Back to Details

scene-to-art-lab

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

Scene to Art Lab

场景转艺术实验室

Turn a supplied photograph into a deliberately art-directed 3:4 poster rather than a styled photograph. Preserve only the scene features that carry identity, then redesign composition, color, mark, space, graphic structure, and material according to one coherent medium and one poster idea. Do not assume that the whole source image must remain visible or receive equal artistic treatment.
Use image generation for every requested transformation. A supplied image plus a request to transform it is sufficient consent; do not reconfirm. Inspect the image before editing when it is available as a local file.
将提供的照片转化为经过精心艺术指导的 3:4 海报，而不是仅对照片进行风格化处理。仅保留承载主体识别特征的场景元素，然后依据一种连贯的媒介和统一的海报概念，重新设计构图、色彩、笔触、空间、图形结构与材质。不要预设整张源图都必须完整可见，也不要让所有细节获得同等程度的艺术处理。
每次请求的转换都必须使用图像生成。用户提供了图像并要求对其转换，即构成充分授权，无需再次确认。如果图像可作为本地文件访问，请在编辑前先检查图像。

Core workflow

核心工作流

  1. Read the source image.
  2. Choose or infer one primary art mode.
  3. Define a concise poster proposition: what the viewer should notice or feel first.
  4. Select one poster layout system and redesign the composition for the selected medium and a 3:4 canvas.
  5. Write a scene-specific transformation brief containing every mandatory output constraint.
  6. Generate the image directly.
  7. Inspect the result when possible; revise once if identity, poster hierarchy, layout, edge loss, or medium behavior failed.
  8. Return the image with a short Chinese rationale. Provide the full generation prompt only when requested.
If the user names a mode, follow it. If not, infer the strongest match from the source and explain the choice briefly after generation. Ask a question only when two materially different interpretations are equally plausible and the choice cannot be inferred from the request or image.
  1. 读取源图。
  2. 选择或推断一种主要艺术模式。
  3. 定义简洁的海报命题：明确观者应首先注意或感受到什么。
  4. 选择一种海报版式系统，并针对所选媒介和 3:4 画布重新设计构图。
  5. 编写针对当前场景的转换简报，其中包含所有强制输出约束。
  6. 直接生成图像。
  7. 尽可能检查结果；如果主体识别特征、海报层级、布局、边缘消隐或媒介表现未达到要求，则修订一次。
  8. 返回图像并附一段简短的中文说明。仅在用户明确要求时提供完整的生成提示词。
如果用户指定了艺术模式，应严格遵循；如果未指定，则从源图中推断最匹配的模式，并在生成后简要说明选择。只有当两种在实质上不同的解读同样合理，且无法根据请求或图像推断时，才向用户提问。

Read the source

读取源图

Create a compact internal Source Map:
  • Identity anchors: up to three features that must remain recognizable.
  • Spatial logic: horizon, overlap, pose, perspective, direction, or scale relationship that organizes the scene.
  • Energy: still, lyrical, graphic, tense, playful, monumental, intimate, or kinetic.
  • Shape hierarchy: dominant silhouette, secondary masses, and expendable detail.
  • Color opportunity: colors worth preserving, suppressing, exaggerating, or replacing.
  • Material opportunity: edges, light, texture, rhythm, or empty space suited to an artistic medium.
Preserve identity through selected relationships, not through indiscriminate detail. For portraits, protect facial identity, expression, pose, and hand count. For products and vehicles, protect proportions, signature graphics, perspective, and enough of the silhouette to remain identifiable unless the user requests distortion. Permit non-defining contours, peripheral bodywork, background objects, and expendable detail to dissolve, crop out, merge into the ground, or remain unrendered. Never add logos, people, or objects that are absent unless requested or clearly necessary for the chosen concept. Poster typography is the one default addition governed by the typography system below.
建立一份精简的内部来源映射（Source Map）：
  • **识别锚点：**最多三个必须保持可辨识的特征。
  • **空间逻辑：**用于组织画面的地平线、遮挡关系、姿态、透视、运动方向或尺度关系。
  • **能量感：**静态、抒情、图形化、紧张、嬉趣、纪念碑式、私密或动感。
  • **形状层级：**主导轮廓、次要体块以及可舍弃的细节。
  • **色彩机会：**值得保留、抑制、夸张或替换的颜色。
  • **材质机会：**适合特定艺术媒介的边缘、光线、纹理、节奏或留白。
通过经过选择的关系保留主体身份，而不是无差别地保留所有细节。对于人像，应保护面部身份、表情、姿态和手部数量。对于产品和载具，应保护比例、标志性图形、透视关系，以及足以维持辨识度的轮廓，除非用户明确要求变形。非定义性轮廓、周边车身或外壳、背景物件和可舍弃细节可以消融、被裁切、融入底色，或保持未呈现状态。除非用户提出要求，或所选概念显然需要，否则不得添加源图中不存在的标志、人物或物件。海报排版是唯一默认新增的元素，并受下述排版系统约束。

Mandatory output contract

强制输出规范

Apply these requirements to every generation unless the user explicitly overrides a specific item:
  • Generate a 3:4 vertical image. State 
    vertical 3:4 aspect ratio
    in the final generation prompt and compose for that canvas rather than merely resizing the source.
  • Reconstruct the composition according to the chosen medium. Do not automatically preserve the source crop, framing, centering, background coverage, or distribution of detail.
  • Create a poster, not merely an artistic rendering: establish a poster proposition, a thumbnail-readable silhouette, an intentional layout system, and a clear hierarchy of dominant image, supporting graphic field, negative space, and optional type zone.
  • Derive a fresh functional palette from each source image. Do not reuse a fixed house palette merely because a previous poster succeeded.
  • Establish one focal anchor and preserve only the minimum identity cues needed around it.
  • Use selective rendering. Keep one controlled area resolved; simplify, fragment, wash out, omit, or leave other regions as substrate or negative space.
  • Include at least one intentional lost-edge transition where a non-critical contour disappears into paper, pigment, ink, paint, halftone, shadow, or background color.
  • Keep the artwork edge open in at least one region. Avoid a uniformly filled rectangular image, decorative border, poster frame, or full-bleed filter effect unless explicitly requested.
  • Reserve an intentional type zone that functions as active negative space, not an empty leftover margin.
  • Add scene-specific English poster typography by default unless the user explicitly requests no text. Read and apply references/typography.md. Use one short condensed uppercase title and, when compositionally useful, zero to three smaller editorial text groups derived from the visible subject, mood, motion, place, material or poster proposition. Keep one unified typography family across a series while allowing scale, position, angle, masking and overlap to respond to each source image.
  • Never add signatures, seals, watermarks, invented logos, fake credits, dates, brands, or factual claims. If the user supplies exact wording, preserve it exactly. If exact spelling is mission-critical and the image model cannot render it reliably, generate a clean type-safe zone and recommend composing the final lettering separately.
Treat 3:4, poster proposition, intentional layout, hierarchy, medium-led recomposition, selective rendering, the default typography system, type-safe negative space, and intentional edge loss as hard acceptance criteria, not optional style suggestions.
除非用户明确覆盖某项要求，每次生成都必须遵守以下规定：
  • 生成 3:4 竖版图像。在最终生成提示词中写明 
    vertical 3:4 aspect ratio
    ，并针对该画布进行构图，而不是简单缩放源图。
  • 根据所选媒介重构构图。不要自动沿用源图的裁切、边框、居中方式、背景覆盖范围或细节分布。
  • 创作的是海报，而不只是艺术化渲染：建立明确的海报命题、在缩略图尺寸下仍可辨识的轮廓、有意图的版式系统，以及主导图像、辅助图形场、负空间和可选文字区之间的清晰层级。
  • 从每张源图中提炼新的功能性配色。不要仅因上一张海报效果良好就沿用固定的品牌配色。
  • 建立一个视觉焦点，并仅在其周围保留维持辨识度所必需的最少特征。
  • 采用选择性渲染。保留一个受控区域的完整呈现，同时简化、碎裂、淡化、省略其他区域，或将其留作底材与负空间。
  • 至少设置一处有意设计的边缘消隐过渡，使非关键轮廓消融于纸张、颜料、油墨、绘画、网点、阴影或背景色之中。
  • 至少让一个区域保持开放的画面边缘。除非用户明确要求，避免生成四边均匀填满的矩形画面、装饰性边框、海报画框或全幅滤镜效果。
  • 预留一个有意图的文字区，使其作为主动负空间发挥作用，而不是剩余的空白边缘。
  • 默认加入针对当前场景的英文海报排版，除非用户明确要求不要文字。读取并应用 references/typography.md。使用一个简短、窄体、全大写的标题；在构图适用时，可加入零至三组较小的编辑文案，内容应源自可见主体、氛围、运动、地点、材质或海报命题。系列作品应使用统一的字体家族，同时允许根据每张源图调整字号、位置、角度、遮罩和重叠关系。
  • 绝不添加签名、印章、水印、虚构标志、虚假署名、日期、品牌或事实性陈述。如果用户提供了准确文案，应逐字保留。如果任务对精确拼写有严格要求，而图像模型无法可靠呈现，则生成干净、无干扰的文字安全区，并建议在后期单独排入最终文字。
必须将 3:4 比例、海报命题、有意图的版式、层级结构、由媒介主导的重构、选择性渲染、默认排版系统、文字安全负空间和有意边缘消隐视为硬性验收标准，而非可选的风格建议。

Choose an art mode

选择艺术模式

Read references/art-modes.md after selecting a mode. Use its medium grammar, composition moves, color logic, failure checks, and prompt vocabulary.
  • Transparent watercolor: atmospheric light, fluid edges, pigment blooms, paper showing through.
  • Pop screenprint: bold shape separation, limited high-impact palette, halftone, registration play.
  • Expressive painting: directional brush energy, impasto hierarchy, emotional color construction.
  • Ink-wash: tonal economy, breathing space, calligraphic structure, controlled diffusion.
  • Relief print: carved silhouettes, forceful black-white rhythm, tactile cut marks.
  • Editorial surrealism: one legible conceptual transformation, believable visual logic, restrained symbolism.
  • Controlled hybrid: one primary mode plus one supporting process. Never blend more than two modes.
Interpret “波普” as pop-art/screenprint. If the user writes “波谱” in an art-style context, treat it as likely referring to 波普 unless frequency-spectrum imagery is explicitly requested.
选定模式后，读取 references/art-modes.md。使用其中定义的媒介语法、构图手法、色彩逻辑、失败检查项和提示词词汇。
  • **透明水彩：**氛围光、流动边缘、颜料晕染、纸面透出。
  • **波普丝网印刷：**大胆的形状分离、有限但高冲击力的配色、网点、套印趣味。
  • **表现性绘画：**方向性笔触能量、厚涂层级、情绪化色彩构建。
  • **水墨：**克制的明度、呼吸空间、书法性结构、可控的扩散效果。
  • **凸版印刷：**雕刻式轮廓、有力的黑白节奏、可感知的刻痕。
  • **编辑超现实主义：**一个清晰可辨的概念性变形、可信的视觉逻辑、克制的象征手法。
  • **可控混合媒介：**一种主要模式加一种辅助工艺。不得混合超过两种模式。
将“波普”理解为波普艺术或丝网印刷。如果用户在艺术风格语境中写的是“波谱”，通常应按“波普”处理，除非用户明确要求频谱图像。

Build the transformation brief

构建转换简报

Resolve these decisions before prompting:
  1. Output: vertical 3:4 aspect ratio, intended poster context, viewing distance, and whether the user has overridden the default English typography.
  2. Preservation: exact identity anchors and spatial logic to retain.
  3. Poster proposition: one short visual statement or emotional promise; never a vague style label.
  4. Layout system: choose one system from references/poster-systems.md and map the focal image, supporting field, type zone, and eye path.
  5. Transformation: what becomes simplified, exaggerated, dissolved, repeated, displaced, or recolored.
  6. Medium behavior: physical marks, edge behavior, layering, pigment/ink/paint interaction, and substrate.
  7. Composition: focal point, eye path, dominant and subordinate masses, source elements to move/crop/scale/omit, and active negative space.
  8. Coverage map: resolved focal region, simplified support region, untouched or substrate-led region, and the exact contours selected for lost-edge treatment.
  9. Graphic devices: select no more than two source-derived devices such as crop bar, repeated contour, enlarged detail, color block, halftone field, registration echo, or directional rule. Do not add arbitrary decoration.
  10. Palette: extract the source's dominant color, identity accent, environmental color, and substrate opportunity, then build a source-specific 3–6 color functional palette. Preserve or strategically intensify recognition-bearing colors such as a lamp, garment, brake caliper, painted graphic, sign, material finish, or atmospheric light. Suppress muddy or expendable colors. Permit hue shifts and stronger value, saturation, or temperature contrast when they clarify hierarchy, but never impose a fixed palette from an earlier poster.
  11. Typography: read references/typography.md, derive one short English title and zero to three optional supporting text groups from the visible image, assign exact wording and hierarchy, choose a source-responsive title direction, and define any controlled foreground-background interaction between the title and subject. Keep the locked typography family consistent across a series while varying placement according to each composition.
  12. Exclusions: generic filter look, centered product-shot layout, equal detail, edge-to-edge uniform treatment, stray or misspelled text, watermark, unwanted frames, fake mockups, extra anatomy, and style-specific failure modes.
Use medium-native verbs. Watercolor should bleed, pool, granulate, glaze, lift, and reserve paper; screenprint should separate, overprint, misregister, repeat, and halftone. Avoid relying on artist names. Describe visual properties and processes instead.
编写提示词前，先解决以下决策：
  1. **输出：**竖版 3:4 比例、目标海报应用场景、观看距离，以及用户是否覆盖了默认英文排版要求。
  2. **保留：**需要准确保留的识别锚点和空间逻辑。
  3. **海报命题：**一句简短的视觉陈述或情绪承诺，不得使用模糊的风格标签。
  4. **版式系统：**从 references/poster-systems.md 中选择一种系统，并映射焦点图像、辅助场、文字区和视线路径。
  5. **转换：**明确哪些元素将被简化、夸张、消融、重复、移位或重新着色。
  6. **媒介表现：**物理笔触、边缘行为、图层关系、颜料或油墨的相互作用，以及底材。
  7. **构图：**焦点、视线路径、主导与从属体块、需要移动、裁切、缩放或省略的源图元素，以及主动负空间。
  8. **覆盖映射：**完整呈现的焦点区域、简化的辅助区域、未处理或由底材主导的区域，以及被选作边缘消隐处理的具体轮廓。
  9. **图形手法：**最多选择两种源自源图的图形手法，例如裁切条、重复轮廓、放大细节、色块、网点场、套印回声或方向性线条。不得添加任意装饰。
  10. **配色：**提取源图的主色、识别性强调色、环境色和底材表现空间，然后构建针对当前源图的 3–6 色功能性配色。保留或策略性强化具有识别意义的颜色，例如灯光、服装、制动卡钳、涂装图形、标识、材质表面效果或氛围光。抑制浑浊或无关键的颜色。允许改变色相，并增强明度、饱和度或冷暖对比以强化层级，但不得强加来自上一张海报的固定配色。
  11. **排版：**读取 references/typography.md，根据可见图像推导一个简短英文标题和零至三组可选辅助文案，明确其具体措辞与层级，选择与源图呼应的标题方向，并定义标题与主体之间受控的前景与背景关系。在整个系列中保持锁定的字体家族一致，同时根据每张构图调整位置。
  12. **排除项：**通用滤镜观感、居中的产品图布局、均等细节、边缘到边缘的均匀处理、错误或多余文字、水印、不需要的边框、虚假样机、错误人体结构，以及特定媒介的常见失败效果。
使用符合媒介特性的动词。水彩应表现出渗化、积聚、颗粒、罩染、提色和留出纸白；丝网印刷应表现出分色、叠印、套印偏移、重复和网点。避免依赖艺术家姓名。应描述视觉属性和工艺过程，而不是直接模仿具体艺术家。

Composition rules

构图规则

  • Read references/poster-systems.md before prompting. Choose one layout system; do not average several systems together.
  • Establish one dominant visual idea that is readable at thumbnail size.
  • Build at least three clearly different spatial zones: dominant image, supporting graphic/material field, and active negative/type-safe space. These may overlap but must remain hierarchically legible.
  • Create scale contrast. The focal anchor should be decisively larger, sharper, darker, brighter, or more saturated than supporting elements.
  • Integrate the title with source geometry. Permit controlled masking, overlap, vertical stacking, edge cropping or one source-derived diagonal when it strengthens depth and hierarchy. Keep the title understandable at thumbnail size and never hide several essential letters.
  • Recompose for the 3:4 canvas using crop, scale shift, asymmetric placement, overlap, repetition, compression, or expanded negative space when appropriate to the medium.
  • Let one area carry detail and let other areas simplify, disappear, or breathe. Do not distribute finish evenly.
  • Change at least two of these dimensions so the result is a true reinterpretation: shape, color, material, scale, space, repetition, or narrative meaning.
  • Connect artistic effects to source geometry. Do not scatter arbitrary splashes, dots, collage scraps, or symbols as decoration.
  • Connect color to source evidence. Let the image determine the palette family, then art-direct it for contrast and hierarchy; do not mechanically apply habitual red-blue-cream, black-yellow, or any other preset combination.
  • Keep medium behavior spatially selective: sharp versus lost edges, dense versus open marks, opaque versus transparent layers.
  • Use lost edges structurally: dissolve contours on the shadow side, motion side, atmospheric side, or where the subject meets a similar value/color. Keep defining edges near the focal anchor legible.
  • Allow peripheral forms to terminate before the canvas edge or merge into exposed substrate. The artwork need not depict every source element.
  • Preserve user-requested wording exactly. Otherwise derive restrained scene-specific English copy and apply the locked typography system. Use one dominant title and no more than three supporting text groups. State every permitted word exactly in the generation prompt and forbid all other lettering. Prefer fewer accurate words over corrupted subtitles, fake metadata or decorative filler text.
  • Avoid the common pseudo-poster solution of placing an unchanged centered subject over a textured background. At least two structural relationships from the source must change: crop/scale, placement, overlap, repetition, figure-ground, or negative-space distribution.
编写提示词前，读取 references/poster-systems.md。选择一种版式系统，不要将多种系统平均混合。
  • 建立一个主导视觉概念，使其在缩略图尺寸下仍能立即被理解。
  • 建立至少三个明显不同的空间区域：主导图像、辅助图形或材质场，以及主动负空间或文字安全区。这些区域可以重叠，但必须保持清晰的层级关系。
  • 建立尺度对比。焦点锚点应在尺寸、清晰度、明暗、亮度、饱和度或细节密度上明显强于辅助元素。
  • 将标题与源图几何关系整合。可使用受控遮罩、重叠、纵向堆叠、边缘裁切，或一条源自源图的对角线来增强深度和层级。标题在缩略图尺寸下仍应易于理解，且不得遮住多个关键字母。
  • 根据需要运用裁切、比例偏移、不对称布局、重叠、重复、压缩或扩展负空间，为 3:4 画布重新构图。
  • 让一个区域承载细节，同时让其他区域简化、消失或留出呼吸空间。不要平均分配完成度。
  • 至少改变形状、色彩、材质、尺度、空间、重复或叙事意义中的两个维度，确保结果是真正的重新诠释。
  • 将艺术效果与源图几何关系建立联系。不要把任意的泼溅、圆点、拼贴碎片或符号当作装饰散布在画面中。
  • 让色彩来自源图依据。由图像决定配色家族，再通过艺术指导强化对比和层级；不要机械套用常见的红蓝米、黑黄或任何其他预设组合。
  • 在空间上有选择地体现媒介表现：锐利边缘与消隐边缘、密集笔触与开放笔触、不透明图层与透明图层应形成对比。
  • 在结构上使用边缘消隐：在阴影侧、运动侧、大气侧，或主体与相近明度或颜色的交界处消解轮廓。焦点锚点附近的定义性边缘必须保持清晰。
  • 允许周边形体在到达画布边缘前终止，或融入裸露底材。作品不必呈现源图中的每一个元素。
  • 准确保留用户指定的文案。其他情况下，应推导克制且与场景相关的英文文案，并应用锁定的排版系统。使用一个主导标题和不超过三组辅助文案。在生成提示词中逐字列出所有允许使用的文字，并禁止出现其他文字。宁可使用更少且准确的文字，也不要生成拼写错误的字幕、伪造的元数据或装饰性填充文字。
  • 避免伪海报的常见做法，即把未经改变的主体居中放在纹理背景上。至少改变源图中的两种结构关系：裁切或缩放、位置、重叠、重复、图底关系或负空间分布。

Hybrid rules

混合媒介规则

Use a hybrid only when it creates a clear division of labor:
  • primary mode controls composition, shapes, and dominant material;
  • supporting mode contributes one limited behavior such as halftone, ink contour, collage displacement, or metallic accent;
  • assign each process to specific regions;
  • keep one palette and one focal hierarchy across both processes.
Reject combinations that merely list styles. If removing the supporting mode does not change meaning, depth, rhythm, or focus, remove it.
只有当混合媒介能够形成明确的职责分工时，才使用混合方式：
  • 主要模式控制构图、形状和主导材质；
  • 辅助模式只贡献一种有限表现，例如网点、墨线轮廓、拼贴位移或金属质感点缀；
  • 为每种工艺指定具体应用区域；
  • 两种工艺共享同一套配色和焦点层级。
拒绝仅罗列风格的组合。如果移除辅助模式不会改变作品的含义、深度、节奏或焦点，就应将其删除。

Prompt structure

提示词结构

Write the final image-generation prompt in four compact sections:
  1. Source fidelity: identify the supplied image, identity anchors, and spatial logic that must survive.
  2. Poster direction: state 
    vertical 3:4 aspect ratio
    , the poster proposition, chosen layout system, thumbnail silhouette, focal anchor, scale contrast, eye path, type zone, and source elements to crop, move, enlarge, repeat, suppress, or omit.
  3. Material, graphic, and type system: specify physical medium behavior, substrate, palette, selective coverage map, no more than two source-derived graphic devices, exposed negative space and at least one intentional lost-edge transition. State the exact title and every permitted micro-copy group, define their hierarchy, direction and approximate positions, specify any controlled title-subject masking or overlap, apply the unified typography rules from references/typography.md, and forbid all other lettering.
  4. Guardrails: state what must not change and forbid unchanged centered-photo layouts, uniform full-image treatment, arbitrary decoration, misspelled or additional text, frames, and relevant mode-specific failures.
Describe the desired image as a finished artwork, not as instructions to add a filter. Do not mention this skill or internal analysis in the prompt.
最终的图像生成提示词应包含四个紧凑部分：
  1. **源图保真度：**说明所提供的图像、必须保留的识别锚点，以及需要延续的空间逻辑。
  2. **海报方向：**写明 
    vertical 3:4 aspect ratio
    、海报命题、所选版式系统、缩略图尺寸下可辨识的轮廓、焦点锚点、尺度对比、视线路径、文字区，以及需要裁切、移动、放大、重复、抑制或省略的源图元素。
  3. **材质、图形与文字系统：**指定物理媒介表现、底材、配色、选择性覆盖映射、不超过两种源自源图的图形手法、裸露负空间，以及至少一处有意设计的边缘消隐过渡。明确写出准确标题和每一组允许使用的短文案，定义其层级、方向和大致位置，指定标题与主体之间任何受控的遮罩或重叠关系，应用 references/typography.md 中的统一排版规则，并禁止出现其他文字。
  4. **防护要求：**说明不得改变的内容，并禁止使用未经改变的居中照片布局、全图均匀处理、任意装饰、拼写错误或额外文字、边框，以及与当前模式相关的失败效果。
将目标图像描述为一件完成的作品，而不是一项“添加滤镜”的指令。提示词中不要提及本 skill 或内部分析。

Safety and privacy

安全与隐私

  • Send only the necessary reference image and final prompt to the image-generation service.
  • Do not browse for, upload, save, or share the user's source image elsewhere.
  • Do not save source or generated images into project files unless requested.
  • Do not imitate a living artist's exact style. Translate such requests into high-level traits, medium, era, palette, and composition.
  • Avoid inventing sensitive personal attributes or changing identity unless the user asks.
  • 仅向图像生成服务发送必要的参考图像和最终提示词。
  • 不要在其他地方浏览、上传、保存或分享用户的源图。
  • 除非用户提出要求，不要将源图或生成图像保存到项目文件中。
  • 不要模仿在世艺术家的具体风格。将此类请求转化为高层次特征、媒介、时代、配色和构图。
  • 不要虚构敏感的个人属性，也不要在用户未要求时改变人物身份。

Quality gate

质量关卡

Before accepting the result, check:
  • the result is composed and generated as a vertical 3:4 image;
  • the result reads immediately as an authored art poster rather than a filtered photograph or framed illustration;
  • one clear poster proposition and one layout system organize the image;
  • dominant image, supporting field, and active negative/type-safe space form a legible hierarchy;
  • scale contrast and eye path remain clear at thumbnail size;
  • the selected identity anchors remain legible;
  • the composition has been meaningfully redesigned for the selected medium rather than copied from the source by default;
  • the medium behaves physically rather than as a uniform digital overlay;
  • the image has a clear focal hierarchy, selective rendering, and intentional empty or substrate-led areas;
  • at least one non-critical contour uses a convincing lost edge while defining focal edges remain readable;
  • the canvas is not uniformly filled or boxed by an unwanted frame;
  • the palette is coherent and structurally useful;
  • the palette is recognizably derived from the current source image, protects meaningful identity accents, and is not copied from a previous poster by default;
  • the transformation feels source-specific rather than template-like;
  • graphic devices derive from source geometry and do not behave as arbitrary decoration;
  • the English title is concise, legible, correctly spelled, source-relevant and rendered in the locked condensed uppercase editorial form unless the user requested another treatment or no text;
  • supporting copy uses the same typography family, contains no more than three restrained text groups and remains subordinate to the focal image and title;
  • title direction and any image-title overlap respond to visible source geometry, strengthen hierarchy and preserve thumbnail readability;
  • no additional lettering, watermark, mockup frame, or unwanted object appears;
  • hands, faces, product geometry, and vehicle proportions remain credible where relevant.
If the first result misses a defining requirement, revise the prompt with a precise correction and regenerate once. Do not repeatedly generate minor variants unless the user requests exploration.
接受结果前，检查以下事项：
  • 结果是经过构图设计并生成的 3:4 竖版图像；
  • 结果应立即被识别为一件经过创作的艺术海报，而不是经过滤镜处理的照片或带框插画；
  • 一个明确的海报命题和一种版式系统组织整幅图像；
  • 主导图像、辅助场和主动负空间或文字安全区形成可辨识的层级；
  • 尺度对比和视线路径在缩略图尺寸下仍然清晰；
  • 所选识别锚点保持可辨识；
  • 构图已针对所选媒介进行了实质性重新设计，而不是默认沿用源图；
  • 媒介呈现出物理过程感，而不是均匀的数字叠加效果；
  • 图像具有清晰的焦点层级、选择性渲染，以及有意留空或由底材主导的区域；
  • 至少一个非关键轮廓采用了令人信服的边缘消隐，同时焦点的定义性边缘仍然清晰；
  • 画布没有均匀填满，也没有被不需要的边框包围；
  • 配色连贯且在结构上发挥明确作用；
  • 配色可辨识地源自当前源图，保护了具有意义的识别性强调色，并且没有默认沿用上一张海报的配色；
  • 转换过程具有源图专属特征，而不是模板化处理；
  • 图形手法源自源图几何关系，而不是充当任意装饰；
  • 英文标题简洁、清晰、拼写正确、与源图相关，并采用锁定的窄体全大写编辑风格，除非用户要求采用其他处理方式或不添加文字；
  • 辅助文案使用相同字体家族，不超过三组克制的文字组，并从属于焦点图像和标题；
  • 标题方向及任何图像与标题的重叠关系呼应可见的源图几何关系，在强化层级的同时保持缩略图可读性；
  • 没有额外文字、水印、样机边框或不需要的物件；
  • 在相关场景中，手部、面部、产品几何结构和载具比例保持可信。
如果首次结果未达到某项决定性要求，应使用精确的修正说明修订提示词，并重新生成一次。除非用户要求进行探索，不要反复生成细微变体。