12 条首次生成的实际输入

以下英文正文逐字取自 2026-09-09 的实际生成输入。这些是本实验编写的原创提示词，未复制外部图库的完整提示词。每条是本轮单图任务的具体要求，其中数量、风格、画幅和参数不是本 Skill 的默认规定。模型为当时可用的内置 image_gen；没有保存固定 seed 或可锁定的模型构建标识，因此不能承诺重新生成相同像素。

首次生成未附外部或内部参考图；五次编辑使用各自的 v1 首图，输入见 REVISIONS.md。本页 03/04/07/08/12 的原始输入对应 v1，画廊展示的是修订后的 v2。

01 character-asset

Use case: stylized-concept. Asset type: single character reference image. Create exactly one original clay stop-motion fox postman, full body strict left side profile, facing left, on a seamless pale beige background. The fox has a compact ochre-orange clay body, cream muzzle, a wide heavy tail, and a small triangular notch missing from the left ear. It wears a short deep navy coat with exactly one round brass button on the chest and carries one worn brown mail satchel across the back; the strap presses naturally into the coat and the bag has believable weight. Preserve the strict side silhouette: do not turn the face or torso toward camera. Both feet touch the floor with a soft contact shadow; leave breathing room in front of the nose and behind the tail. Large soft light from upper left, matte clay with subtle hand-pressed irregularities, cloth and leather still interpreted as crafted clay with different surface response. Single image only, no grid, no alternate views, no extra buttons, badges, hat, letters, extra characters, text, logo, watermark, photoreal fur, glossy vinyl, duplicate limbs or duplicate tail.

02 scene-asset

Use case: stylized-concept. Asset type: single environment reference image. Create exactly one empty old railway waiting hall viewed at adult eye height from the entrance, looking straight into the room. Lock the layout: the left wall has exactly three tall windows in a row; the back wall has one large round clock centered above one continuous row of worn wooden benches; the right wall has one closed ticket window with a shallow counter. Keep a central walking aisle from camera to the back wall. Deep one-point perspective with connected floor, walls and ceiling; all doors and structural lines physically coherent. Overcast neutral daylight enters only from the three left windows, producing soft rightward shadows and readable detail under the benches and around the ticket window. Materials: scuffed stone floor, aged plaster, dark wood, oxidized metal window frames, each reacting differently to light. No people, animals, luggage, posters, signage, readable text, extra windows, duplicate clocks, furniture collage, warped architecture, orange sunset cast, fog, supernatural light, logo or watermark.

03 prop-asset

Use case: product-mockup. Asset type: single prop reference image. Create exactly one original palm-sized portable radio on a low warm-gray studio plinth, front face clearly visible with a slight three-quarter angle. Lock the product: cream-white rounded rectangular body; exactly one orange circular knob on the left side of the front face; exactly three horizontal rows of small circular speaker holes on the right; no screen. Keep the silhouette simple and the body thickness believable. Fine matte molded-plastic texture with a subtle injection seam, soft wide reflections, crisp circular recesses, and a continuous contact shadow on the plinth. Neutral softbox from upper left; muted gray-green background; orange knob is the only saturated accent. No second radio, text, numbers, labels, logo, watermark, antenna, handle, cable, screen, extra knob, extra buttons, glowing holes, exploded view, annotations, grid, floating product, mirror duplication or scratched vintage distress.

04 cyberpunk-design

Use case: photorealistic-natural. Asset type: cinematic still. In a cramped underground prosthetic clinic, a tired female mechanic in a faded charcoal work shirt adjusts the exposed external elbow actuator of a seated adult courier. The courier's natural arm enters a clearly engineered mechanical forearm; show the boundary between skin, a soft pressure liner, articulated metal linkage, and braided cable without fusion or body horror. The mechanic's left hand stabilizes the forearm while her right hand tightens one visible fastener; both look at the same contact point, nobody looks at camera. Eye-level medium shot through the edge of a hanging translucent divider, with a clear path to the back door. A cool cyan task lamp above the workbench illuminates hands and actuator; one tiny magenta status LED is the only strong accent. Worn metal, rubber, woven cloth, skin and wet concrete each respond differently. Background monitors remain dim and unreadable. Photorealistic, grounded future, readable shadows. No floating HUD, no text, no logos, no full-body neon outline, no rainbow signs, no extra limbs, no cable merging into skin, no weapon, no beauty retouching, no plastic skin, no smoke hiding the repair, no watermark.

05 epic-design

Use case: historical-scene. Asset type: cinematic story still. A vast mountain-valley floodgate is being manually held open for the last civilians crossing before rising water reaches the causeway. In the lower foreground, two exhausted gatekeepers lean their full body weight into a timber-and-iron capstan, feet braced in mud, rope taut. Midground: a family and pack animal hurry across the stone causeway, varied poses and spacing. Background: the colossal gate towers climb the cliffs and continue out of frame; counterweight chains descend into masonry slots; far valley settlements and storm clouds prove distance. Low adult-height wide view, strong diagonal from capstan to causeway to gate, people small but readable. Overcast daylight with one restrained warm torch near the mechanism, desaturated stone, earth and deep blue-gray; wet stone, rope fibers, iron oxidation and mud contact are visible. The architecture has credible supports and load paths. No aerial view, no cloned crowd, no decorative flags, no fantasy glow, no text, no logos, no weapons, no heroic posing, no impossible rope paths, no floating feet, no game-render gloss, no watermark.

06 fantasy-design

Use case: illustration-story. Asset type: cinematic fantasy still. Inside an abandoned glasshouse at dusk, a young adult apprentice kneels beside a cracked bronze basin containing one amber seed. The seed is the only magical light source. Thin luminous roots grow outward from the seed through real soil and enter the trunk of one dormant twisted tree; wherever the roots make contact, a small band of bark warms and the first few leaves unfold. The apprentice's hand has stopped just above the basin and their eyes follow the root path, showing surprise without theatrical posing. Wide medium view with broken glasshouse ribs framing the scene, foreground shards and soil, apprentice and basin in the middle, tree and misted panes behind. Cool blue-gray dusk outside, restrained amber light local to seed, roots, fingertips and bark contact. Bronze patina, damp soil, cracked glass, woven clothing and bark remain distinct. No generic aura, no floating runes, no multiple seeds, no extra characters, no castle, no dragon, no random neon, no text, no logo, no plastic tree, no light without a source, no watermark.

07 horror-design

Use case: photorealistic-natural. Asset type: cinematic horror still. A modest family dining room in clear overcast morning light. An adult woman stands just inside the doorway holding a folded dish towel, frozen mid-step, looking down at a dining chair that has been pulled away from the table. A line of fresh wet bare footprints crosses the dry wooden floor from the closed back door and stops beneath the empty chair; there are no footprints leading away. The woman, chair, last footprint and closed door must all be readable in one frame. Human-eye camera at waist height, slightly off-center one-point composition; foreground doorframe creates mild obstruction but does not hide evidence. Natural gray daylight from a left window, one warm ceiling bulb switched off, shadows open and readable. Real wood grain, damp footprint sheen, cotton towel and ordinary painted walls. No visible monster, ghost face, blood, gore, black void, green filter, jump-scare figure, extra person, text, logos, floating objects, excessive fog, Dutch angle, horror poster typography or watermark.

08 noir-design

Use case: photorealistic-natural. Asset type: cinematic noir story still. Night inside a small dockside accounting office. A middle-aged accountant in a wrinkled white shirt has paused with one corner of an open paper ledger held just above a metal wastebasket containing a small controlled flame. He notices a fresh wet handprint on the inside surface of the closed window beside him; rain streaks remain outside, so the inside handprint is the concrete clue. His eyes move from the handprint toward the locked office door reflected faintly in the glass; his other hand tightens on the ledger. Medium-wide eye-level frame, desk edge in foreground, accountant and ledger in warm desk-lamp light, handprint in cooler window light, locked door still readable in background. High contrast with open shadow detail, deep blue-gray, worn brown, and a restrained amber practical light. Paper fibers, condensation, scratched desk metal and damp glass are distinct. No gun, detective cliché hat, blinds pattern, smoke, exaggerated Dutch angle, dead-black face, unreadable clue, extra people, text overlays, logos or watermark.

09 romance-design

Use case: photorealistic-natural. Asset type: cinematic romance still. At a quiet railway platform before departure, two fictional adult lovers are separated by the closed window of an old train carriage. The woman stands inside the softly lit carriage and the man stands outside on the cool platform. Their open palms align on opposite sides of the glass but do not physically touch; both look at the other's eyes, restrained and unsmiling. Show condensation around their hands and a narrow clear patch between them, with subtle reflections that do not duplicate faces. Medium shot from the platform at human eye level, the vertical window frame keeps them in adjacent spaces, empty platform recedes behind the man with purposeful negative space. Cool blue-gray station daylight, small warm carriage practical lights, natural skin, worn wool coat, glass thickness and metal frame. No kiss, hug, wedding styling, pink filter, dreamy bloom, heart shapes, extra people, text, logos, melodramatic crying, plastic skin, duplicated hands or watermark.

10 war-design

Use case: photorealistic-natural. Asset type: military cinematic story still. Inside an original near-future aircraft maintenance hangar, a female technician with short black hair and transparent safety goggles has stopped inspecting an open service port after finding one small orange tamper seal attached across an internal latch where it should not be. A male pilot in the same deep-gray work uniform still supports the heavy detached access panel with both hands; a pale square cloth patch is clearly sewn on his right sleeve. The technician points toward the seal without touching it, and both look at the seal, not camera. The open port, orange seal, supported panel, both faces and the clear route to a lit exit are readable in one continuous space. Medium shot from an oblique front angle. Neutral skylight defines uniforms and aircraft skin; one warm work lamp illuminates only the port and hands, making the seal the restrained accent. Woven cloth, transparent goggles, painted metal, rubber liner and fasteners remain distinct; panel weight affects hands and posture. No weapons, combat, tactical HUD, national insignia, readable labels, extra crew, floating cables, random crates, smoke, orange-blue wash, heroic pose, text, diagrams, logos or watermark.

11 wuxia-design

Use case: historical-scene. Asset type: cinematic wuxia action still. In a rain-soaked stone courtyard, an adult woman swordsman in a dark indigo robe uses a cloth-wrapped wooden scabbard held with both hands to stop an adult male opponent's descending single-edged saber. Freeze the exact contact instant: saber edge presses against the scabbard near its upper third; her feet are planted wide on wet stone, front knee bent, hips and shoulders receiving the force; his raised arms, forward step and torso drive downward. Both weapons, four hands, contact point, foot placement and opposing force directions remain clearly visible. Low side three-quarter camera, medium-wide view along the courtyard wall; a red lantern under the eave is the only warm accent and casts a matching reflection on wet stone. Overcast cool daylight, rain streaks angled by one wind direction, clothing hems and loose hair respond consistently, water sprays outward from the weapon impact and planted feet. Real cloth weight, wet stone, wood compression and steel sheen. No floating bodies, wire-fu pose, extra fighters, duplicated weapons, blade through hands, broken anatomy, random slow-motion debris, heavy fog, text, logos or watermark.

12 xianxia-visual-director

Use case: photorealistic-natural. Asset type: 16:9 live-action xianxia cinematic still. Human-height camera on a broad weathered stone ritual platform with open edges and absolutely no railings, looking slightly upward through a colossal bronze armillary instrument that crosses two cloud layers and continues beyond the upper frame. A single fictional adult immortal, seen from a 45-degree rear angle and about 4% of frame height, walks slowly toward the instrument; layered ivory-and-muted-teal robes have two unequal rear hems, with hair and cloth moving in one wind direction. Fixed five layers: a near-camera chipped stone arch fragment frames the left edge; the human-bearing platform and tiny adult form occupy the lower foreground; the ruling bronze armillary and suspended pale-stone observatory dominate the middle and upper frame; lighter bridges, floating terraces and smaller observatories recede behind; an extreme-distance higher realm and pale celestial river remain visible above a low cloud sea. Four scale proofs are unmistakable: the instrument ring is wider than hundreds of bodies, a support pier is eight adult heights thick, repeated bridge bays diminish into distance, and the main ring disappears through cloud and out of frame. Preserve 40–60% content-bearing air through ring openings, cloud gaps, void under bridges and distant sky. Treat foreground stone, actor and robe as practical photography; middle architecture as physical set extension; impossible far realm as invisible VFX with progressive loss of contrast, chroma and detail. 65mm large-format logic, 50mm spherical lens, natural deep focus falloff, soft highlight roll-off, restrained grain. High-altitude morning light from camera right, cool transparent shadows, near-neutral pale stone, dark patinated bronze, narrow muted-teal cloth accent. No moon gate, no white-jade corridor, no front-facing portrait, no cloned figures, no dense roof carpet, no global gold cast, no neon, no perfect CGI surfaces, no artificial HDR, no uniform fog, no text, logos or watermark.