处于一种摇摇欲坠的临界状态。

*   **Emotion & Atmosphere**：
    *   **整体基调**：**Isolation**、**Desolation**、**Mystery**。
    *   **深层情感**：这种迷雾中的静止画面容易引发**存在主义焦虑**。那个渺小的人影面对巨大的、废弃的工业遗迹，强化了人类在自然或不可知力量面前的渺小与无力。

*   **Semiotics & Metaphor**：
    *   **电线杆**：通常象征着沟通、连接和现代文明。在这里，它们是断裂和废弃的，象征着**沟通的失败**或**文明的遗迹**。
    *   **迷雾**：象征着**未知**、**迷失**和**隔离**。它遮蔽了视野，让人看不清前路，也看不清来路。
    *   **孤独的观察者**：那个微小的人影可以被看作是观众的**Avatar**，迷失在这个充满迷雾的异世界中。

*   **Synesthesia**：
    *   **听觉**：看着这张图，仿佛能听到**风吹过枯草的沙沙声**，或者远处沉闷的**低频嗡嗡声**，亦或是**绝对的死寂**（耳鸣般的安静）。
    *   **触觉**：能感受到**潮湿粘腻的空气**粘在皮肤上，以及刺骨的**阴冷**。
    *   **嗅觉**：空气中似乎弥漫着**潮湿泥土**、**腐烂植被**和**生锈金属**的气味。

**总结**：
这张图像是一幅极具表现力的**反乌托邦风景画**。它通过**迷雾**这一核心元素，成功地将现实世界异化，利用**废弃的工业符号**（破损的电线杆）和**孤独的人影**，构建了一个关于**遗忘、断联和孤独**的视觉叙事。它不仅仅是在展示一个场景，更是在唤起一种深入骨髓的寒意和对未知的敬畏。

End of analysis example.

Build the prompt after analysis:

  1. Integrate the above analysis results
  2. If the user specifies a target style, prioritize descriptions of that style
  3. Generate the prompt

Text input

  1. Parse the core visual elements in the description
  2. Identify implicit style tendencies (such as "dreamy" implying soft lighting)
  3. Add reasonable details to make the prompt complete

Mixed input (the user provides both an image and a prompt)

You need to carefully identify the user's specific intent; for example, the user may want to:

  • Generate a new prompt based on the image description
  • Merge the image and text description to create a richer prompt

Detailed analysis

  • Specific rather than general: do not say "beautiful lighting", but rather "golden hour side lighting casting long shadows"
  • Layered description: from overall composition → subject details → background environment
  • Consistent style: ensure that style, color, and lighting descriptions are coordinated with each other
  • Avoid excessive stacking: the principle of moderation for quality words

Style definition techniques

Reference artists/works

When a style needs to be precisely defined, you may cite:

  • Artist：in the style of [Artist Name] or style mixing (such as:Katsuhiro Otomo meets Moebius）
  • Artwork: reference the aesthetic characteristics of a specific work (such as:Blade Runner 2049 cinematography aesthetic）
  • Genre/Movement：Art Nouveau, De Stijl, Ukiyo-e, Bauhaus

Principle: 1-2 references are enough; too many will confuse the style. Prioritize the references most directly relevant to the user's intent.

Lens and photography parameters

For realistic or photographic styles, you may add:

  • Lens type：wide-angle lens, 85mm portrait lens, fisheye, macro
  • Aperture/depth of field：shallow depth of field, f/1.8, bokeh background
  • Composition terminology：Dutch angle, worm's eye view, bird's eye view, rule of thirds
  • Film/Equipment：shot on Kodak Portra 400, Hasselblad medium format, vintage Polaroid

Principle: Use only when the photographic feel is the core of the style; illustration/concept art types do not need to be forced to add it.

Negative Prompt

When generating realistic types or when specific elements need to be excluded, it is recommended to add a negative prompt:

  • General exclusions：low quality, blurry, distorted, deformed, ugly, duplicate, watermark, signature
  • Portrait exclusions：bad anatomy, extra limbs, missing fingers, mutated hands, poorly drawn face
  • Style exclusions: Exclude opposite features according to the target style (e.g., for realistic style, excludecartoon, anime, illustration）

Principle: Negative prompts are added only when the user explicitly requests them or the style requires them; usually 3-5 are enough, and too many will dilute the effect.

Weight and Emphasis

When specific elements need to be strengthened or weakened:

  • Emphasis：(keyword) or (keyword:1.2) increase weight
  • Weakening：[keyword] or (keyword:0.8) reduce weight
  • Avoid excess: A weight range of 0.5-1.5 is appropriate; too high will cause distortion

When to apply: Use when the description contains multiple elements that need to be distinguished by primary and secondary importance, or when certain words are easily ignored.

Aspect Ratio Suggestions

Recommended ratios according to compositional intent:

  • 1:1 — avatar, icon, symmetrical composition
  • 16:9 / 2:1 — landscape, panorama, cinematic feel
  • 9:16 — full-body portrait, vertical poster
  • 4:3 — illustration, document illustration
  • 21:9 — ultra-wide cinematic

Principle: The ratio serves the compositional intent; a wide frame strengthens the sense of extension, a square frame focuses on the center, and a vertical frame strengthens the sense of depth. One recommended ratio is enough.

By default, each request outputs only theChinese version, with English annotations after professional terms (e.g., 赛博朋克(Cyberpunk), 景深(Depth of field)). Output the English version only when the user explicitly requests it. Directly output the corresponding prompt (plaintext\n<提示词>\n) is sufficient.

In the Chinese version,annotate professional terms in English, such as:

  • Art styles (e.g., 赛博朋克(Cyberpunk), 巴洛克(Baroque))
  • Technical terms (e.g., 景深(Depth of field), 三分法(Rule of thirds))
  • Materials/textures (e.g., 厚涂(Impasto), 丝网印刷(Screen printing))

Prompt References

  • Banana (default): Do not write negative prompts; the total prompt length is around 1000 characters, refer tobanana prompt examples
  • Qwen-Image: The prompt is around 500 characters, refer toqwen-image prompt examples
  • Midjourney: Add --ar [比例]、--stylize [值]、--v 6 and other parameters
  • SD/FLUX: Add a negative prompt
  • DALL-E: The description is more natural-language-like

Other Requirements

  • Overall, the prompt should not exceed 2000 characters