菜单
Use when crafting fantasy cover art, fantasy cover concepts, book-cover prompts, fantasy image-generation workflows, or pixel-review refinements for fantasy fiction.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Use when drafting, revising, line editing, dialogue, POV voice, prose texture, cadence, human-feeling prose, and prose audit.
Use when any serial work needs lifecycle, state, spoilers, ledgers, handoff, logs, or commit discipline. It owns planning, grounding, and updates; story-engine owns story design and developmental review; authors-voice owns composition and prose audit.
Use when premise, arc, chapter, scene, pacing, character pressure, reader pull/anticipation, promise movement, structural variation, or developmental review work needs to be designed.
| name | crafting-fantasy-cover-art |
| description | Use when crafting fantasy cover art, fantasy cover concepts, book-cover prompts, fantasy image-generation workflows, or pixel-review refinements for fantasy fiction. |
Crafting-fantasy-cover-art turns a spoiler-safe fantasy brief into a commercially coherent cover-concept package. It is for reader-contract design, generation prompts, image review, and layout guidance. It is not a prose skill and it does not own canon or story lifecycle.
Resolve or clearly mark assumptions about:
Do not block on low-risk omissions. Infer sensible defaults when the omission is unlikely to change the design materially, and label the assumption.
State the reader contract before writing prompts: who the book is for, what fantasy shelf it belongs on, what emotional promise it makes, what differentiator must be visible, and which facts are off-limits. If the brief is missing a low-risk piece, infer it and mark it as an assumption instead of stopping.
Current fantasy-cover trends shift faster than durable principles. When the task depends on market fit, scan recent comparable covers in the closest shelf first. If that research is unavailable, say so and rely on durable principles from the bundled reference rather than claiming a trend scan happened.
Each direction must make a different strategic choice, not just a palette swap. Use a different dominant focal strategy, such as:
Each direction should explain why that strategy fits the brief and what reader expectation it creates.
Do not leave the hook as abstract lore. Turn it into an action, object, posture, tool, mark, environment cue, or magical consequence that a thumbnail can actually read. Spoiler-safe invention is allowed, but proposed visual devices are design choices, not canon. Label invented cover devices as proposed rather than factual.
Pick the direction that best satisfies the reader contract, age-market register, and thumbnail recognition while avoiding generic genre shorthand. Say why the rejected directions lose on this brief.
Every prompt should be a coherent image specification, not a pile of aesthetic nouns. Default to a tall portrait composition and keep the art text-free. Specify, in this order:
Typography must normally be added later. Reserve exact quiet fields with value, detail, and light constraints, not vague “leave space” language. Name where the title lives and where the byline lives, for example by specifying a low-detail upper band for title and a calmer lower band or margin band for byline, with no faces, hands, props, particles, architecture, or bright hotspots in those zones.
If image-generation tools are available in the current environment, generate the recommended direction and keep the prompt plus every model, size, seed, and generator detail the tool actually returns together with the output. Do not invent metadata that the generator did not return. Do not drop the prompt after the image is made.
Review the result at full size and as a real thumbnail. Create a separate thumbnail file—120 × 180 px for a 2:3 portrait, or the equivalent 120 px width for another ratio—and inspect it at 100% display size rather than imagining the full image smaller. Check whether the title zone is actually quiet, the byline zone is actually usable, the differentiator survived, and the image still reads at small size. Revise only the failures that are visible in the pixels. If the first render misses the target, produce one corrected candidate rather than endless variants.
The skill should leave the typography to later composition unless the user explicitly asks otherwise. State the title zone, byline zone, and any collision risks clearly. If the prompt can only support typography with overlays or panels later, say that directly. Do not treat an ebook front as a print-ready full wrap.
Use a short final gate:
Use references/cover-principles.md in this skill directory for the evidence-derived rationale, focal-strategy trade-offs, prompt anatomy, rendering review details, technical guidance, and source-role caveats.
