AI 陷阱 👇
分享链接 · https://openimages.ajiang.me/p/2098249012369641790/

camera_lens · motion-blurcamera_lens · telephotocomposition · dynamic-diagonalcomposition · eye-levelhuman_attributes · long-hairlighting · natural-daylight
提示词(原文,逐字保留)
AI pitfalls 👇
Seedance 2.5: 1 render ~$5.55
Naming a move doesn't beat what the model has seen most.
I wrote "one-foot slalom." It skated every cone on two feet instead. The term lost to the model's prior. Fix (untested): describe the position, not the move - "only the right skate down, left skate lifted off the floor."
A term is a request. The model answers with what it's seen most.
Prompt 👇
15-second, 16:9 landscape, hyper-realistic live-action short. A young woman, Alessia, runs one continuous real-time speed-slalom inline skating race, filmed by a single locked static camera. Use @reference as her locked visual identity - preserve her face, low ponytail, body proportions and outfit exactly across every shot, never reinterpret or let her face turn generic. Matte natural skin. Outfit, identical every frame: a powder-blue fitted cycling-style short-sleeve top with a short front zipper pulled open partway to the lower chest, in thick opaque lined fabric with no protruding detail; a cream pleated skort at mid-thigh with a wide high elastic waistband, worn over fitted cream undershorts of the same colour that fully cover her upper thighs so nothing is exposed when the skort lifts; blue ribbed socks; white inline speed skates, low boot, single wheel frame, pink accents.
She is a focused competitive skater racing at full speed. The technique is fast one-foot slalom throughout, at racing pace: she clears the whole cone line balanced on a single skate, gliding and carving on one foot while the other leg lifts and crosses over, foot-to-foot, only one wheel frame on the floor at a time. The weaving is rapid and sharp, cones flicking past quickly under her single skate, an urgent high-speed race rhythm, never a slow careful balancing act. This is real fast one-foot slalom, never a two-footed squat over the cones.
FIXED CAMERA, never moving, never panning, never zooming: one locked static camera low near eye level, set at an angle to the lane. The line of small black slalom cones recedes on a strong diagonal from the far upper-left down toward the near lower-right foreground, real vanishing-point perspective, never a flat head-on view. She starts far back up the diagonal and skates down the cone line toward the near foreground, growing larger in frame only because she gets closer. Her travel direction is constant, the 180-degree line held.
Real ordinary-scale indoor arena: glossy pale reflective floor, a plain deep-blue back wall, a small digital race timer at the side. A sparse casual mixed crowd stands well behind low barriers, static and unchanging, standing still watching, no one walking through, no phones raised filming, no individual singled out or in focus, kept small and soft in the far background, never a packed stadium. Ordinary indoor sports-hall lighting only, plain and even, nobody specially lit, no dramatic beams, no god-rays, no studio or beauty light. Face shadows stay dark, do not fill.
Full-speed explosive real-time skating, never slow motion, never speed ramp: hard motion blur, ponytail streaming, the pleated skort lifting and fluttering from the speed and the deep crouch. Multi-beat coverage all from the one fixed camera as she races down the line.
0-2.5s | She is far up the diagonal in a still side-on ready stance, knees bent, weight loaded, a long empty run-up of open floor ahead of her before the first distant cone. The timer triggers and she launches, driving hard fast first strides across the empty floor toward the start of the cone line.
2.5-7s | She hits the cones and enters fast one-foot slalom, mid-distance, carving the line rapidly on a single skate, the other leg crossing over quickly, torso folded low in a deep compressed racing crouch, hips and glutes pushed back by the crouch, skort flared up, cones flicking past under her one skate at speed.
7-11s | Closer and larger, still fast one-foot slalom, low and urgent down the line, deep lean, rapid sharp foot-to-foot rhythm on the single support skate, clearly closing distance quickly toward the near foreground.
11-13s | LONG-LENS CLOSE on her face in the near foreground, chest-up, still driving fast and low through the cones: effort raw on her face, jaw set, eyes forward, hair flying. Her face large and sharp. This is the identity-lock beat.
13-15s | She clears the very last cone at speed and breaks into a finishing flourish right at that cone, the last cone still beside her skate in frame: she drops into a deep one-legged squat gliding on the single skate, the other leg extended, both arms flung wide open and up in a big graceful confident spread, chin up. Hold this spread-armed one-legged pose over the last cone as the final frame.
Every beat shows clear cause and real physics: real edge grip, real lean, real body weight over one skate, real racing speed. Single continuous real-time take, cuts decisive.
AUDIO: all sound captured in-camera by a single fixed camera, raw, never a studio mix, never dubbed. Only the hard rhythmic clatter and roll of inline wheels on the hard floor, the scrape of the carving skate, her sharp breathing and effort, a very faint sparse ambient murmur with no cheering. Kept low and real. No music, no BGM, no cinematic sound design, no effect sound.
Avoid: subtitles, text overlay, watermark, anime, 3D-CG, plastic skin, over-beautifying, two-footed squat over the cones, both skates on the floor through the cones, slow careful balancing, tentative slow weaving, skort exposing underwear, upskirt exposure, bare thigh exposure under the skort, slow motion, speed ramp, freeze, moving camera, panning camera, zooming camera, head-on symmetrical view, feet-first opening, sprinter's block start, background crowd walking around, phones filming, music, BGM, crowd roar, crowd cheering, commentary, effect sound, boom.
中文译文
AI 陷阱 👇
Seedance 2.5:1 次渲染约 5.55 美元
给动作命名,拗不过模型最常看到的样子。
我写了"单脚障碍滑"。它却用双脚滑过每一个锥筒。术语输给了模型的先验。修正(未测试):描述姿势,而不是动作——"只有右轮滑着地,左轮滑离开地面。"
术语是一种请求。模型会用它最常见的内容来回答。
提示词 👇
15 秒,16:9 横屏,超写实真人短片。一名年轻女性 Alessia 进行一场连续实时速度障碍轮滑比赛,由一台锁定不动的固定摄像机拍摄。使用 @reference 作为她锁定的视觉身份——在每一帧都精确保留她的脸、低马尾、身材比例和服装,绝不重新诠释,也不让她的脸变得普通。自然哑光肌肤。服装,每一帧完全一致:粉蓝色修身骑行风格短袖上衣,短拉链拉到下半胸,厚实不透明内衬面料,无突出细节;奶白色百褶短裙至大腿中部,宽高弹腰带,内穿同色紧身打底短裤,完全覆盖大腿上部,这样短裙掀起时不会暴露任何东西;蓝色罗纹袜子;白色直排轮速滑鞋,低帮,单轮架,粉色点缀。
她是一名专注的比赛滑手,全速滑行。整个过程是快速单脚障碍滑,比赛节奏:她用单只轮滑鞋保持平衡滑过整条锥筒线,一脚滑行和切弯,另一条腿抬起交叉,脚到脚切换,一次只有一只轮架着地。穿梭快速利落,锥筒在她单只轮滑鞋下快速掠过,呈现紧迫的高速比赛节奏,绝不是缓慢、小心的平衡动作。这是真正的快速单脚障碍滑,绝不是在锥筒上方双脚蹲行。
固定摄像机,绝不移动,绝不摇摄,绝不变焦:一台锁定不动的固定摄像机,低位接近眼高,与赛道成角度设置。小黑色障碍锥筒的排列沿强对角线从远端左上角向近端右下前景延伸,真实消失点透视,绝不是平直的正对视角。她从对角线远端出发,沿锥筒线滑向近端前景,在画面中变大仅仅是因为她更近了。她的行进方向保持不变,180 度直线保持。
真实普通规模的室内场馆:浅色光滑反光地板,纯深蓝色背景墙,侧面小型数字比赛计时器。一小群稀疏、随意、混合的人群站在低屏障后方,完全静止不变,站着观看,无人穿行,没人举手机拍摄,没有个体被突出或清晰聚焦,在远背景中保持小而柔和,绝不是拥挤的体育场。仅使用普通室内体育馆照明,朴素均匀,没人被特别打光,没有戏剧性光束,没有光柱,没有摄影棚或美容光。面部阴影保持深色,不要提亮。
全速爆炸式实时滑行,绝不慢动作,绝不变速:强烈运动模糊,马尾飘扬,百褶短裙因速度和深蹲而起舞飞扬。多节拍画面全部来自同一台固定摄像机,记录她沿直线飞驰而下。
0-2.5s | 她在对角线远端,以静止侧身预备姿势站立,膝盖弯曲,重心压低,在第一个远端锥筒前有一段长长的空旷滑跑地面。计时器触发,她启动,以强硬快速的前几步跨过空旷地面,冲向锥筒线起点。
2.5-7s | 她切入锥筒,进入快速单脚障碍滑,中距,在单只轮滑鞋上快速切弯线路,另一条腿快速交叉,上半身低伏于深压比赛蹲姿,臀部和臀部因蹲姿而向后推,短裙扬起,锥筒在她单只轮滑鞋下高速掠过。
7-11s | 更近更大,仍然是快速单脚障碍滑,低姿态紧迫沿线下行,深倾,单支撑轮滑鞋上快速利落的脚到脚节奏,明显快速接近近端前景。
11-13s | 在近端前景用长焦特写她的脸,胸部以上,仍快速低姿态穿越锥筒:脸上呈现原始的用力感,下颌紧咬,眼睛向前,头发飞扬。她的脸大而清晰。这是身份锁定节拍。
13-15s | 她高速滑过最后一个锥筒,并在该锥筒处爆发完成动作,最后一个锥筒仍在她轮滑鞋旁边留在画面中:她下蹲成深深的单腿滑行姿态,滑在单只轮滑鞋上,另一条腿伸展,双臂大幅度张开向上舒展成优雅自信的姿态,下巴抬起。在最后这个锥筒上方保持这个张开双臂的单腿姿势作为最后一帧。
每个节拍都呈现清晰的因果关系和真实物理:真实的轮缘抓地、真实的倾斜、真实的体重压在单只轮滑鞋上、真实的比赛速度。单一连续实时镜头,切点果断。
音频:所有声音由一台固定摄像机原位采集,原始录制,绝不是摄影棚混音,绝不后期配音。只有硬质地面上直排轮的硬朗节奏性撞击声和滚动声、切弯轮滑鞋的刮擦声、她急促的呼吸和用力声、极其微弱稀疏的环境低语,没有欢呼声。保持低调真实。没有音乐,没有背景音乐,没有电影感音效设计,没有效果声。
避免:字幕、文字叠加、水印、动画、3D-CG、塑料皮肤、过度美化、在锥筒上双脚蹲行、穿越锥筒时双脚着地、缓慢小心平衡、试探性缓慢穿梭、短裙露出内裤、裙底走光、短裙下裸露大腿、慢动作、变速、定格、移动摄像机、摇摄摄像机、变焦摄像机、正面对称视角、脚先开场、短跑起跑器起跑、背景人群四处走动、手机拍摄、音乐、背景音乐、人群吼叫、人群欢呼、解说、效果声、吊杆话筒。
来源与署名
原文由 @markh_app 发布在 X:查看原推文。 本页逐字保留原文并提供机器翻译的中文解读;版权归原作者所有。
同工具的更多提示词
- Seedance 2.5 提示词 标题: 致命战斗中的自拍 时长: 15 秒 画幅: 16:9 风格: 高品质日本动画电影,流畅的 2D 动画,极其流畅的角色…@tokyo_Valentine
- 制作一个30秒的超写实电影质感音乐视频,三位年轻成年歌手在夜晚霓虹闪烁的城市中演唱一首情感丰富的现代歌曲。视频应呈现高预算专业音乐视频的质感,具备真实的人物、…@ChillaiKalan__
- 角色:@[image1]。视觉风格:暗黑幽默的哥特式故事书插画,与参考图保持一致——风化的石头、苔藓般的绿色、风暴般的灰色天空、绘画般的质感,以及诡异而温暖…@airina_xyz
- SEEDANCE 2.5 — 动作捕捉 / 角色替换提示词 创建一个 11 秒、4:3 的动作捕捉视频,要求口型精准同步。 核心目标 仅将 @vid1 中位于…@Korgamiai
- 以图片 #1 作为男人、狗狗以及整体美术方向的唯一视觉参考。保留男人极度瘦高的剪影、苍白狭长的脸庞、凌乱的黑色头发、黑色外套以及笨拙而忧郁的肢体语言。保留那只…@aimikoda
- GM 😃 星期六! 祝大家周末愉快——这次附带一个挑战 🔥 希腊神话——或其他。 (任何工具) ⭕️ 以前从没做过——但我 挑战‼️ 你来创建这个场景 🟧 🟧…@SarantuyaSimone