
seedance-2.5Seedance 2.5 is the long-form and editing tier of ByteDance Seedance, served straight from the official Volcengine Ark API — same channel as Seedance 2.0, no reseller layer and no Volcengine account on your side. It generates 4-30 seconds in a single pass (pass duration: -1 and the model picks the best length itself), accepts up to 50 reference assets in one job — 30 images + 10 videos + 10 audios, each clip 2-30s, with all videos together at most 30s and all audios together at most 30s — and adds two task types the 2.0 line does not have: video editing (replace, add or remove objects, or edit the audio track) and video extension (extend a clip forward or backward, over multiple rounds). An audio-only reference is allowed here; on 2.0 it still needs an image or video alongside it. Output is 480p or 720p — there is no 1080p and no 4K on 2.5, so stay on Seedance 2.0 when you need 1080p — in 21:9, 16:9, 4:3, 1:1, 3:4, 9:16 or adaptive (default), delivered as mp4 (default) or mov (H.264 + yuv444p + PCM, for higher colour fidelity when grading or keying; some players cannot open mov). Native synced audio (voice, sound effects and music), the web search tool and real-person video through the asset:// library work exactly as on 2.0, and prompts are natively multilingual (Chinese, English, Spanish, Indonesian, Malay, Thai, Arabic, Portuguese, Vietnamese, Japanese and Korean). Three constraints are enforced upstream: video editing, video extension and first-frame / first+last-frame jobs all force aspect_ratio "adaptive" (the output keeps the source ratio and cannot be overridden), video editing additionally forces duration: -1 (the output keeps the source clip length and may come out up to 0.4s shorter), and the source video for an edit must itself be 4-30s — Ark reports violations asynchronously as InvalidParameter.TaskTypeConstraint after the task has queued, so our API rejects them at create time and you get an immediate 400 instead. Billing follows real token usage (usage.completion_tokens), charged only on success and refunded on failure, which works out to 480p $0.134/s and 720p $0.300/s for a generation without a reference video. A job that carries a reference video — video editing or video extension — bills on a different basis: (source seconds + output seconds), at 480p $0.080 and 720p $0.180 per second. The lower per-second rate is not a discount, because every second of the source clip is billed alongside the output: editing a clip into an output of the same length costs 20-25% MORE than generating that length outright, so generate rather than route through editing when cost is what matters. Editing a 4.0-second 480p clip that returns 3.7 seconds long comes to (4.0 + 3.7) x $0.080, about $0.62. Credits are frozen at create time against the source clip duration we read from the file itself, and the difference is refunded once upstream reports the real token count.
4-30s from a single call, or duration: -1 to let the model pick the best length — Seedance 2.0 stops at 15s
30 images + 10 videos + 10 audios (2.0 allows 9 + 3 + 3). Each clip 2-30s, all videos together ≤ 30s, all audios together ≤ 30s. Audio-only reference is allowed here.
Replace, add or remove objects and edit the audio track, or extend a clip forward and backward over multiple rounds. The source video for an edit must itself be 4-30s.
Editing, extension and first-frame / first+last-frame all force aspect_ratio "adaptive"; editing also forces duration -1 (output keeps the source length, up to 0.4s shorter). We reject violations at create time with a 400 instead of failing asynchronously upstream.
No 1080p and no 4K on 2.5 — use Seedance 2.0 for 1080p. mov is H.264 + yuv444p + PCM for grading and keying, but some players cannot open it.
Register a face to get an asset:// reference, then use it as a reference or first frame. Raw real-person images are rejected upstream. Use only with consented subjects.
480p $0.134/s, 720p $0.300/s without a reference video, charged only on success and refunded on failure. A reference-video job uses a lower per-token rate but bills input seconds + output seconds, so editing or extending a clip as long as the output costs about 20% more than generating the same length from scratch.
Loading library…
mp4 兼容性最好;mov 是 H.264 + yuv444p + PCM,色彩还原更准,适合调色 / 抠像 / 合成,但部分播放器打不开。
Render synced audio alongside the video
Enrich your prompt with real-time web knowledge before generation (T2V only)
Generated video will appear here
Provide URLs and click Generate
Seedance 2.5 is a Video Generation API provided by bytedance. Seedance 2.5는 ByteDance Seedance 제품군에서 장척 영상과 편집을 담당하는 등급으로, 훠산팡저우(Volcengine Ark) 공식 API에 직접 연결해 제공합니다 — Seedance 2.0과 같은 채널이며 중간 리셀러가 없고, Volcengine 계정도 필요 없습니다. 한 번의 생성으로 4~30초를 만들고(duration에 -1을 넣으면 모델이 알맞은 길이를 직접 정합니다. 2.0은 최대 15초), 한 작업에 참조 에셋을 최대 50개까지 넣을 수 있습니다 — 이미지 30 + 영상 10 + 오디오 10(2.0은 9 + 3 + 3)이며, 각 소재는 2~30초, 영상은 모두 합쳐 30초 이내, 오디오도 모두 합쳐 30초 이내여야 합니다. 2.5에서는 오디오만 참조로 넣는 것도 허용됩니다(2.0은 이미지나 영상을 함께 넣어야 합니다). 여기에 영상 편집(화면 속 사물을 교체·추가·삭제하고 오디오 트랙을 수정)과 영상 연장(앞이나 뒤로 늘리며 여러 차례 반복 가능)이라는 두 가지 작업이 새로 생겼습니다. 출력은 480p 또는 720p입니다 — 2.5에는 1080p도 4K도 없으니 1080p가 필요하면 Seedance 2.0을 쓰세요. 화면비는 21:9 / 16:9 / 4:3 / 1:1 / 3:4 / 9:16과 adaptive(기본값), 포맷은 mp4(기본값) 또는 mov(H.264 + yuv444p + PCM으로 색보정·키잉 시 색 재현이 좋지만 일부 플레이어에서는 열리지 않습니다)입니다. 화면에 동기화된 네이티브 오디오(음성·효과음·배경음악), 웹 검색 도구, asset:// 에셋 라이브러리를 통한 실존 인물 영상은 2.0과 동일하게 동작하며, 프롬프트는 중국어·영어·스페인어·인도네시아어·말레이어·태국어·아랍어·포르투갈어·베트남어·일본어·한국어를 기본 지원합니다. 업스트림 제약이 세 가지 있습니다. 영상 편집, 영상 연장, 첫 프레임 / 첫+마지막 프레임 작업은 aspect_ratio가 무조건 adaptive로 고정되고(출력이 원본 비율을 그대로 따르며 바꿀 수 없습니다), 영상 편집은 duration까지 -1로 고정됩니다(출력이 원본 길이를 따르며 최대 0.4초 짧아질 수 있습니다). 또 편집에 쓰는 원본 영상 자체도 4~30초여야 합니다. 업스트림은 작업이 큐에 들어간 뒤에야 InvalidParameter.TaskTypeConstraint를 비동기로 돌려주기 때문에, 저희 API는 생성 시점에 막고 즉시 400을 반환합니다. 과금은 실제 토큰 사용량(usage.completion_tokens) 기준이며 성공 시에만 청구하고 실패하면 환불합니다. 참조 영상이 없는 경우 480p 초당 $0.134, 720p 초당 $0.300에 해당합니다. 다만 참조 영상이 있는 작업은 토큰 단가는 낮아지지만 입력 초 + 출력 초로 과금되므로, 결과물과 같은 길이의 소재를 편집하거나 연장하면 같은 길이를 처음부터 생성할 때보다 약 20% 더 비쌉니다. 권리와 동의를 확보한 대상에만 사용하세요. Through APIMODELS platform, you can access this model via a unified API with transparent pay-as-you-go pricing. Current pricing: 480p / s: $0.134, 720p / s: $0.3.
Related model: Seedance 2.0 — the previous generation — cheaper and it does 1080p, from $0.092/s
Quickly generate brand promotion videos for ad campaigns and social media marketing.
Create compelling short-form video content for platforms like TikTok, Instagram, and YouTube.
Generate product feature demonstrations and tutorials to improve user conversion.
Produce course explanations, knowledge explainers, and training videos at low cost.
Seedance 2.5 is available through APIMODELS at: 480p / s: $0.134, 720p / s: $0.3. Billing is pay-as-you-go — you only pay for what you generate.
Sign up at APIMODELS, get your API key, and call our unified API endpoint. We provide detailed API documentation with code examples in cURL, Python, and Node.js.
APIMODELS offers the same Seedance 2.5 model through our aggregation platform. We provide a unified API interface so you do not need separate accounts for each provider - one API key to access all models.
한 번의 생성으로 4~30초까지, 이어 붙이지 않고 만들 수 있습니다. duration에 -1을 넣으면 그 장면에 맞는 길이를 모델이 직접 정합니다. 출력은 480p 또는 720p이고, 2.5에는 1080p도 4K도 없습니다. 1080p가 필요하면 Seedance 2.0을 쓰세요. 화면비는 21:9 / 16:9 / 4:3 / 1:1 / 3:4 / 9:16과 adaptive(기본값)이며, 파일 형식은 mp4(기본값) 또는 mov(H.264 + yuv444p + PCM) 중에 고를 수 있습니다. mov는 색보정과 키잉에서 색 재현이 좋지만 일부 플레이어에서는 열리지 않는다는 점을 유의하세요.
한 작업에 최대 50개입니다: 이미지 30 + 영상 10 + 오디오 10(Seedance 2.0은 9 + 3 + 3). 각 소재는 2~30초이고, 영상은 모두 합쳐 30초 이내, 오디오도 모두 합쳐 30초 이내여야 합니다. 2.5는 오디오만 참조로 넣는 것도 되지만, 2.0은 이미지나 영상을 함께 넣어야 합니다. 실존 인물은 여전히 에셋 라이브러리를 거칩니다: 얼굴을 등록해 asset:// 참조를 받은 뒤 참조 이미지나 첫 프레임으로 전달하세요. 실존 인물 사진을 그대로 넣으면 업스트림 심사에서 거부됩니다. 권리와 동의를 확보한 대상에만 사용하세요.
영상 편집은 화면 속 사물을 교체·추가·삭제하고 오디오 트랙까지 수정할 수 있고, 영상 연장은 앞뒤로 늘리며 여러 차례 반복할 수 있습니다. 이 두 작업(그리고 첫 프레임 / 첫+마지막 프레임 작업)은 aspect_ratio가 무조건 adaptive로 고정되어 출력이 원본 비율을 그대로 따르므로, 전달한 비율은 무시됩니다. 영상 편집은 duration까지 -1로 고정되어 결과물이 원본 클립 길이를 따르며 최대 0.4초 짧아질 수 있습니다. 업스트림은 작업이 큐에 들어간 뒤에야 InvalidParameter.TaskTypeConstraint를 비동기로 알려주기 때문에, 저희는 생성 시점에 막고 곧바로 400을 돌려드립니다.
업스트림이 알려주는 실제 토큰 사용량(usage.completion_tokens)으로 과금하며, 성공했을 때만 청구하고 실패하면 환불합니다. Volcengine의 계산식은 tokens = (입력 영상 초 + 출력 초) x 가로 x 세로 x 프레임레이트 / 1024입니다. 참조 영상이 없으면 480p 초당 $0.134, 720p 초당 $0.300에 해당합니다. 참조 영상이 있는 작업은 토큰 단가는 낮지만 출력 초에 더해 입력 초까지 과금되고, 손익분기점은 입력 = 출력 x 0.667입니다. 즉 결과물과 같은 길이의 소재를 편집하거나 연장하면 같은 길이를 처음부터 생성할 때보다 약 20% 더 비쌉니다. 할인 등급이 아니라 계산 방식이 다른 것뿐입니다.
2.5는 더 싼 2.0이 아니라 장척과 편집을 위한 등급입니다: 한 번에 30초, 작업당 참조 에셋 50개, 영상 편집과 영상 연장, mov 출력을 지원합니다. 1080p를 가진 쪽은 2.0이고 가격도 더 쌉니다(480p 초당 $0.092, 720p 초당 $0.197, 1080p 초당 $0.492). 720p끼리 비교하면 2.5가 약 52% 비쌉니다. 긴 호흡의 장면, 이미 찍어둔 영상의 재작업, 참조 소재가 많은 작업에는 2.5를, 1080p가 필요하거나 비용을 아끼고 싶을 때는 2.0을 고르세요. 둘 다 같은 훠산팡저우(Volcengine Ark) 공식 채널과 같은 API 키로 동작하므로, 바꾸는 건 model 문자열 하나면 됩니다.
On APIMODELS, Seedance 2.5 runs alongside 60+ models on one API key and one balance, so choosing is about fit, not lock-in. It supports 4-30s in One Pass、Up to 50 Ref Assets、Video Edit + Extend、480p/720p、mp4 / mov, and you can weigh it on price and capability against other Video Generation models, then switch by changing a single model-name string — no new account or integration. Browse every Video Generation option with live pricing at apimodels.app/models.
Seedance 2.5 supports: 4-30s in One Pass、Up to 50 Ref Assets、Video Edit + Extend、480p/720p、mp4 / mov. See the APIMODELS docs for full parameters and call examples.
Yes. APIMODELS exposes Seedance 2.5 through a single unified API and one key — no separate provider accounts, and no need to handle each provider's regional network access yourself.
We support Stripe (Visa, Mastercard, and other international cards) and Alipay. Credits are available instantly after payment.
Prompts shared by their authors — copy and adapt them. Each one credits its author and links back to the original post.
Seven-shot bedroom candid — cat, phone call, doorbell
Montage, multi-shot candid observational footage. Do not use a single camera angle or continuous take. Handheld documentary style with the feeling of accidental real-life capture. Slightly imperfect framing, subtle handheld shake, tiny reframing adjustments, gentle exposure breathing, and autofocus that settles half a beat late. Realistic skin texture, soft indoor natural light, film grain, shallow depth of field. Relaxed breathing, natural blinking, restrained and authentic performance. Total of 7 shots. The woman from @ Image 1 is wearing a matching cotton pajama set consisting of a scoop-neck sleeveless top and loose shorts made from the same fabric and design. Barefoot, she lies on her stomach across a lived-in, slightly messy bed in her bedroom, browsing a mobile shopping page on a black smartphone. A house cat jumps onto the bed seeking attention, and she affectionately plays with it. Moments later, a friend calls. She answers, chats, and eventually bursts into laughter. The doorbell rings, so she ends the call and gets up to collect her dinner. Shot 1 (0–2s): She is already moving. One thumb scrolls through a shopping page while she frowns slightly, thinking. She quietly mutters in Korean: "아, 뭐 살랬더라…" ("Ah... what was I going to buy again?"). Lips synchronize naturally. Her bare feet gently sway behind her. A low handheld camera glides along the edge of the mattress, with soft bedding partially obscuring the foreground. Shot 2 (2–4s): The house cat jumps onto the bed, compressing the blanket as it walks toward her forearm and lets out a single meow. The camera gives a slight jolt from the impact, then quickly reframes from the phone to an over-the-shoulder view that includes her face, hands, phone, and cat. Shot 3 (4–6s): She turns warmly toward the cat, stroking it once from its forehead to its shoulders. In a gentle, affectionate tone she says: "우리 애기 왔어?" ("Did my baby come?"). Lip sync is clear and natural. The camera pushes in briefly through the cat's softly blurred foreground toward her face and hand. Shot 4 (6–8s): A friend's ringtone sounds. She glances down, confirms the caller, swipes to answer, and brings the phone to her ear. The camera moves in a loose semicircle from a slightly tilted overhead angle, capturing the entire answering motion. Shot 5 (8–11s): Her friend begins talking. She responds first with confusion, then disbelief: "어, 왜? 진짜 거짓말." ("Huh? Why? No way, you're kidding."). Immediately afterward she naturally bursts into laughter. The cat kneads the blanket beside her. A close handheld side angle keeps both her phone and profile within the same focal plane. Shot 6 (11–13s): A clear doorbell interrupts the laughter. Both she and the cat turn toward the bedroom door. She braces one hand on the mattress to get up. The camera reacts a fraction too late, briefly turning toward the door before naturally correcting back to her rising movement. Shot 7 (13–15s): Still on the phone, she hurriedly says: "어, 나 밥 왔다. 끊어." ("My food's here. I'll hang up."). She ends the call with her thumb, gets off the bed, and walks toward the door while the cat follows across the blankets. The camera slides low beneath her elbow and tilts upward as she stands in one continuous motion. Sound: No music. Only raw production sound: bedding rustling, quiet breathing, finger taps on the phone, the soft impact of the cat jumping onto the bed, one natural meow, faint purring, a single short incoming ringtone, call connection tone, the friend's voice through the phone, the woman's four Korean lines with accurate lip sync, a genuine brief laugh, one clear doorbell, the tap ending the call, and soft cat footsteps. No subtitles, no on-screen text, no logo, no watermark. Never render a reference sheet or duplicate the subject.
by oggii_0
Handheld DV morning dog-walk vlog
CAMERA: Handheld DV 16mm daily vlog footage. The video MUST begin with her holding the camera at arm's length in selfie mode while stepping outside her apartment building with her dog. The first 20–30 seconds are entirely handheld. Later she occasionally places the camera on a park bench, low stone wall, picnic table, or the ground for wider shots. Keep subtle handheld shake, drifting composition, autofocus hunting, rushed reframing, uneven zooms, exposure breathing, brief accidental face cropping, and imperfect framing throughout. The camera itself is never visible. LOOK: Warm analog tape texture with gentle film grain, slightly softened sharpness, subtle halation around sunlight, realistic skin tones, low contrast, tiny exposure shifts, and natural motion blur. It should feel like authentic footage from someone's everyday life rather than a polished commercial. STYLE: A relaxed morning lifestyle vlog. Quiet, cozy, and spontaneous. She occasionally laughs at her dog, pauses to look around, adjusts the leash, brushes hair away from her face, and speaks naturally in short sentences with comfortable pauses. CHARACTER: EMMA — a beautiful white woman in her mid-20s. Long light brown hair in a messy ponytail, green eyes, minimal makeup, oversized gray hoodie, black biker shorts, white sneakers, and a small crossbody bag. She is walking a happy golden retriever. SETTING: A peaceful suburban neighborhood on a sunny morning. Tree-lined sidewalks, quiet residential streets, birds singing, a small park with benches, green grass, and soft golden morning light. Very few people are around. SCENES: The vlog opens in selfie mode. Emma holds the camera while leaving her apartment building with the golden retriever excitedly pulling on the leash. "Good morning." She smiles. "Someone couldn't wait." The dog eagerly sniffs everything as they walk down the sidewalk. She laughs quietly. "He has to inspect every single tree." Still holding the camera, she walks into a small neighborhood park. The dog suddenly stops and stares at a squirrel. "Oh... there we go." She smiles and gently shakes her head. She places the camera on a nearby bench for a wider angle while throwing a tennis ball. The dog happily chases after it. "Worth waking up early." She picks the camera back up. Walking slowly through the park, she looks up at the trees for a moment. "It's actually really peaceful out here." The dog returns with the ball but drops it halfway. She laughs. "Close enough." She sits on the bench while the dog lies beside her. She scratches behind his ears. "I think he's happier than I am." She stands up and continues walking. The camera stays in selfie mode as they head toward home. "Coffee is definitely next." She smiles into the lens. "See you later." She gives a small wave and ends the recording.
by maxxmalist
Thirty seconds of hand-to-hand combat in a rain-flooded metro maintenance hall
生成一段完整30秒、16:9横屏、24fps、写实电影级质感的现代近身格斗视频。使用我上传的两组人物参考图:第一名人物固定为"主角",第二名人物固定为"敌人"。严格继承参考图中两人的面部、年龄、发型、体型、身高比例、服装、鞋子、配饰和整体气质。全程不得交换身份,不得变脸、改变服装颜色、改变体型或生成第三名参战者。 【核心风格】 参考《一个人的武林》所呈现的现代硬派武术电影气质:动作迅猛、贴身、凶狠,攻防转换极快,拳脚具有真实接触感和明确受力反馈,但不照搬电影中的人物、场景和具体镜头。以高质量ACT动作游戏的第三人称战斗镜头诠释两人对决,摄影机主要跟随主角,位于主角肩后、腰后或侧后方,使观众产生操控主角迎战强敌的沉浸感;关键格挡和重击时,可以短暂切入侧面中景或近景。 视频不要求真正一镜到底,可以在30秒内进行多次自然切镜,但整场战斗必须像一个连续长镜头般流畅。利用人物身体遮挡、立柱擦镜、快速摇镜、撞击震动和动作匹配完成隐形切镜。切镜后必须延续上一镜的动作惯性、人物位置、身体朝向、伤势和攻击方向,禁止瞬移、换位错误、人物突然恢复站姿或摄影机无理由越轴。 【场景】 深夜暴雨,一座已经停运的高架地铁检修站。站厅呈狭长矩形,地面是被雨水浸湿的深灰色防滑地砖,分布少量浅水洼,能够反射冷白顶灯和红色维修警示灯。左侧是一排金属检票闸机和关闭的卷帘门,右侧是半透明钢化玻璃护栏,护栏外能够看到高架轨道、雨幕、车灯和模糊城市建筑。站厅中段有三根粗大的混凝土立柱,柱脚带黄黑警示条,尽头是一段通往废弃站台的短楼梯。 冷白灯管是主要光源,其中两盏接触不良、偶尔闪烁;红色警示灯在墙面提供少量轮廓光。斜吹进来的雨水、潮湿薄雾、地面积水、立柱、护栏和闸机都要参与战斗。场景空间在所有镜头中保持一致,不得随切镜改变门、立柱、闸机和玻璃护栏的位置。 开场时主角位于站厅中段偏左,背后靠近检票闸机,面朝右前方;敌人位于右前方约三米处,背后是玻璃护栏。前半段战斗向站厅深处移动,中段绕过第二根立柱,后半段反向压回闸机区域。除非通过环绕镜头明确展示换位,否则敌人始终位于主角前方,保持清楚的行动轴线。 【动作原则】 动作由现代实战拳法、肘击、膝撞、低位踢击、近身控制、擒拿拆解和短距离摔法组成。主角冷静、精准、动作幅度小,善于格挡后立即反击;敌人力量更大、进攻性更强,擅长连续压迫和贴身撞击。敌人不是等待挨打的木桩,主角也不能全程无伤碾压。主角必须经历一次明显失势和一次危险化解,最后依靠判断和重心控制夺回主动。 每次进攻都必须呈现起点、运行路径、接触点和结果;格挡必须真正改变攻击方向;受击者的头部、肩膀、躯干、脚步和重心按照力量方向产生反馈。禁止隔空挥拳、拳脚穿透身体、没有接触却自行后退、无故旋转、连续空翻、悬浮和夸张飞行。动作速度快,但关键接触点必须清楚,不得用混乱残影掩盖动作。 【0—4秒:敌人抢攻】 第一帧已经处于战斗中,不进行对视、摆造型或绕圈试探。摄影机位于主角右后方约一米,接近肩部高度,采用轻微广角的第三人称ACT跟随视角。主角位于画面左侧偏中央,敌人从右前方快速逼近。前景是主角潮湿的肩膀和抬起防守的手臂,中景清楚呈现双方身体,背景是玻璃护栏、雨幕和城市灯光。 敌人右脚踩入水洼,水花向后飞散,同时右直拳攻击主角面部。主角小幅向左偏头,使拳锋擦过脸侧,左手从内侧拍开敌人手腕,右前臂立即架住敌人跟进的左摆拳。敌人不停顿,顺势用右肩和身体重量向前撞击。主角被迫后退半步,鞋底在湿地面短暂滑动,但立刻稳住。 敌人抬膝攻击主角腹部,主角双肘向内压住膝部,将冲击引向身体左侧,同时右脚切入敌人支撑腿外侧,为反击建立位置。摄影机跟随主角后退,敌人始终处于正前方。动作快速但不使用慢镜头。 【4—8秒:第一次反击】 敌人的膝部被压开,主角手臂从镜头前掠过形成遮挡,自然切换到两人右侧的平行移动机位。采用略低的中景,清楚展示脚步、髋部和拳肘路径。 主角左手控制敌人的右前臂,右拳从胸前短距离击中敌人右侧肋部。敌人的上半身因受力向左收缩,呼吸短暂中断,但立刻用左肘横击主角太阳穴。主角抬右臂挡住肘尖,前臂在碰撞中明显震动;随后转动肩膀,以左肘短击敌人胸口,再用右脚低位扫踢敌人左侧小腿。 敌人的左脚被扫得横移,但用右脚及时撑住,没有夸张腾空。他抓住主角上臂,将主角猛烈推向第一根混凝土立柱。摄影机随主角急速后退。主角后背接近立柱时用左掌撑住柱面卸力,沿柱边旋转躲避。敌人紧随而来的重拳擦过主角肩膀,砸中柱面,产生沉闷撞击、少量灰尘和疼痛反馈,但立柱不得碎裂。 敌人击中立柱时,镜头产生一次短促撞击震动。主角快速绕到立柱另一侧,摄影机紧贴主角背后绕柱跟随。立柱短暂遮满画面,完成隐形切镜。遮挡结束后,主角位于立柱右侧,敌人从左侧追出,两人沿站厅纵深向楼梯方向移动。 敌人连续使用左右直拳压迫主角。主角边后撤边格挡:右掌向外拨开第一拳,左前臂贴住第二拳。敌人第三次攻击改为右腿中段踢击。主角来不及完全避开,只能收紧左臂保护肋部。踢击真实命中护臂,力量让主角向右踉跄两步,肩膀撞上玻璃护栏。玻璃发出沉闷震颤却没有破碎,护栏外的雨夜灯光因撞击产生短暂拖影。 敌人跨步逼近,左手压住主角肩膀,右肘攻击面部。主角在最后一刻低头,肘部从头顶掠过;主角左手抓住敌人上臂,右手控制手腕,尝试将其压向玻璃护栏。敌人凭借力量强行抽臂,以额头短促撞击主角眉骨。主角头部向后偏移,额角出现轻微擦伤,目光短暂失焦。 【13—17秒:主角失势】 头撞形成明确节奏重音。快速切入主角侧面近景,表现接触瞬间、飞散雨水和真实眩晕,随即拉回能够看清全身动作的中景,不使用长时间慢动作。 敌人抓住主角上身,将他从玻璃护栏方向拽回站厅中央,右腿绊住主角左腿,试图完成向前摔投。主角身体倾斜,在倒地前用右脚重新踩稳,并以左手勾住敌人后颈,双方进入短暂贴身角力。 摄影机从主角侧后方环绕约九十度,明确展示两人换位:敌人转到画面左侧,主角转到画面右侧,主角背后变为第二根立柱和闸机方向。敌人以力量将主角压在立柱边缘,进行一次肩撞和一次短膝击。主角用大腿与前臂挡住主要冲击,但仍因疼痛弯腰。敌人抬起右肘,准备从上方向下砸击主角后背。镜头靠近主角肩后,使抬起的肘部在画面中形成强烈威胁。敌人的右肘快速下砸。主角不能凭空躲避,而是在肘击启动时向敌人身体内侧贴近,使肘尖从背后落空。主角左手托住敌人肘关节,右手控制其手腕,以肩膀作为支点向前旋转,破坏敌人的肩线。摄影机跟随主角完成半环绕,背景中的立柱、闸机和警示灯提供清楚的空间参照。 敌人为避免手臂被锁,顺势转身,以左拳反击。主角放开控制,低头闪过拳头,起身时用右肩撞击敌人胸口,将其推离立柱。主角立即以左直拳攻击敌人面部防守,迫使敌人抬高手臂;随后右拳从手臂下方击中腹部,再用左肘短击锁骨区域。三次攻击快速递进,但必须分别清楚,不能生成多条手臂。 敌人后退两步,脚后跟撞到检票闸机底座,随即用右腿横扫主角腰部。主角向前贴近,压缩踢击距离,以左臂承受大腿近端冲击,同时右手环住敌人腰侧,左脚踏到敌人支撑脚外侧。低机位镜头清楚交代主角如何控制敌人重心。 【22—26秒:环境摔投】 主角转动髋部并向斜前方发力,将失去支撑的敌人摔向闸机。敌人的肩背先撞在金属闸机侧面,闸机挡板被撞开并发出刺耳金属声,随后敌人落在湿地面,水花向外扩散。敌人的身体必须表现真实重量,不能轻飘飘飞行或在空中翻转多圈。 摄影机沿敌人落地方向快速俯冲,在触地时转为略高的斜俯视ACT视角。主角没有停下摆姿势,而是跨过被撞开的挡板继续追击。敌人翻身用右脚蹬向主角膝部,主角侧移避开。敌人借蹬腿动作翻滚起身,左手撑地,右脚踩稳,还未完全站直便挥出右摆拳。 主角从拳头内侧进入,左臂贴住敌人右臂,右掌推击敌人下颌。敌人向后仰头卸掉部分力量,同时左拳击中主角侧腹。主角明显收腹受击,却保持贴身距离,用右膝短促撞击敌人大腿前侧,进一步破坏其稳定性。两人呼吸加重,衣服逐渐湿污,但动作不能突然变得疲软。 【26—30秒:高潮反制】 摄影机回到主角左后方约一米的第三人称ACT跟随位置。主角位于画面中央偏左,敌人在正前方,背景保持被撞开的闸机、闪烁顶灯和红色警示灯。 敌人最后一次猛烈前冲,先用左拳虚晃主角面部,真正攻击是右肘横扫。主角没有被虚招骗出大幅动作,而是用右掌压下敌人左手,在右肘接近时向前踏入攻击内圈,左前臂架住敌人上臂,使肘尖从主角脑后掠过。主角抓住敌人后颈,右脚勾住其前脚踝,以肩膀向前下方施压。 敌人重心向后倒去,却抓住主角衣服试图将主角一同拖倒。主角立刻松开后颈,身体向左转出抓握,以右肘从极短距离击中敌人胸口偏肩位置,再用左掌推开敌人上胸。敌人向后撞在已经打开的闸机挡板上,挡板进一步弯曲。敌人半跪落地,一只手撑住湿地,另一只手仍保持防守,没有昏迷,也没有飞出画面。 主角受惯性影响向前半步,迅速重新稳住重心,双手保持防守,没有背对敌人、庆祝或摆胜利造型。摄影机缓慢移动到主角右侧,以主角肩膀作为前景,将焦点落在半跪喘息、暂时失去进攻能力的敌人身上。红色警示灯闪烁一次,远处传来雷声。最后一帧保持双方继续互相锁定的危险状态,让主角暂时占据优势,但为下一段战斗留下自然衔接。 【摄影、画面与声音】 保持电影级实拍质感与ACT游戏镜头的融合。主要采用肩后跟随、侧后跟随和腰部高度移动机位;关键命中使用短暂侧面中景或近景。摄影机移动快速但具有真实重量,不得漂浮、瞬移、疯狂旋转或持续无规则抖动。撞击时只使用极短、低幅度震动,震动方向与力量方向一致。 动作中允许适量运动模糊,但人物面部、手脚和接触点在关键时刻保持清晰。禁止鱼眼、长时间慢动作、子弹时间、冻结画面、速度线、游戏血条、准星、按钮提示和文字UI。ACT感来自主角中心构图、第三人称跟随、空间推进和及时的动作反馈。 超写实皮肤、肌肉牵拉、衣料褶皱、湿润头发、汗水和雨水反射;真实金属、混凝土、玻璃与湿地砖材质。冷蓝灰色为主色,冷白顶灯作为主光,红色警示灯作为局部轮廓光。高对比但保留暗部细节,高光不过曝,地面反射自然。禁止蜡像皮肤、塑料CG质感、噪点、脏污颗粒、锐化白边和压缩色块。 生成严格同步的现场声音:雨击顶棚、远处雷声、脚踩积水、湿地摩擦、衣料摆动、急促呼吸、拳掌击中皮肉、前臂格挡、肘膝碰撞、身体撞柱、玻璃震颤和闸机金属回响。无需对白。音乐仅保留极其克制的低频脉冲,不能盖过格斗音效。 【负面约束】 禁止身份互换、面孔漂移、服装变化、身体比例突变、多余人物、复制人物、多手多脚、关节反折、身体穿透和人物粘连;禁止隔空受击、动作没有惯性、敌人静止等待、无故滑行、瞬移、悬浮、武侠轻功、气功、能量特效、夸张冲击波和墙体爆炸;禁止主角全程无伤碾压,禁止敌人只会挨打;禁止无意义空镜、英雄登场、长时间对视、重复招式和结尾突然黑屏。 最终效果必须呈现清晰的动作结构:敌人强势抢攻、主角短暂反击、绕柱追击、主角失势、危险拆解、重新夺回主动、利用闸机完成摔投和最终反制。30秒全程高强度、无废秒,刺激而不混乱,凶狠但遵守人体力学,摄影机始终以主角为核心。
by lansenai
We curate copy-ready prompt libraries — every entry shows its full text and a sample result, ready to adapt.
How to get access, regional availability, and how this model compares with its alternatives.