Higgsfield 入圍/最多讚/最多看的片,是怎麼做世界觀、角色卡、場景卡的

資料來源:這 68 部作品公開的專案紀錄,共 65,666 個生成工作、每一條提示詞、每個用到的模型和卡片。不是猜的,是他們真的按下去的紀錄。

一、他們做一部片花多少工

一部片中位數 652 個生成、做 25.8 天。
其中:做角色卡(人物三視圖、參考表)6,131 個(9%),做場景板(定場、環境)4,029 個(6%),一般圖 56%,影片 27%。先把人和地方做穩,再拍片。

二、他們怎麼做「卡」:先做圖,再存成卡,拍片時用 @ 叫出來

68 部裡有 50 部在拍片時掛了卡片(reference_elements)。一部片中位數 26 張卡,最多 142 張。卡的命名習慣:char_(角色)、loc_(地點)、prop_(道具)。分類:角色 37,550 次、環境 14,240 次、道具 18,487 次。

角色卡怎麼生:三視圖,同一個人,中性灰底

模型:nano_banana_2 | 片:
Three studio photographs of the same 60 year old East Asian woman arranged side by side on a flat neutral mid-grey studio backdrop, a film character sheet: close-up portrait photo on the left, full-body front photo in the center, full-body back photo on the right, the same real person in all three, consistent across panels. Soft directional cinematic studio lighting from one side, gentle natural shadow falloff, clean neutral cinematic look.

The woman: sixty years old, East Asian features — the 
模型:nano_banana_2 | 片:
Three studio photographs of the same 60 year old East Asian man arranged side by side on a flat neutral mid-grey studio backdrop, a film character sheet: close-up portrait photo on the left, full-body front photo in the center, full-body back photo on the right, the same real person in all three, consistent across panels. Soft directional cinematic studio lighting from one side, gentle natural shadow falloff, clean neutral cinematic look.

The man: sixty years old, East Asian features — the clea
模型:seedream_v5_pro | 片:
Three studio photographs of the same 60 year old East Asian man arranged side by side on a flat neutral mid-grey studio backdrop, a film character sheet: close-up portrait photo on the left, full-body front photo in the center, full-body back photo on the right, the same real person in all three, consistent across panels. Soft directional cinematic studio lighting from one side, gentle natural shadow falloff, clean neutral cinematic look.

The man: sixty years old, East Asian features — the clea
模型:nano_banana_2 | 片:
Edit the character sheet: replace the wardrobe on the figure in all three panels while keeping the face, hair, and body structure of the person in image 1 identical and unchanged.

CHANGE — WARDROBE ONLY across all three panels:
Replace the current olive crewneck and cargo trousers with the outfit from image 2 and image 3:
An extremely oversized deconstructed cotton sweatshirt in heavily sun-faded dark charcoal-brown — the fabric washed and aged to a dusty warm brown-grey, the surface showing la
模型:text2image_soul_v2 | 片:
A film character reference sheet displays three vertically-aligned studio photographs of the same East Asian man in his early twenties. The left panel presents a close-up, straight-on portrait, tightly cropped at the shoulders, with the subject’s lean facial features, sharp cheekbones, ungroomed medium-length straight black hair, sparse stubble, and natural complexion exhibiting visible pores and unretouched skin texture. In the central panel, a full-body front view (from feet to neck) reveals t
模型:seedream_v5_pro | 片:
Three studio photographs of the same 60 year old East Asian woman arranged side by side on a flat neutral mid-grey studio backdrop, a film character sheet: close-up portrait photo on the left, full-body front photo in the center, full-body back photo on the right, the same real person in all three, consistent across panels. Soft directional cinematic studio lighting from one side, gentle natural shadow falloff, clean neutral cinematic look.

The woman: sixty years old, East Asian features — the 

場景板怎麼生:定場鏡頭、環境、同一地點多角度

模型:seedance_2_5 | 片:
SCENE CONTEXT
A V-22 tiltrotor lifts vertically off a wet fogbound airfield apron with nacelles pointing skyward, climbs to altitude, retracts its landing gear, rotates its nacelles from vertical to horizontal, banks and disappears into the fog. No people are visible. One unbroken take.

ACTIVE REFERENCES
<<<27401b20-82d3-492d-9996-4d5841231bf1>>>: military tiltrotor aircraft, matte dark grey airframe, twin engine nacelles at the wingtips, three-bladed proprotors, twin vertical tail fins, four-w
模型:seedance_2_5 | 片:
Ho ridefinito il ruolo del riferimento: ora è dichiarato esplicitamente come **mappa di posizione, non inquadratura**. La camera segue la shotlist, 85mm, campo più stretto. Ho riscritto camera, blocking e luce per il taglio ravvicinato.

---

**SCENE CONTEXT**
A man sits alone on the bench of an empty red locker room, dressing, seen entirely from behind. One shot, twenty seconds, one continuous action, locked-off tight frame on 85mm. Observational documentary realism — this is a man putting on h
模型:seedance_2_5 | 片:
```text
SCENE CONTEXT
A man sits alone on the bench of an empty red locker room, finishing dressing, seen entirely from behind. He closes his tactical vest, bends forward to lace his boots, and pulls on his gloves. One continuous 20-second take on an 85mm prime, the camera creeping forward on an extremely slow dolly in. Observational documentary realism — a man putting on his gear, not a hero suiting up.

ACTIVE REFERENCES
<<<4e88e359-d3ea-4e93-b865-5dde291291c4>>> — CHARACTER: 40yo male, dark m
模型:seedance_2_5 | 片:
```text
SCENE CONTEXT
A man sits alone on the bench of an empty red locker room, finishing dressing, seen entirely from behind. He closes his tactical vest, bends forward to lace his boots, and pulls on his gloves. One continuous 20-second take on an 85mm prime, the camera creeping forward on an extremely slow dolly in. Observational documentary realism — a man putting on his gear, not a hero suiting up.

ACTIVE REFERENCES
<<<4e88e359-d3ea-4e93-b865-5dde291291c4>>> — CHARACTER: 40yo male, medium
模型:seedance_2_5 | 片:
SCENE CONTEXT
A man puts on his uniform alone in a dark red locker room. He pulls on the jacket, closes the vest, pulls on his gloves, and shuts the locker. He never touches a helmet — there is no helmet anywhere in this sequence. Six handheld shots, twenty seconds, shot close on a 50mm, the camera following him throughout.

ACTIVE REFERENCES
<<<4e88e359-d3ea-4e93-b865-5dde291291c4>>>: 40yo male, dark medium-length hair swept back, short dark beard with grey, deep-set tired eyes, fine lines at t
模型:seedance_2_5 | 片:
SCENE CONTEXT
A woman running with a newborn clamped to her chest reaches her apartment door. One hand digs into her jacket pocket for keys while the other arm never leaves the baby. The keys fall. The metallic impact on the vinyl floor makes her freeze and look back down the corridor. The baby begins to cry. She snatches the keys from the floor, unlocks the door, enters, pulls it shut behind her.

ACTIVE REFERENCES
<<<fefe33ff-55d0-4b09-a679-a821ca87e201>>>: 30yo woman, straight dark hair loose

三、拍片時的提示詞長什麼樣(這是最值得學的)

提示詞中位數 2,668 字。5,546 條用「場景脈絡 → 掛卡 → 鏡頭 → 光線 → 聲音 → 不要什麼」的段落結構。掛卡的方式是把卡片 ID 寫進提示詞,並附一行外觀說明,讓模型知道畫面裡誰是誰。
| seedance_2_5
SCENE CONTEXT
A woman sits at a table, waiting for an answer she cannot influence. The shot holds on her hands resting low on the curve of her belly. Her face is out of frame. The hands carry the entire performance.

ACTIVE REFERENCES
<<<f1b0e546-f77e-446b-b62e-ddce709cae6d>>>: 38yo woman, freckled fair skin. White sleeveless wrap vest with a wide ribbed V-panel over a black long-sleeve t-shirt. Seated at the table, torso in strict left-facing profile. Framed from the base of the V-panel down to the top of the thigh — her head, face, chin and throat are entirely above the top frame edge and never enter the shot. Both forearms come together low across the front of her body, exactly as in the hand reference: the near hand (her left, camera side) rests over the lowest curve of the belly, fingers relaxed and slightly curled, pointing screen-left and angled down; the far hand sits just beneath and behind it, partly hidden, supporting from below. A plain metal ring on the ring finger of the near hand. Black sleeve fabric covering both forearms to the wrist; only the hands are bare. 100% matches the reference.

<<<11c38e23-63fe-4c77-8037-2f7645a25948>>>: small residential apartment — grey-green walls, large window screen-left, cold grey residential tower block beyond the glass, dark red curtain at the window edge, pale round table. Geography, materials, palette and light direction only. No device visible in the frame.

LOCATION MAP
<<<f1b0e546-f77e-446b-b62e-ddce709cae6d>>>: seated, torso in strict left-facing profile, occupying screen-center and screen-right. The white vest and t
| seedance_2_5
SCENE CONTEXT
A woman sits at a table, waiting for an answer she cannot influence. The shot holds on her hand resting on the curve of her belly. Her face is out of frame. The hand carries the entire performance.

ACTIVE REFERENCES
<<<f1b0e546-f77e-446b-b62e-ddce709cae6d>>>: 38yo woman, freckled fair skin, light hair. White wrap vest over black t-shirt, wide-leg black trousers. Seated at the table, body in strict left-facing profile. Framed from the base of the sternum to the top of the thigh — her face and head are entirely above the top frame edge and never enter the shot. Her right hand rests across the front of her belly, fingers pointing screen-left, back of the hand toward camera, forearm entering the frame from the lower screen-right corner. 100% matches the reference.

<<<11c38e23-63fe-4c77-8037-2f7645a25948>>>: small residential apartment — grey-green walls, large window in the background, cold grey residential tower block visible through the glass, dark red curtain partially visible at the window edge. Geography, materials and atmosphere only. No device visible in the frame.

LOCATION MAP
<<<f1b0e546-f77e-446b-b62e-ddce709cae6d>>>: seated, torso in strict left-facing profile, screen-center. The curve of her belly reads as a clean silhouette edge against the soft background at screen-left. Her right hand on the belly is the sharpest plane in the frame.
Display: screen-left, off frame, above and forward of the hand. Never visible.
Table edge: lower screen-left, a soft dark horizontal shape at the bottom corner of the frame, out of focus.
Camera: tripod, positioned sc

四、他們的器材設定(Cinema Studio 欄位)

欄位最常用
Camera35mm Film×1362、Fine Film×47、Modern×35、Auto×33
LensClean Sharp×773、Anamorphic×363、Vintage Anamorphic×135、Halation Vintage×67
影片長度(秒)20秒×3345、15秒×2449、30秒×2385、10秒×1736、5秒×1292
畫幅16:9×35747、21:9×15633、3:4×1758、9:16×1233

五、用的模型

用途模型(次數)
全部nano_banana_2×16,316、seedance_2_5×16,189、seedream_v5_pro×9,374、gpt_image_2×7,691、soul_cinematic×2,494、cinematic_studio_soul_location×1,541、nano_banana_flash×1,369、seedream_v5_lite×1,332、text2image_soul_v2×1,046、cinematic_studio_video_4_0×1,003、soul_cinema_studio×962、seedream_v4_5×843、seedance_2_0×840、cinematic_video_editor×486

六、前 20 部逐片

片讚生成數天數角色卡場景板掛卡的鏡頭主要模型
★90828830.9102108nano_banana_flash、seedance_2_5、seedance_2_0
★74039133.1516888nano_banana_2、seedance_2_5、soul_cinematic
★61164120.5174211220seedance_2_5、cinematic_studio_soul_location、nano_banana_flash
★555616431.1359322493seedream_v5_pro、seedance_2_5、nano_banana_2
★54865526.31049622nano_banana_2、nano_banana_flash、seedance_2_unlimited
★45365431.57380489seedance_2_5、seedream_v5_pro、nano_banana_2
★42339418.3342seedance_2_5、seedream_v5_pro、nano_banana_2
★418107830.4135255418seedance_2_5、cinematic_studio_soul_location、text2image_soul_v2
★41035625.8229280seedance_2_5、gpt_image_2、seedance_2_0
★394312526.18941481252seedream_v5_pro、nano_banana_2、seedance_2_5
★37440226.0582224seedance_2_5、seedream_v5_pro
350157931.128549226soul_cinematic、soul_cinema_studio、gpt_image_2
★344154634.811111108nano_banana_2、seedance_2_5、cinematic_studio_video_4_0
★343216824.852103701nano_banana_2、seedance_2_5、soul_cinema_studio
★29782123.935270seedance_2_5、seedream_v5_pro、gpt_image_2
★29513715.2000cinematic_studio_video_4_0、nano_banana_flash、cinematic_video_editor
★2936624.7000seedance_2_5
★28013825.827722nano_banana_2、seedance_2_5、seedream_v4_5
★279110825.878240478cinematic_studio_soul_location、cinematic_studio_video_4_0、seedance_2_5
★265694026.541614nano_banana_2、gpt_image_2、seedance_2_5

七、搬到和心貓治村:照這個順序做

  1. 角色卡(已完成 20 張):他們用真人照生「三視圖同一人中性灰底」。我們的 20 隻動物已有官方照+360 轉身表,下一步用 nano_banana_2 或 seedream 把每隻生成「特寫/全身正面/全身背面」三格參考表,存回卡片(每卡多圖)。
  2. 場景卡(已完成 23 張):他們用 soul_location 或 Cinema Studio 的 Location 風格生「定場+多角度」板。我們用真實照片為底,每個地點補 3 個角度(定場、中景、細節)+一張夜景。
  3. 道具卡(還沒有):他們平均一部片十幾張道具卡。和心村該做:貓食堂餐盤、焚火台、桑拿柴爐、住箱吊床、賣店風鈴、鐵壺、柴燒鍋。
  4. 拍片提示詞用固定段落:場景脈絡 → 掛卡(@角色 @地點)→ 鏡頭(35mm Film+Clean Sharp 或 Anamorphic)→ 光線 → 聲音 → 不要什麼。長度 2,000 字上下是常態。
  5. 影片長度:主流 15 到 30 秒一段,16:9 或 21:9。
  6. 時間預算:一部認真的片約一個月、六百多次生成。駐村 3 天 2 夜做得出預告片,一個月做得出短片。