image-prompt
Compare original and translation side by side
🇺🇸
Original
English🇨🇳
Translation
ChineseImage Prompt (brief · craft · router)
Image Prompt (brief · craft · router)
The foundation of the visual cluster. When a post needs an image, this skill decides what image
(the brief), describes it with model-agnostic craft, and routes to the right tool — then hands
off. It sits above the tool-specific mini-skills (, , ,
), which own their model's specific tricks.
nano-bananaideogramfluxveo-3Think of it as the visual equivalent of /: the shared decision layer
the tool skills assume you've already worked through.
brand-profilecontent-pillars视觉技能集群的基础。 当帖子需要图片时,这个 skill 决定需要哪类图片
(brief),用与模型无关的方式描述它,并路由到正确的工具——然后移交给
特定工具的 mini-skill(、、、
),由它们负责各自模型的特定技巧。
nano-bananaideogramfluxveo-3可以将它视为 / 的视觉对应物:这是一个位于特定工具
skill 之上的共享决策层,后续工具 skill 默认你已经完成了这一步。
brand-profilecontent-pillarsStep 0 — Gate: should this even be AI-generated?
第 0 步——门槛:这是否应该由 AI 生成?
AI generation is one option, not the default. A real photo (your actual product/space/team), a
screenshot, UGC, or a chart often wins on authenticity and clarity — and AI must never
fabricate your real people/place. Reach for generation when you need something that doesn't exist, must
match a brand look, needs legible in-image text, or a consistent series — and no real asset fits. If a
real asset is better, say so. (See .)
references/the-image-brief.mdAI 生成只是一种选择,并非默认方案。真实照片(你的实际产品、空间或团队)、
截图、UGC 或图表,在真实性和清晰度方面往往更胜一筹——AI 绝不应
虚构你的真实人物或地点。只有当你需要现实中不存在的素材、必须匹配某种品牌风格、
需要图片中包含清晰可辨的文字,或需要生成一套风格一致的系列,并且没有合适的
真实素材可用时,才应考虑生成。如果真实素材更好,就明确说明。
(参见 。)
references/the-image-brief.mdStep 1 — Write the brief
第 1 步——编写 brief
Read , then answer four things: job/purpose (hero / explainer / quote graphic /
product / lifestyle / thumbnail / background / ad), the one thing it must convey, format/aspect
ratio for the destination, and the brand look. Output a one-line brief. See
.
brand-profile.mdreferences/the-image-brief.md读取 ,然后回答四个问题:用途/目的(主视觉 / 解释图 / 金句图 /
产品图 / 生活方式图 / 缩略图 / 背景图 / 广告图)、必须传达的唯一核心信息、
发布渠道所需的格式/宽高比,以及品牌视觉风格。用一行输出 brief。参见
。
brand-profile.mdreferences/the-image-brief.mdStep 2 — Describe it (model-agnostic)
第 2 步——描述图像(与模型无关)
Turn the brief into a specific, natural-language prompt — subject · composition · lighting ·
colour · style · mood · detail · constraints — and drop quality-token spam ("4k, masterpiece").
Specificity beats length; design for any text overlay (negative space, contrast). See
.
references/prompt-anatomy.md将 brief 转化为具体、自然语言的 prompt——主体 · 构图 · 光线 ·
色彩 · 风格 · 氛围 · 细节 · 约束——并删掉无意义的 quality token 堆砌
(“4k、masterpiece”)。具体性比长度更重要;还要为任何文字叠加层进行设计
(负空间、对比度)。参见 。
references/prompt-anatomy.mdStep 3 — Route to the tool
第 3 步——路由至工具
Pick by the image's dominant requirement:
- Text / typographic → or
ideogram(both strong — honest pick by tooling).nano-banana - Photoreal portrait/product → or
nano-banana(Midjourney external). Surreal/painterly → Midjourney.flux - Editing ("change one thing, keep the rest") → (conversational) or
nano-banana(Kontext).flux - Consistent character/product set → (multi-image references) or
nano-banana(multi-reference).flux - Search-grounded infographics → . Brand-exact hex / open weights / license control →
nano-banana.flux - Vector/SVG → Recraft. Video → .
veo-3
When two fit, say both work and pick by the deciding factor — don't fake a winner. See
.
references/choosing-the-tool.md根据图片的主导需求进行选择:
- 文字/排版 → 或
ideogram(两者都很强——根据工具支持情况如实选择)。nano-banana - 照片写实的人像/产品 → 或
nano-banana(Midjourney 为外部工具)。超现实/绘画风格 → Midjourney。flux - 编辑(“只改一处,其余保持不变”)→ (对话式)或
nano-banana(Kontext)。flux - 一致的角色/产品系列 → (多图参考)或
nano-banana(多参考图)。flux - 基于搜索结果的信息图 → 。品牌色值必须精确匹配 / open weights / license control →
nano-banana。flux - 矢量图/SVG → Recraft。视频 → 。
veo-3
当两个工具都适合时,说明两者都可用,并根据决定性因素进行选择——不要伪造
唯一的赢家。参见 。
references/choosing-the-tool.mdStep 4 — Hand off, then ship
第 4 步——交接,然后发布
Hand off to the tool's mini-skill for model-specific craft (and for
the API). After generation: verify in-image text/data, disclose AI per platform/region, then
upload to WoopSocial Media → attach via . WoopSocial publishes; it doesn't
generate.
tools/integrations/<tool>.mdscheduling-and-queue移交给相应工具的 mini-skill,应用针对该模型的具体技巧(API 则参见
)。生成后:核验图片中的文字和数据,按照
平台/地区要求披露 AI,然后上传至 WoopSocial Media → 通过
附加到帖子。WoopSocial 负责发布;它不负责生成。
tools/integrations/<tool>.mdscheduling-and-queueCross-cutting (always)
贯穿始终的要求
- Disclose AI per platform/region (Meta "Made with AI", EU AI Act). Watermark behaviour varies by tool (SynthID on Google tools; none enforced on Ideogram) — don't rely on it for disclosure.
- No real people / no IP — no real identifiable individuals, copyrighted characters, or trademarked logos; original directions only.
- Verify in-image text and data before publishing.
- Accessibility — contrast + legibility on mobile; leave room for overlays.
- 按照平台/地区要求披露 AI(Meta “Made with AI”、EU AI Act)。不同工具的水印行为 各不相同(Google 工具使用 SynthID;Ideogram 不强制添加水印)——不要依赖水印完成披露。
- 不使用真人/IP——不得使用真实可识别人物、受版权保护的角色或注册商标 Logo; 只采用原创方向。
- 发布前核验图片中的文字和数据。
- 无障碍访问——确保移动端的对比度和可读性;为叠加内容预留空间。
Quality bar — self-check
质量标准——自检
- Did I gate (real asset vs AI) before generating?
- Did I write a brief (job + the one thing + format + brand) and a specific, natural-language prompt (no token spam)?
- Did I route by the dominant requirement with an honest which-tool read, and hand off?
- Did I design for overlays/accessibility, and cover disclosure, IP, and verification?
- Did I keep WoopSocial honest (publishes via Media; doesn't generate)?
- 我是否在生成前完成了 gate(真实素材与 AI 的选择判断)?
- 我是否编写了 brief(用途 + 唯一核心信息 + 格式 + 品牌)和具体的自然语言 prompt(没有无意义的 token 堆砌)?
- 我是否根据主导需求完成了路由,如实判断工具,并完成交接?
- 我是否为叠加文字/无障碍访问进行了设计,并处理了披露、IP 和核验?
- 我是否确保 WoopSocial 的职责清晰(通过 Media 发布;不生成图像)?
Edge cases & pushback
边界情况与异议处理
- "Just write a cool image prompt" → build the brief first; purpose before prompt.
- "4k masterpiece ultra-detailed" → rewrite into a specific scene; drop quality tokens.
- "Generate my real cafe/staff" → use a real photo; don't fabricate real people/place.
- "Base it on [studio]'s characters" / a real person → refuse IP/real people; original direction.
- "Which model — A or B?" → honest, tooling-based answer; both can be right.
- "Generate it in WoopSocial" → WoopSocial doesn't generate; route to a tool, then upload + attach.
- “直接写一个很酷的图像 prompt” → 先编写 brief;先明确目的,再写 prompt。
- “4k masterpiece ultra-detailed” → 改写为具体场景;删掉 quality token。
- “生成我的真实咖啡馆/员工” → 使用真实照片;不要虚构真实人物或地点。
- “以 [studio] 的角色为基础”或使用真人 → 拒绝使用受版权保护的 IP 或真实人物;转为原创方向。
- “该用模型 A 还是 B?” → 根据工具支持情况如实回答;两者都可能合适。
- “直接在 WoopSocial 里生成” → WoopSocial 不生成图像;先路由至合适的工具,再上传并附加素材。
Related
相关内容
- ,
nano-banana,ideogram,flux— the tool-specific mini-skills this routes/hands off to.veo-3 - — the connection/API for each tool;
tools/integrations/*— the visual brand.brand-profile - ,
carousel-writer,reels-script/caption-writer— consumers (image-prompt-pack / overlay copy).hook-writer - — upload the asset to Media and attach to a post.
scheduling-and-queue
- 、
nano-banana、ideogram、flux——本 skill 用于路由并交接的特定工具 mini-skill。veo-3 - ——各工具的连接/API;
tools/integrations/*——视觉品牌。brand-profile - 、
carousel-writer、reels-script/caption-writer——下游调用方 (image-prompt-pack / 叠加文案)。hook-writer - ——将素材上传至 Media 并附加到帖子。
scheduling-and-queue
References
参考资料
- — the "should this be AI?" gate and the four-question brief.
references/the-image-brief.md - — the universal, model-agnostic anatomy of a strong prompt.
references/prompt-anatomy.md - — the router/decision tree and honest which-tool calls.
references/choosing-the-tool.md - — end-to-end flows, weak→strong, the don't-generate gate, and honest scope.
references/examples.md
- ——“这是否应该由 AI 生成?”的 gate 和四问 brief。
references/the-image-brief.md - ——高质量 prompt 的通用、与模型无关的结构。
references/prompt-anatomy.md - ——工具路由器/决策树,以及对工具选择如实判断的方法。
references/choosing-the-tool.md - ——端到端流程、从弱到强的改写示例、不要生成的 gate,以及如实界定能力范围的方法。
references/examples.md