# 模仿生图skill > 创建和优化中文壁纸类图像生成提示词。用于分析参考图、提炼壁纸风格、编写手机壁纸、头像、背景、可爱动物图标、旅行风景、治愈插画、清新生活方式视觉、复古海报壁纸等提示词；当用户明确要求生成图片时，先给出精炼的七大板块提示词卡片，再进行图片生成。默认输出分析、最终提示词和负面提示词；参考图只作为风格和氛围来源，生成时默认做原创变化，避免一比一复刻。 - Skill: `chenkuop9909-creator/skill` (Agent Skill) - Install (CLI): `npx skillmds@latest add chenkuop9909-creator/skill` - Raw SKILL.md: https://api.skillmd.com/api/skills/chenkuop9909-creator/skill/raw - Safety review: pending (external: skill-scanner PASS, skillspector PASS) - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Coding & Dev Tools - Author: chenkuop9909-creator (https://skillmd.com/u/chenkuop9909-creator) - Updated: 2026-09-22 - Page: https://skillmd.com/skills/chenkuop9909-creator/skill --- # Wallpaper Prompt Designer ## Overview Use this skill to analyze a wallpaper reference image or a short wallpaper theme and turn it into a high-quality Chinese prompt. The default output is: 1. multi-dimensional analysis 2. final prompt 3. negative prompt Do not generate an image unless the user explicitly asks to generate or edit an image. This skill is for prompt design, prompt standardization, and reusable image-generation prompt refinement. When the user explicitly asks to generate an image from a reference, treat the reference as style DNA rather than a layout to copy. Preserve the appealing mood, style, color memory, and usability, but create a same-series new image with clear differences. ## Core Rule: 同源不同图 For reference-image generations, default to `同源不同图`: - Keep: overall style, mood, color personality, lighting logic, texture, and wallpaper usability. - Change: at least three of subject placement, camera angle, foreground objects, background setting, time of day, color accents, props, architecture, species/object type, or environmental details. - Avoid: same composition, same object positions, same crop, same text/logo layout, same foreground-to-background arrangement, or near-duplicate framing. - Add: one or two original visual ideas that fit the category, such as a different road curve, balcony, window light, flower slope, small boat, clouds, sign shape, seasonal detail, or alternate viewpoint. - Phrase prompts with `同类质感 / 同类氛围 / 同系列原创画面`, not `一模一样 / 复刻 / 完全相同`. If the user explicitly says they want the image to be very close to the reference, still avoid copying watermarks, platform UI, real brands, and exact protected logos. ## Workflow ### 1. Identify the input type Determine whether the user provided: - **reference image**: analyze visible style, composition, colors, lighting, subject, texture, and mood. - **theme only**: infer a suitable wallpaper direction from the theme. - **existing prompt**: rewrite it into a clearer wallpaper prompt. - **multiple references**: extract shared traits and note differences before writing a prompt. - **generation request**: if the user says to generate, create the image prompt and provide a compact reusable prompt card before calling image generation. ### 2. Decide the wallpaper category Classify the request into the closest category. Use more than one only when needed. - landscape / travel wallpaper - city / street / architecture wallpaper - cute animal avatar or icon - dreamy hand-drawn illustration - fresh lifestyle background - minimal clean background - retro poster-style wallpaper - healing text-lockscreen background - digital planner / desk / iPad-style background ### 3. Analyze before prompting For reference images, always analyze these dimensions: - **画风**: photography, film look, retro poster, flat illustration, oil pastel, thick paint, soft hand-drawn, semi-realistic illustration, etc. - **主体**: main object, scale, pose, expression, foreground object, repeated motif. - **构图**: vertical framing, centered subject, three-part structure, leading lines, aerial view, large negative space, foreground-middle-background depth. - **色彩**: dominant colors, accent colors, saturation, warm/cool balance, color personality. - **光影**: morning light, golden hour, soft backlight, strong sunlight, haze, sparkle, shallow depth of field. - **氛围**: healing, fresh, cute, happy, relaxed, dreamy, nostalgic, cinematic, quiet, vacation-like. - **质感**: film grain, soft focus, air perspective, hand-painted texture, acrylic brush, oil pastel, paper texture, glossy light. - **适用性**: whether it works as lockscreen, home screen, avatar, chat background, or collection cover. Keep analysis concise but concrete. Avoid vague praise such as “很好看” unless paired with a reason. ### 4. Build the final prompt The final prompt must be in Chinese by default. It should be a single coherent paragraph unless the user asks for multiple variants. Include these elements when relevant: - aspect and usage: usually `竖版 9:16 手机壁纸` or `圆形构图头像`. - subject: what the image contains. - composition: how the subject is placed and how the eye moves. - color system: dominant colors and accent colors. - light and atmosphere: mood and lighting. - style and texture: photography, illustration, poster, hand-drawn, film, etc. - detail density: rich but not cluttered. - usability: clean, suitable for wallpaper, no distracting overlays. - originality direction: what is intentionally changed from the reference. Do not include platform words such as 小红书头像、抖音头像、微信头像 unless the user explicitly asks. Describe the image itself instead: `圆形构图`、`竖版手机壁纸`、`头像插画`、`壁纸背景`. ### 5. Add a negative prompt Always include a negative prompt unless the user asks for only the final prompt. Default negative prompt themes: - text and overlays: 文字、字母、数字、时间、日期、logo、水印、平台标识、UI界面、手机边框 - quality issues: 模糊、低清晰度、噪点、变形、比例错误、过度锐化、脏污画面 - style mismatch: 写实风 where illustration is requested, 3D风格 where flat illustration is requested, 卡通过头 where realistic scenery is requested - clutter: 杂乱背景、过多装饰、元素堆砌、强烈刺眼颜色 - unwanted subjects: 人物特写、多人、车辆 if irrelevant, 复杂招牌文字 if not requested Customize the negative prompt for the category. For scenery, exclude text, UI, watermarks, people, harsh colors, and clutter. For avatars, exclude realistic animal photos, complex backgrounds, multiple animals, text, and watermarks. If the user wants graphical logos or signboards, allow original non-branded graphic logos, abstract icons, badges, color blocks, and unreadable decorative sign shapes. Still exclude real brands, readable text, platform marks, watermarks, and copied trademarks. ### 6. Generation prompt card When the user explicitly asks to generate an image, send a compact prompt card before image generation. Use this format: ```markdown ## 生成提示词精简版 **画风：** ... **主体：** ... **构图：** ... **色彩：** ... **光影：** ... **氛围：** ... **质感：** ... **原创变化：** ... ``` Rules for the compact prompt card: - Keep each block to one short sentence or phrase; avoid long paragraphs. - Make it more refined than the full analysis, not just a copy of the analysis. - Include `原创变化` whenever generating from a reference image. - In `原创变化`, name at least three concrete differences from the reference. - Include whether logos/signboards should exist, and specify `原创图形 logo / 不可读装饰性招牌` when logos are desired without real brands. - If the environment requires image generation as the last action in the turn, output this card immediately before calling image generation. ## Category Guidelines ### Landscape and travel wallpaper Prioritize vertical composition, atmosphere, color memory, and real usability. Include: - clear foreground / middle ground / background - sky or clean upper area for lockscreen usability - light direction and time of day - texture such as ocean sparkle, mountain haze, film grain, or air perspective Originality ideas: - change the viewpoint, coastline curve, foreground flowers, weather, season, house style, road/path shape, or distant object - keep the emotional color but avoid the same crop and object positions Avoid: - visible watermarks, UI, platform text, screenshot artifacts - over-crowded foreground - random people unless requested ### Cute animal avatar or icon Prioritize a memorable subject, centered composition, simple expression, and clean background. Include: - round or centered composition - low-saturation background - cute expression, blush, soft body shape - clean line art or soft hand-drawn texture - a few small detail elements only when useful Originality ideas: - change pose, accessory, background motif, color accent, or facial expression - preserve cuteness and texture, not the exact animal layout Avoid: - platform labels - complex scenery - realistic pet photography unless requested - too many animals or accessories ### Dreamy hand-drawn illustration Prioritize tactile texture, dreamlike details, and emotional softness. Include: - oil pastel, acrylic, thick paint, crayon, soft brush, paper texture - flowers, butterflies, stars, sparkles, soft glow when relevant - naive / childlike linework if reference image suggests it Avoid: - overly polished vector look if the user wants hand-drawn warmth - hard shadows, harsh neon colors, rigid geometry ### Fresh lifestyle background Prioritize clean daily-use feeling and soft color. Include: - cream white, pale blue, mint, pale pink, beige - desk, coffee, flowers, iPad, notebook, window light, small objects when relevant - open negative space and balanced arrangement Avoid: - messy desk, too many props, heavy text, harsh contrast ### Retro poster-style wallpaper Prioritize bold color blocks, sunlit atmosphere, grain, and graphic structure. Include: - palm trees, signage shapes, abstract logo badges, clouds, coastline, vintage street when relevant - saturated but controlled blue, cream yellow, orange-brown, green - poster-like edges, slight grain, clean silhouette Originality ideas: - change the street angle, sign stack, car type, palm tree placement, building style, cloud shape, or distant scenery - use original graphic logos instead of real brands or copied lettering Avoid: - real brand names, legible random text, watermark, excessive typography ## Output Format Use this structure by default: ```markdown ## 分析 **画风：** ... **主体：** ... **构图：** ... **色彩：** ... **光影：** ... **氛围：** ... **质感：** ... **适用性：** ... ## 最终提示词 [one complete chinese prompt] ## 负面提示词 [negative prompt] ``` If the user explicitly asks to “只输出提示词”, output only the final prompt blocks without analysis. If the user asks to generate an image, output `## 生成提示词精简版` first, then generate the image. ## Quality Rules - Keep prompts specific and visual. Avoid empty adjectives unless attached to concrete visual choices. - Do not mention unavailable tools, generation platforms, or platform-specific account labels. - Do not promise exact recreation of copyrighted or branded images. Extract style and composition instead. - Prefer “同类质感 / 同类构图 / 类似氛围 / 同系列原创画面” over copying exact logos, names, watermarks, or layouts. - For user-provided reference images with watermarks or UI screenshots, instruct the prompt to generate a clean original image with no watermarks, text, UI, or screenshot borders. - When the reference image contains visible text as part of a scene, avoid asking the model to reproduce exact text unless the user explicitly needs it. - When the user asks for logos on signs, use original abstract graphic logos or unreadable decorative sign shapes; do not request real brands or readable copied lettering. - For follow-up iterations, preserve what the user liked, change only the requested parts, and add one or two original visual ideas so the result is not a near-duplicate.