秋のちくわ狩りが最盛期　専門家『見た目では判断できない』

プロンプト

Wan 3.0各シーンと採用画像の実プロンプト · revision 1

## prompt_wan3_harvest.txt
12-second16:9 photorealistic Japanese regional TV field report, quiet autumn forest, natural overcast light and restrained documentary color. Image1 Yotan canonical identity: blond hair, dark round glasses, black leather jacket, trousers, boots; NO guitar or strap. Image2 Fukugyun canonical identity: black hair, black blazer, white printed shirt, trousers, white shoes; NO badge or lanyard. Portrait panels define faces, body panels define clothing; NEVER generate headless people, multiple copies, sheet labels or white studio backgrounds. Image3 exact initial composition and woodland environment: Yotan left, Fukugyun right, wicker basket between, crouching beside chikuwa patch. Image4 defines chikuwa growth, small 9-12cm beige hollow fish cakes with brown toasted belts, irregular emerging angles, half buried in damp brown leaves, soil and moss. No mushroom caps, faces, roots or cartoon limbs. Only these TWO people, no Sobaya.
Audio1 voice-timbre ONLY Yotan; Audio2 voice-timbre ONLY Fukugyun. Do not repeat reference words or timing. All speech generated in Japanese with synchronized visible mouths.
Shot1 [0-3s editorial guide]: Begin with composition of Image3, genuine motion immediately: Yotan slowly pulls ONE chikuwa free from ground; soil crumbs fall, he lifts it. Fukugyun puts another already picked chikuwa into basket. No grabbing dirt clumps as food. Food stays consistent.
Shot2 [3-9s editorial guide]: Cut to medium chest-up two-shot, Yotan holds his chikuwa at chest level and speaks naturally exactly 「今年は肉厚ですね。焼き目もきれいについてます」 (ことしはにくあつですね。やきめもきれいについてます). Calm matter-of-fact harvest interview, light pleasant expression, no exaggerated laughing. Fukugyun listens with mouth closed. Hold until Yotan finishes and closes mouth.
Shot3 [9-12s editorial guide]: Cut or gently reframe toward Fukugyun face chest-up, he replies naturally exactly 「今夜はおでんギュン」 (こんやはおでんぎゅん). Preserve final GYUN, not desu/ne. Yotan remains silent. End with closed mouth and small pleased nod.
Time ranges are editing guidance, not instructions to stretch speech. Natural speaking speed and short pauses take priority. No extra words, no overlaps, no narrator, no music or laughter track. Quiet woodland air, light leaf rustle and clothes movement only. No text, clock, logo, subtitles or watermark. Top corners and bottom fifth reserved for later graphics. Keep food above bottom fifth when presenting it.
Fukuchan's skin quality and apparent age must match the close-up portraits in his canonical character sheet. Keep his well-cared-for, clean, even complexion and natural fine skin texture. Preserve the sheet's subtle forehead lines, eye-area wrinkles, smile lines and nasolabial folds, including their natural changes with expression. Do not add or exaggerate dark spots, freckles, blemishes, rough pores or mottled skin beyond the sheet. Do not erase all wrinkles, de-age him, reshape his face, or give him plastic, waxy or porcelain skin. Preserve the scene's lighting and natural facial shading without turning shadows into blemishes.


## prompt_wan3_expert.txt
One continuous 12-second 16:9 photorealistic Japanese university expert remote interview, fixed chest-up camera, no cuts. Image1 is the exact face, outfit, office and composition: preserve this same Okayaman, dark gray blazer, light gray shirt, bookshelf and microscope. Face is identity reference, not a frozen expression. Audio1 is voice-timbre reference ONLY for Okayaman, never repeat reference words or timing. Generate his Japanese voice and mouth motion together.
Shot 1 [0-4 seconds, editorial guide only]: Looking into lens, calmly and naturally say exactly once 「そうですね。注意が必要です。」 (そうですね。ちゅういがひつようです). Normal natural conversational speed, no drawn-out words, no added greeting or name. Show synchronized lips, jaw and cheeks unobscured until speech ends and lips close. Mouth closes in pauses. Do not stretch speech to fill four seconds.
Shot 2 [4-8 seconds, editorial guide]: SAME uninterrupted camera and posture. Silent attentive listening, mouth closed, one slight nod and natural blink. No speech, no fake lip movements.
Shot 3 [8-12 seconds]: SAME camera, subtly furrow brow and hold a mildly concerned, restrained listening reaction. Mouth remains closed, natural breathing/blinks. Do not turn away, smile broadly or gesture.
Reusable clean footage: NO chikuwa, no handheld object, no episode-specific imagery, no other people, no text/captions/logo/clock/name tag. No music, laughter, narration or offscreen voice. Quiet neutral office room tone only. Keep face and clothes consistent throughout. Neutral believable daylight. Leave top corners and bottom quarter clear for later news graphics. Keep natural Japanese delivery and short pauses more important than target shot boundaries.

## prompt_wan3_expert_listening.txt
A single continuous SIX-second photorealistic 16:9 silent listening reaction shot. Image1 is exact Okayaman identity, fixed chest-up composition, neutral office, dark gray blazer and light gray shirt. Preserve face, hair, mustache, beard and clothing. His lips remain gently TOUCHING and fully CLOSED for ALL six seconds. No speaking, no murmuring, no lip articulation at any moment. He is ONLY LISTENING to someone off camera. First three seconds: calm attentive eyes into lens, one very small nod with closed lips. Last three seconds: eyebrows slightly tighten in mild concern, then hold attentive expression and blink naturally. Mouth stays closed even when nodding, no smile showing teeth. Static camera and steady scene. Real subtle breathing, no frozen photograph. No cuts or text, logo, captions, clock, name badge, chikuwa or extra people. Quiet office room-tone audio ONLY, absolutely no human voices or speech, no music, no effects. This is a reusable B-roll listening clip with no dialogue at all.

## prompt_wan3_theft.txt
18-second16:9 photorealistic Japanese local news field report. ONE continuous real-time camera shot with NO cuts, all causally connected. Image1 Sobaya canonical human identity: muscular large adult, short black hair, neutral gray skin, rigid white mask with round black eyes, black forehead dot, red cheek markings, narrow mouth slot, white fitted T-shirt black trousers white sneakers. NO beer mug. Mask stays correctly attached and UNCHANGED every frame: never removed, lifted, translucent, replaced, stretched or turned into human face. Chest fabric retains muscles and natural folds but no nipple protrusions/outlines/point shading. Image2 Fukugyun identity, black hair black blazer white printed shirt black trousers white sneakers, no badge/lanyard. Portrait defines face, body panels clothes only, never render sheet layouts or duplicates. Image3 shows the mid-action food/hand/mouth spatial relationship and same woodland, NOT opening frame: Sobaya must be ABSENT initially and enter later. Image4 Sobaya's human quadrupedal front posture, palms and feet with knees raised. Image5 rear quadrupedal escape posture. Image6 final collapsed prone/rear pose, facing AWAY from camera. Image7 exact purple poison chikuwa with dark purple toasted belt magenta and yellow-green mottling, ordinary hollow food shape, no glow or smoke. Image8 forest floor, scattered ordinary beige chikuwa, damp brown leaves and moss. Maintain one location, plausible scale and continuous screen direction.
Audio1 Fukugyun voice-timbre ONLY. Audio2 Yotan voice-timbre ONLY for the ONE offscreen reaction. Never repeat reference words. No Sobaya speech, no animal vocalizations.
Opening [0-4s editorial guide]: Medium shot of crouching Fukugyun on LEFT, looking at news camera with face unobscured, holding ONE purple chikuwa horizontally by only its leftmost end in his right fingertips at knee height. Its right end projects toward the empty RIGHT side. He says exactly 「こちらも、おいしそうなのが採れたわよー」 (こちらも、おいしそうなのがとれたわよー) in his natural voice, relaxed lightly feminine playful ending, not shouting. Generate lips/jaw/cheeks in sync; finish entire sentence before action. Natural speed, no forced stretching. Sobaya is OUT OF FRAME at start.
Action [4-8s editorial guide]: Sobaya rapidly ENTERS from RIGHT on all fours, using BOTH HUMAN PALMS and FEET, knees raised, like a predatory bear crawl but clearly the same muscular HUMAN. He lunges his masked mouth to the free end of chikuwa and snatches it with his mouth through the unchanged mouth slot. Fukugyun releases his fingertips from food as it moves into Sobaya's mouth. Do NOT bite fingers, do NOT receive food with hand. Show a clean readable transfer of ONE chikuwa from fingers to mouth; no duplicate food remains in hand. Both Sobaya hands support locomotion. His rigid mask never opens or distorts; food can be held at its existing mouth slot. No blood or wounds.
Escape [8-13s editorial guide]: Sobaya pivots away and scampers quickly ON ALL FOURS into the woodland upper right, mouth carrying the same purple chikuwa. Camera reacts a fraction late, pans RIGHT and takes a short forward step to FOLLOW his back continuously; Fukugyun leaves frame LEFT naturally. Sobaya never stands upright or becomes an animal. All four limbs move physically with alternating support. Keep him fully visible, genuine travel of several meters, not running in place.
Collapse [13-18s editorial guide]: While fleeing AWAY, his arms suddenly buckle and he falls forward onto leaf litter, BACK toward camera, ending close to Image6. Camera follows and tilts slightly down, stays BEHIND him, never cuts to front or sees his face. Then two or three unmistakable short involuntary spasms through shoulders, back and legs while lying down; he does not stand, roll toward camera or resume crawling. A quiet OFFSCREEN Yotan says only 「えっ……」 after the collapse. Hold rear view long enough to read spasms before end. Purple food may fall immediately in front of his masked head, never reappears with Fukugyun.
Hard constraints: exactly one Sobaya and one Fukugyun visible; Yotan never enters frame. One continuous camera following the fleeing body. Correct temporal order and one food transfer. No talking Sobaya, no music, no comic sounds or laughter. Woodland ambience, footsteps/leaf crunch, brief scuffle, dull soft fall only. Normal documentary exposure and no horror effects. NO readable text, subtitles, logos, clocks, labels. Leave upper corners for later graphics. Natural dialogue/physics take priority over exact editorial shot times.
Fukuchan's skin quality and apparent age must match the close-up portraits in his canonical character sheet. Keep his well-cared-for, clean, even complexion and natural fine skin texture. Preserve the sheet's subtle forehead lines, eye-area wrinkles, smile lines and nasolabial folds, including their natural changes with expression. Do not add or exaggerate dark spots, freckles, blemishes, rough pores or mottled skin beyond the sheet. Do not erase all wrinkles, de-age him, reshape his face, or give him plastic, waxy or porcelain skin. Preserve the scene's lighting and natural facial shading without turning shadows into blemishes.


## prompt_imagegen_expert.txt
Use case: identity-preserve. Create a photorealistic 16:9 reusable Japanese university expert remote TV interview clean plate. Image 1 is the STRICT face identity reference for Okayaman: match same person, facial proportions, black medium parted hair, mustache and chin beard, age. Chest-up seated at center, looking directly at camera, lips gently closed, calm attentive neutral expression. Dark charcoal casual blazer over plain light gray shirt, no tie. Modest authentic Japanese university research office: shelves with unlabeled books and paper files, subtle botanical specimen and small microscope in soft focus, neutral daylight. No chikuwa or any episode-specific props. Static eye-level camera, natural webcam interview perspective but broadcast quality. Head fully inside frame, face centered in upper middle, enough bottom 25 percent space for captions without covering face. Natural skin texture and identity fidelity, no beautification, no dramatic lighting. No text, logos, badges, graphics or watermarks. This is reusable reference for future talking, listening and reaction video clips, so straightforward symmetrical composition and uncluttered neutral academic background.


## prompt_imagegen_forest_v2.txt
Create a single exceptionally credible documentary photograph for a Japanese regional TV news field report, landscape 16:9. This is the fictional autumn harvest of naturally growing chikuwa. The reference is the subject concept ONLY; REPLACE its overly glossy CGI advertising look, oversized symmetrical upright tubes, glowing autumn colors and dramatic shallow focus entirely. Photograph a real damp Japanese secondary woodland floor on an overcast late-autumn afternoon, from a crouching person's oblique downward viewpoint using a normal 40mm news camera lens, moderate depth of field f/5.6, neutral white balance and restrained saturation. Humble messy brown oak leaves, faded ochre leaves, decomposing wet leaf litter, tiny twigs, coarse soil, patches of dull moss, sparse background green undergrowth. There are only FIVE ordinary small chikuwa fish cakes, actual plausible snack size 9-12 cm long and 2.5 cm thick relative to oak leaves, emerging irregularly from underneath leaf litter as if wild mushrooms. Preserve exact unmistakable finished chikuwa food shape: creamy off-white hollow cylindrical fish cake with a patchy medium-brown toasted belt and open round bore; no mushroom caps, roots, faces or limbs. Slight dents and wrinkles, matte porous fish paste, lightly damp but NOT lacquered, waxy, glassy or oily. Main pair at center left, one leaning 35 degrees and the other short with lower half buried; a third at center right partially concealed by a dry leaf; two small distant tubes scattered naturally. Imperfect nonuniform placement, subtle dirt on lower surfaces, credible occlusion and contact shadows, believable size. An ordinary harvest wicker basket barely visible at the far upper right edge establishes scale, no people. Forest floor fills image, no scenic orange forest vista or sunbeams. Authentic unstyled field evidence, quiet flat natural daylight, slightly imperfect exposure, realistic fine sensor texture. Make the extraordinary subject look mundane and physically photographed. Important tubes in central 60% so news graphics can fit top and bottom. NO text, logos, UI, dramatic grading, bokeh balls, oversharpening, oversaturation, glow or illustration.


## prompt_imagegen_comparison_harvest.txt
Create a photorealistic documentary photograph of freshly harvested chikuwa specimens, landscape 16:9. Reference image defines the TWO exact types and left/right order: LEFT ordinary ivory beige hollow chikuwa with brown toasted belt; RIGHT unmistakably purple poison chikuwa with black-purple toasted belt, magenta and yellow-green mottling. KEEP the poison colors obviously different; the visual joke is the TV report claims they cannot be distinguished. Change the pristine studio product setting into a real outdoor harvest inspection. Exactly two whole ordinary snack-sized tubes laid horizontally side by side on a weathered gray-brown wooden field table beside a Japanese autumn woodland. Both freshly pulled from soil, a little moist dark earth clinging to their lower ends and surface creases, a few scattered leaf fragments and soil crumbs on the table. A harvest wicker basket and dry fallen leaves only at extreme upper edge, softly out of focus. No roots, stems, mushroom caps, limbs, faces or new structures. Real cooked fish-cake surface, modest damp sheen rather than plastic varnish, subdued natural overcast light, neutral color balance. Both specimens fully visible with open round holes visible, equally sized and equally lit. Match reference composition precisely: left tube occupies x4%-47%, right x53%-97%, main specimen pixels between y32%-68%, generous background margin top and bottom for later crop. Table fills background. Serious local news field photography, not food advertising. No text, labels, logos, graphics, borders, magic glow, smoke, utensils or plates.


## prompt_imagegen_scene_harvest.txt
Photorealistic Japanese local news harvest scene, single16:9 landscape photograph. Image1: exact identity Yotan, blond hair, dark round glasses, black leather jacket black shirt trousers boots. Image2: exact identity Fukugyun, black medium hair, black blazer white printed shirt black trousers white shoes. STRICTLY preserve each person's actual face from their photograph, no mixing faces. Fukugyun has clean well-cared-for skin with natural age and expression lines; no added freckles, spots, roughness, waxy smoothing or de-aging. Image3: exact forest floor and chikuwa growth reference: damp brown oak leaves, soil, moss, green undergrowth, overcast light; ordinary snack-sized hollow beige chikuwa with brown toasted bands grow irregularly from leaf litter. Exactly these TWO men crouch naturally beside the patch, Yotan on LEFT, Fukugyun on RIGHT, woven harvest basket between them. Yotan reaches toward one chikuwa still rooted in the litter, Fukugyun watches. Their faces clearly visible, neutral pleased expressions as ordinary mushroom harvest, not broad comedy. No guitar, guitar strap, amplifier, name badge or lanyard. Full heads, hands and chikuwa visible; medium-wide eye level from crouching reporter, enough skyless woodland behind. About five scattered chikuwa, not oversized. Natural subdued documentary photography, no orange glow or exaggerated gloss. No other people or masked man. Leave bottom fifth for news overlay but do not render text, logos, panels or watermark.


## prompt_imagegen_scene_theft.txt
Create a single16:9 photorealistic Japanese local news field shot, preparation for a continuous action. Image1 is composition and matching woodland location: same crouching black-haired Fukugyun in black blazer white printed shirt black trousers white sneakers, woven basket and ordinary chikuwa patch. Image2 is the ONLY facial identity source for Fukugyun, preserve exact actual face and clean well-cared-for skin with natural wrinkles and age; no added spots or de-aging. Image3 is Sobaya quadrupedal posture and identity: large muscular HUMAN with gray skin, white rigid mask with black dots/red marks, white T-shirt black pants white sneakers. Image4 defines the purple poison chikuwa shape/color. In this shot exactly TWO characters: Fukugyun LEFT, crouching, facing camera and holding ONE purple poison chikuwa horizontally in his right hand at knee height (about45cm above ground). His fingertips hold only the far-left end, the other end extends freely toward RIGHT. Sobaya on the RIGHT about one meter from the chikuwa, on hands and feet knees elevated, face lifted toward free end preparing to bite. He has NOT bitten yet, a visible gap between mouth and chikuwa. Human anatomy two arms two legs, no beer. Fukugyun calmly presents it as appetizing ordinary harvest. Sobaya's mask stays intact, no widening mouth slot, no teeth. No Yotan or extra people in this composition. Space behind Sobaya leads diagonally into woodland upper right for escape. Medium-wide view with both whole bodies and the connecting hand/food/mouth relationship visible. Do not crop heads or hands. Ground is wet brown leaf litter and sparse green plants, overcast natural light. White T-shirt has natural chest muscles and cloth folds but NO nipple protrusions/points/shadows. No text, logos, clocks, captions, borders, no horror lighting.


## prompt_imagegen_sobaya_quadruped_front.txt
Create a single landscape 16:9 full body action reference photograph of EXACTLY ONE Sobaya, matching the supplied character sheet's white rigid mask (black round eyes, black forehead dot, red vertical markings, narrow horizontal mouth slot), short black hair, large muscular human body, uniformly neutral gray exposed skin. Plain fitted white short sleeve T-shirt, black trousers, white sneakers. NO beer mug. He is on all fours like a fast predatory HUMAN bear crawl: BOTH palms flat on the ground, BOTH feet/toes supporting him, BOTH knees visibly raised above ground, hips slightly raised, face lifted forward. Three-quarter front view so all four limbs and intact white mask are readable. Exactly two human arms, two human legs, normal fingers; NOT an animal, not baby crawling, no extra limbs. Keep the natural chest bulk and fabric folds, absolutely no nipple bumps, nipple outlines or paired point shadows through T-shirt. Dull brown wet autumn leaf litter ground and softly blurred ordinary Japanese woodland, flat overcast daylight, no orange cinematic glow. Camera at knee height, full body with margin, no text, labels, split panels or additional people. Identity and clothes from sheet, do not copy sheet layout or props.


## prompt_imagegen_sobaya_quadruped_rear.txt
Create one 16:9 full-body action-reference photo of exactly ONE Sobaya in the same ordinary damp autumn Japanese woodland. Identity, muscular human build, short black hair, gray exposed skin, white fitted T shirt, black trousers, white sneakers from reference. Camera behind him at knee height, THREE-QUARTER REAR view, he is moving AWAY along leaf-covered ground with his BACK toward camera. Human quadrupedal bear-crawl posture on palms and feet, knees lifted clearly off ground, both arms reach forward while both human legs push; exactly two arms and two legs, natural human anatomy. Full body entirely visible with wide environmental margin. His white rigid mask remains correctly attached to the front of head but is mostly hidden by rear view; only a tiny side edge if needed. No facing camera, no head twist backward. No beer mug, no chikuwa, no other people. Flat overcast daylight, muddy leaves, subdued natural colors, no text, split panels, labels, cinematic glow. The action reference should clearly communicate a large muscular human escaping quickly on all fours, not crawling on knees.


## prompt_imagegen_sobaya_fallen_rear.txt
Create one landscape16:9 documentary still showing the ending pose of the SAME single Sobaya in this reference woodland. Camera remains BEHIND him, viewing his back, never his face. He has collapsed naturally onto his side/front on damp fallen leaves, head points AWAY from camera toward upper right of frame, broad white T-shirt back toward camera, black trousers, bent legs and white sneakers toward lower left, two normal gray human arms on the ground. Entire body in frame. Exactly one muscular human body, no anatomical distortion. His rigid white mask remains properly worn on his face facing AWAY, not removed or lying on ground. Short black hair. This is a staged fictional slapstick collapse pose, no blood, wounds or bruises. No people, text or objects added. No beer, no chikuwa necessary. Natural overcast documentary lighting, no dramatic grading. Ground and background exactly consistent with reference. Wider camera at standing waist height, not an extreme closeup, body fills about middle half of frame. Reference for subsequent brief shoulder and leg twitching while lying down, not rolling over or standing up.


## prompt_imagegen_forest_poison.txt
Edit reference image 1 (forest_v2.png) into ONE photorealistic documentary image for a fictional Japanese local-news report. Preserve the exact natural Japanese woodland floor, wet fallen oak leaves, dirt, restrained overcast light, plausible food-sized chikuwa growing upright/leaning out of soil like mushrooms, real camera texture and 16:9 composition. Reference image 2 is ONLY the appearance reference for the purple poisonous chikuwa, NOT its tabletop setting. Change exactly ONE of the five growing chikuwa: the shorter foreground tube just right of the large leaning foreground chikuwa, into the poisonous variety matching reference 2: unmistakably hollow chikuwa cylinder, purple ends and magenta/dark purple toasted mottling with small yellow-green patches. Preserve its original size, tilt, open hole, natural buried base, soil and leaf occlusion. All FOUR other chikuwa remain ordinary cream-white with brown toasted bands. Exactly ONE poisonous chikuwa among four normal ones, all growing naturally in the same soil. No harvested pair, wooden board, people, labels, arrows, text, borders or UI. Physically credible damp food texture, no glow or fantasy effects. Keep all main subjects visible centrally for a landscape news inset. Produce exactly one image.