天宫视觉提示词工坊 (Celestial Palace Visual Prompt Studio)
Overview
Generate consistent, cinematic image and video generation prompts for a serene, monumental Eastern fantasy world: believable grand architecture (with restrained vermilion accents on columns, railings and brackets) rising from a bright sea of clouds, with 仙女/神女 (fairy) figures — slightly enlarged for presence but never portrait-dominant — for scale, cool neutral atmosphere pierced by selective warm light. Optional fantasy creatures (巨龙/鲸鱼/星球/空中楼阁) appear only on the more-fantasy route. This skill enforces a fixed visual system so outputs stay coherent across many images without mechanically copying any single reference.
When to use
- User asks for 天宫 / 云上宫阙 / 东方仙境 / 云海古建 / 史诗国风 / 古松宫殿 / 仙山楼阁 image or video prompts.
- User needs a visual style guide, environment design, storyboard, key visual, or image-gen prompt for cloud-sea palaces.
- User wants to review or rewrite an existing celestial-palace prompt for consistency.
Workflow
Step 1 — Lock parameters (ALWAYS ask first if unspecified)
Before writing any prompt, confirm these eight parameters with the user (use AskUserQuestion; do not re-ask any already provided). Note: AskUserQuestion accepts at most 4 questions per call — split the eight across two calls (e.g. 4 + 4) when asking all at once; parameters 2 / 3 / 7 apply only when video is requested.
1. 图片比例 (image aspect ratio): 16:9 / 9:16 / 3:4 / 21:9
2. 视频比例 (video aspect ratio): 16:9 / 9:16 / 21:9 / 4:3 (only if video requested)
3. 视频时长 (video duration): 5s / 10s / 15s / 30s (only if video requested)
4. 图片数量 (image count): 1 / 3 / 5 / custom — when >1, PREFER varying composition family over swapping small props
5. 风格路线 (style route / composition family): 宏大中轴登临 / 以建筑或古松框住云海的全景 / 露台仪式或安静聚会 / 孤身临云远眺 / 从室内通向室外的层叠揭示
6. 自定义元素 / 颜色 (custom elements / colors): ask the user which specific elements or color accents to add beyond the default 2–3 optional picks — e.g. a particular motif (莲灯、玉璧、星点) or a tint (jade-green accent, deeper amber, faint violet haze). Capture free-text input; leave empty if the user wants the default look.
7. 视频运镜偏好 (video camera movement): only if video requested — push-in (推) / pull-out (拉) / lateral pan (横移) / crane up-down (升降) / orbit (环绕) / locked static (固定，仅云动) / let skill decide by composition family. This drives the shot grammar in Step 4.
8. 提示词语言 (prompt language): English (MJ/SD/海外) / 中文 (国内平台：即梦·可灵·通义万相) / 双语对照. Determines whether Step 3–4 output English, Chinese, or both.
Step 2 — Choose composition family
Select one family (or, for multiple images, distribute across families so each image has a distinct dominant idea). See references/style-spec.md for each family's anatomy.
Step 3 — Assemble image prompt(s)
For each image, build the English prompt in layers:
1. Fixed style anchors (mandatory, from style-spec): environment dominates; cool slate-blue + selective amber-gold; volumetric cloud sea; traditional wooden palace with dark layered tile roofs & soaring eaves and restrained vermilion (朱红) accents on columns/railings/lintels/brackets; weathered pines; white carved stone; long stairs/terraces; 仙女/神女 (fairy) figures, flowing robes, near-silhouette, slightly enlarged (3–8% frame height, terrace narrative ≤10%); slow restrained ceremonial motion; cinematic fantasy realism.
2. Composition family framing: foreground frame → mid-scale figure → grand destination → cloud-sea/distant mountains.
3. Add only 2–3 optional elements (from the two-tier optional list in style-spec) — never all at once. Fantasy creatures/celestial bodies (巨龙/鲸鱼/星球/空中楼阁) belong only to the 更幻想 route or an explicit custom request; do not mix them into normal scenes. If the user specified custom elements/colors in Step 1, fold those in (replacing or augmenting the default 2–3 pick) and adjust the color anchor accordingly — e.g. add a jade-green accent, deepen the amber, a faint violet haze, or a restrained vermilion architectural accent — while keeping the cool slate-blue base dominant and the warm tone as selective lighting only.
4. Append clean-frame constraint: no text, subtitles, titles, stamps, signatures, logos, watermarks, black bars, UI overlay, "China", screenshot traces.
5. Append the shared negative prompt (from style-spec) to exclude drift.
6. Add aspect-ratio flag (e.g. --ar 9:16 for Midjourney; for domestic platforms set ratio in the platform UI instead of a flag).
7. Output language per Step 1 #8 — if 中文, write the entire prompt (including clean-frame constraint and negative prompt) in Chinese for domestic platforms (即梦/可灵/通义万相); if 双语, give English then Chinese. Keep the same layered anchors and cool-dominant color rule regardless of language.
Step 4 — Assemble video prompt (if requested)
Write a shot-by-shot prompt (16:9 default) with timed segments (e.g. [0-8s] establishing, [8-18s] push-in, [18-30s] pull-back). Honor the camera-movement preference from Step 1 #7. If "let skill decide", derive motion from the composition family: central-axis → slow crane rise along the stair; framing panorama → lateral parallax across the cloud sea; terrace → gentle push-in to the figures; lone figure → locked static with drifting cloud; interior reveal → slow dolly-through the gateway. Keep all motion slow, ceremonial, and consistent in lighting, with tiny figures for scale and the same clean-frame constraint. In delivery, note that single models often cap at 5–10s; suggest stitching 3×10s or using a long-video model (可灵 1.6 专业版 / Sora / 混元视频 / Wan). Output language follows Step 1 #8 (English / 中文 / 双语), consistent with Step 3.
Step 5 — Self-check
Verify against the delivery checklist in references/style-spec.md before delivering: palace/terrace/stairs/cloud-view is the main subject; ≥3 depth layers; 仙女 figures 3–8% (≤10% terrace), still near-silhouette, not portrait-dominant; cool-dominant color with restrained vermilion accents only; single dominant composition; coherent architecture; distinct materials (pine/cloud/stone/wood); no text/UI/watermark; quiet transcendent mood.
Step 6 — Deliver
Provide a markdown prompt card containing: locked parameters, shared negative prompt, each image prompt (English + short Chinese breakdown), the video prompt (if any), clean-frame constraint list, self-check results, and adjustable knobs (更建筑化 / 诗意 / 叙事 / 写实 / 幻想). Reference references/example-outputs.md for a fully worked 6-image + video example.
Resources
references/style-spec.md— full fixed style anchors, 5 composition families with anatomy, optional elements, rejection offsets, clean-frame constraints, shared negative prompt, delivery checklist.references/example-outputs.md— worked example: 6 image prompts (9:16) covering 4 families + central-axis time variant, plus a 16:9 / 30s video prompt.scripts/andassets/are unused by this skill and may be deleted.
