動画に戻る

窓際族物語 / 制作ノート

THE SOBAYA

Wan 3.0で、ハリウッド映画のようなスケールと迫力に挑戦した短編。巨大化したそば屋が赤坂から飛び立ち、遠くのビール工場へ。セリフに頼らず、超高速移動と衝撃波、着地の重量感、豪快な一杯を描きました。

使用モデル

Wan 3.0生成 v1 · 720p完成版・28.2秒

制作メモ

THE SOBAYA|No BGM版 Wan 3.0でハリウッド映画のようなスケールと迫力に挑戦。巨大なそば屋が赤坂から遥か遠くのビール工場へ飛び、一杯を求める短編です。 480pで試作後、720pで30秒の本編を生成。空中移動と飲酒を部分再生成し、高速通過を編集で短縮。飲酒は27秒地点のフレームを固定して生成しました。キャラクターシート、工場、ラベル、場面構図、森林の参照画像を使用。掲載プロンプトは実際に使用した3本です。 完成版は28.2秒・30fps。横1280×720、Shortsは720×1280。公式サイトには通常比率のNo BGM版のみ掲載しています。 現行版は後付け楽曲なしで、序盤を含む生成音を保持。生成モデルが出した音楽的な音も含まれます。旧版の序盤7.8秒の無音処理を解除し、後半と同じ音量で再編集、末尾の短いフェードを維持しました。映像の再生成はありません。制作素材のBGM編集設定は制作過程の記録であり、現行版には適用していません。

制作素材 19件

IMAGE

そば屋キャラクターシート

採用した参照画像 · character_sobaya_basic_sheet.png
IMAGE

ビール工場

採用した参照画像 · reference_brewery_v3.png
IMAGE

窓際スーパーつらい ラベル

採用した参照画像 · label_super_try_flat_v1.png
IMAGE

遥か遠くの工場

採用した参照画像 · frame_02_distant_brewery_v3.png
IMAGE

加速・跳躍

採用した参照画像 · frame_03_sprint_v3.png
IMAGE

横方向の高速通過

採用した参照画像 · frame_04_side_pass_v3.png
IMAGE

低い視点の着地

採用した参照画像 · frame_05a_low_landing_v3.png
IMAGE

タンクと飲酒のスケール

採用した参照画像 · frame_06_drink_scaled_v3b.png
IMAGE

空中移動の森林・部分再生成

採用した参照画像 · reference_forest_v8.png
IMAGE

飲酒・固定開始フレーム

採用した参照画像 · first_frame_drink_v8_27s.png
DOCUMENT

完成版の音楽編集

採用設定・編集マニフェスト · edit_manifest_28p2_music_finish.json
DOCUMENT

採用カットの構成

採用設定・編集マニフェスト · patch_manifest_v8b.json
DOCUMENT

高速通過の尺調整

採用設定・編集マニフェスト · timing_refinement_v8b.json
DOCUMENT

ショート版レイアウト

採用設定・編集マニフェスト · shorts-manifest.json
DOCUMENT

本編の生成設定

採用設定・編集マニフェスト · wan3_config_v7_native720p.json
DOCUMENT

空中移動の生成設定

採用設定・編集マニフェスト · wan3_config_v8_flyby.json
DOCUMENT

飲酒の生成設定

採用設定・編集マニフェスト · wan3_config_v8_drink.json
DOCUMENT

BGMなし版・効果音のみの編集設定

採用設定・編集マニフェスト · edit_manifest_no_bgm.json
DOCUMENT

現行No BGM版・序盤の生成音を復元

採用設定・編集マニフェスト · edit_manifest_generated_audio_preview.json

採用した比較版の編集設定。後付けBGMなし、生成音を全編保持。

プロンプト

採用本編と部分再生成のプロンプト · revision 1

### prompt_wan3_v9_connected_landing_tracks.txt
Create a 30-second 16:9 Hollywood giant-action film: Sobaya is already standing upright in Akasaka, Tokyo, spots a brewery tens of kilometers away, is next seen ALREADY sprinting at full speed, launches into a physics-defying super-leap, passes across the distant sky above a forest, sends a delayed sonic shockwave toward a ground observer, lands with immense force on the mountain beside the remote brewery, carves long trenches while braking directly before a small storage tank, rips it free and drinks. Photoreal materials, monumental weight, aggressive cinematic staging and readable cause-and-effect. No dialogue or generated music.

REFERENCE INTERPRETATION: All images are visual development, NOT first/last frame locks. Invent more powerful camera movement, framing and action than the stills; never begin with a frozen reference tableau or animate a slideshow. Preserve character identity, brand and story; reinterpret cinematography boldly within the explicitly required POVs. Do not copy accidental proportions, scenery or lettering errors from a secondary reference.
Image 1: authoritative Sobaya character sheet only. Ignore layout, multiple views, captions, white background and beer mug.
Image 2: NEW brewery architecture and industrial tank design, replacing ALL earlier giant-tower tanks. Four silver tanks beside a low red-brick gabled brewhouse, mountain approach on left, clear open access directly to nearest tank.
Image 3: flat brand artwork, authoritative for lettering. Black 労働, red スーパー, large red つらい, black brush-script Madogiwa, smaller black 生, red 窓際スーパーつらい. Paint this on all four steel vessels. No 辛口, 350ml, TRY or invented words.
Image 4: distant telescopic subjective composition for discovery, no visible Sobaya. Not a fixed frame or literal factory scale.
Image 5: running energy and character pose for the ALREADY FULL-SPEED sprint. Do not reproduce it as a still opening.
Image 6: lateral distant airborne passage and depth layers: foreground forest, valley, far ridgeline and sky. The still combines effects; animate the correct delayed temporal order described below.
Image 7: ground-skimming upward monster-movie camera for landing, shoe/rock perspective and impact texture.
Image 8: manageable tank size relative to Sobaya's forearm and two-handed drinking grip ONLY. Ignore its altered factory building, extra background equipment and terrain. Image 2 controls the actual brewery. Image 3 controls lettering; Image 1 controls face/anatomy.

IDENTITY AND SIZE: Exactly one 100-meter Sobaya with canonical muscular bulky build, short black hair, neutral gray skin, correctly attached opaque white mask with round black eye holes, red marks and black forehead dot. Mask stays unchanged throughout, including drinking; never lift, remove, enlarge the mouth slit or expose a face. Fitted white T-shirt with broad chest shading and natural folds but NO nipple projections, outlines or paired point shadows; black trousers and white sneakers. Hands empty until tank extraction.
Each of the FOUR tanks is only 15 meters tall and 7 meters in diameter: about ONE FOREARM in length, about 15 percent of Sobaya's full height. When standing together on the same plane, tank tops are BELOW HIS KNEES. Choose this manageable size FROM THE FIRST FACTORY VIEW. Tank must NEVER shrink when grasped. Compact industrial vessel with steel dome, dished bottom, welded shell, ladder cage, roof guardrail, anchor ring and lower pipe connections. Not a beverage can: no pull-tab, can rim or mug handle. Three tanks remain after exactly one is removed. The held tank retains the same dimensions, ladder and pipes. Exaggerate motion, never inconsistently rescale objects.

GEOGRAPHY: Departure metropolis and destination brewery are tens of kilometers apart. Never show both in one normal wide view. The town layers in the telescopic discovery shot belong to the distant destination region. The intermediate forest is elsewhere along the route; neither city nor factory appears there. One launch, impossible aerial speed and range, no realistic ballistic limit, no teleportation, rockets or wings. At the destination a mountain slope leads through open terrain directly to the nearest tank; no building or road separates the final stopping point from the tank. Keep global travel left-to-right when visible. Mountain damage persists; factory survives the braking.

TIMING: Existing music source 26–56 seconds is added afterward. All times here are output seconds, targets for editorial alignment. Do not generate music. Launch cue 11.71, stop cue21.87, extraction cue25.79. Preserve at least3 seconds of clear drinking at the end. Suggested camera details can be enhanced, but discovery stays character-free, airborne passage stays lateral ground-observer view and initial landing stays foot-level upward.

Shot1 [0–4.87s] ALREADY STANDING: From the first frame giant Sobaya is already fully upright in Akasaka among office buildings, Tokyo Tower recognizable behind him. Low street-level wide view with normal vehicles; head and mask clearly visible. Subtle breathing and a slow head turn toward the distant target. No emergence, rising, growth, crouch or running. Brewery never visible.
Shot2 [4.87–7.80s] FIND: Cut to his pure subjective telescopic vision: distant towns and mountain ridges compress through haze, resolving the remote brewery with four modest industrial tanks. No shoulder/head/person or telescope hardware. No HUD. A rapid optical attention shift conveys supernatural acuity, not physical camera flight to the destination.
Shot3 [7.80–11.71s] FULL-SPEED APPROACH: Hard cut back to Sobaya ALREADY at maximum running speed. Do not show starting to run, standing hesitation or a stationary squat. Show two or three thunderous full strides, body pitched forward, elbows driving and asphalt blasting backward. Use low tracking with strong foreground parallax. Around10.3–11.6s use the quieter musical interval as a brief loaded FINAL STRIDE: plant a foot, compress hip/knee while still moving rapidly forward, swing arms into the leap. This is running takeoff, not an in-place crouch. At11.71s violently launch upward-forward; metropolis drops away behind him.
Shot4 [11.71–17.95s] SIDE PASS AND ARRIVING BLAST: After a brief takeoff view CUT to a stationary human-eye-height observer at a forest hillside edge looking ACROSS a broad valley toward a distant ridgeline. Horizon visible, upper sky and forest canopy visible together; NOT directly underneath Sobaya, NOT looking vertically up, NOT an aerial camera. Foreground trunks frame the sides. Keep the distant sky empty for one second, then Sobaya crosses the ENTIRE screen LEFT TO RIGHT in only 0.35 seconds, like a supersonic fighter jet. He is already fully out of frame before clouds and the forest react. No camera tracking, slow motion or lingering body pose. Maintain the empty sky afterward for the delayed shockwave. He does not approach the lens. Cloud banks split along his far flight corridor behind him. Keep the camera after he exits: a delayed sonic crack and pressure front come from that distant corridor TOWARD THE VIEWER. Distant treetops bend first, then successive bands of forest across the valley, then nearby trees violently whip with a final pressure hit and leaves/dust across lens. Calm foreground before blast arrives, visible propagation through depth, NOT all trees moving simultaneously. Clouds curl apart; condensation and dust reveal pressure without neon rings or energy beams. Do not keep giant hovering onscreen while the forest reacts. Whole event feels enormously fast and far away.
Shot5 [17.933–23.033s] CONNECTED FOOTPRINT BRAKING: Begin 10cm above rocky ground looking upward at descending sneakers. ONE continuous shot pulls back to a low side three-quarter view, always on the same side of the movement axis. Mountain left, intact brewery right; Sobaya slides straight left-to-right. His two soles slam into pristine rock and remain in continuous sliding contact, violently excavating exactly two parallel shoe-width furrows. Each furrow grows only behind its respective heel: original landing point to current sole, with the advancing tip always directly beneath that shoe. Ground ahead of the toes remains intact until crossed. No pre-existing, detached, branching, perpendicular or foreground decorative trenches. Maintain foot spacing and straight direction; no hopping or stepping out. By21.87s decelerate to a full stop directly before the nearest compact tank. Both furrows terminate under the stationary soles and extend backward along the actual route. By23s bend and reach for the tank without moving feet. Factory survives; rocks pile beside the real tracks.
Shot6 [23–30s] SAME TANK, EXTRACT AND DRINK: Use a continuous readable framing for grip and extraction, showing nearest tank and his hands together BEFORE and AFTER it leaves its anchors. He braces and tears that vessel free around25.79s; bolts snap and lower pipe tears open. No size jump at a cut, no transformation. Raise the manageable forearm-length tank in BOTH hands to face height, elbows bent, hands relatively close; do not spread arms wide to handle a torso-sized vessel. Remaining three identical tanks stay behind on the ground. Tilt it so amber beer and white foam pour from broken pipe into mask's EXISTING mouth slit. Keep face, both hands and small industrial tank visible. Continue exuberant drinking from about26.5s to end, natural spill over mask/shirt, stronger flow around29.34s. Mask remains intact and opaque, same small mouth opening. End during the drink, no credits or fade.

AUDIO: Environmental sound and synchronized effects ONLY: running impacts, pavement fracture, colossal launch, supersonic air passage, delayed sonic crack and deep boom approaching observer after the distant flyby, forest roar and branch whips, mountain impact and long grinding scrape, snapping anchors/pipes and beer pouring. No score, rhythm track, speech or narration. BGM is supplied in post.
HARD CONSTRAINTS: One consistent character, one long-range launch, lateral distant pass before blast reaches camera, foot-level landing then long braking, stop directly before compact tank, unchanged industrial tank before/after extraction. No random city attacks, factory explosions, object resizing, cans, duplicated giants, extra limbs, lifted mask or reference-sheet artifacts. No captions, UI or unrelated lettering; required beer artwork is painted on tanks. Single30-second film, no static frame matching requirement.


### prompt_wan3_v10_patch_flyby.txt
Photoreal Hollywood giant-monster action film, 16:9. Exactly one Sobaya, 100 meters tall. Image 1 is authoritative character identity: muscular bulky gray-skinned man, short black hair, attached opaque white mask with round black eye holes, red markings and black forehead dot. Mask never removed or altered. White T-shirt, natural cloth folds and broad chest shading, NO nipple bumps, outlines or paired point shadows. Black trousers, white sneakers. Empty hands. Cinematic natural daylight, immense scale and realistic materials. References guide identity and art direction, NOT mandatory starting frames. No sheet layout, text overlays, dialogue or music. Generate environmental sound and synchronized sound effects only.
Image 2 is the NEW calm forest landscape. Preserve its distinct foreground leaves and bark, varied midground tree crowns, gaps and layered hills. Keep foliage photographically resolved and temporally stable: no muddy granular texture, noisy shimmering, smearing or duplicated tree blobs. The giant is not present in this landscape image; use Image 1 for his identity. Fly across the distant sky above the ridgeline. Maintain crisp stationary trees until the pressure wave reaches them; only moving leaves briefly blur. Do not cover the whole forest in opaque dust. Do NOT hold the giant onscreen. SINGLE CONTINUOUS 7 SECOND LOCKED CAMERA SHOT, no cuts and no tracking. Human eye-height observer on forest hillside looks horizontally ACROSS a broad valley to distant ridge and sky, never straight upward. Foreground trunks and leaves, middle-distance forest canopy, far sky establish layers. City and brewery absent.
TIMING IS CRITICAL: 0.0-1.0 seconds: EMPTY distant sky, calm foreground forest. At 1.0 seconds Sobaya streaks into the FAR LEFT edge already at extreme supersonic speed, his horizontal white-shirted body traverses the ENTIRE width and is COMPLETELY OUT of the RIGHT edge by 1.35 seconds. His entire screen appearance lasts only 0.35 seconds, like a fighter jet blasting past. Motion blur, extreme displacement, just one fleeting recognizable giant silhouette. No actual aircraft. NO slow motion, hovering, flight pose showcase, approach toward camera, or camera following.
1.35-2.3 seconds: giant is gone permanently; his long wake rips the far cloud bank apart, expanding vapor sheets and rolling cloud walls mark his flight corridor. Camera remains fixed on the EMPTY sky and landscape.
2.3-5.5 seconds: delayed sonic crack and deep thunderous BOOM. A pressure front travels from the DISTANT flight corridor TOWARD the observer: far trees bend first, then successive bands across the valley, finally foreground trees whip violently, branches bow and leaves blast across lens. Physical air pressure, no glowing energy rings. A strong camera shudder ONLY when the near blast arrives.
5.5-7.0 seconds: whipping forest aftermath and split cloud corridor, giant remains absent. Main action must be complete by 6 seconds. Sell impossible speed through extremely BRIEF transit followed by delayed environmental devastation, NOT through a long visible flight. Sound: brief fast air rip followed by delayed explosive sonic boom, forest roar and branches snapping; no speech or music.


### prompt_wan3_v10_patch_drink.txt
Continue this exact first frame for 3 seconds as ONE continuous photoreal Hollywood movie shot. Preserve the existing camera, daylight, gray muscular Sobaya, opaque white mask with black eye holes and red marks, white shirt, hands, industrial beer tank, brewery and landscape. This is a seamless continuation at the end of a film, not a new establishing shot. No cut or camera angle change. The mask stays correctly attached and opaque, with exactly its existing small mouth slit. White T-shirt keeps natural chest folds but no nipple bumps or paired point shadows.
ACTION: He actually DRINKS the beer rather than pouring it over himself. In the first half-second, bring the tank's existing pouring edge/outlet right up to the mask's mouth slit, physically touching or almost touching it. Tilt his head back slightly and rotate the tank only enough for a controlled narrow amber stream to travel INTO the mouth slit. Keep the outlet immediately ABOVE and aligned with the slit throughout the drink. His hands stabilize the vessel firmly, making small alignment corrections. The main stream enters the slit and disappears inside; lips/face are not revealed. Sustain this satisfying gulp for the rest of the shot. A little foam may overflow at the contact point and a FEW droplets run down the chin, but the shirt and ground are NOT drenched. No long stream falling past the mouth, no waterfall down the torso, no geyser, no liquid emerging from the chest or tank bottom, no giant foam explosion. Preserve tank size, industrial details and existing painted label; do not transform it into a can or mug. The hands and mouth contact must remain readable, no blocking angle. Natural subtle throat/body swallowing movement, exuberant drinking conveyed through body language. End while still drinking, no fade or credits.
AUDIO: only synchronized restrained beer glugging, gentle pouring and faint environmental wind. No voice, dialogue, music, score or rhythm. No dramatic liquid splashing on ground. Keep the image's high-quality photographic textures and character identity stable.
動画に戻る