用于该视频的四张参考图片。

The four reference images used for the video. I generated the whole sequence in a single pass with Seedance 2.0 on Krea AI. It followed the four-shot CCTV structure better than I expected, although the camera switches and time jumps still needed some cleanup. That one generation wiped out the rest of my Krea credits, so I had ChatGPT on the web finish the MP4 with timestamps and signal glitches. So... where are you all using Seedance without going broke? Full prompt below ↓ Create one continuous 15-second, 4:3 landscape video presented as four separate raw analog CCTV recordings from the same Japanese love hotel on the same night. Treat [Image 1], [Image 2], [Image 3], and [Image 4] as a chronological surveillance storyboard. Each image is the sole authority for the camera position, lens distortion, room geometry, lighting, composition, character placement, clothing, hairstyle, body proportions, handbag, and initial physical state of its assigned shot. The four images are separate recorded shots. Do not morph, blend, interpolate, or transform one location into another. GLOBAL CAMERA AND EDITING RULES Every shot comes from a different fixed ceiling-corner security camera, except Shots 3 and 4, which use the same fixed elevator camera at different moments. Every camera remains completely locked. No pan, tilt, zoom, tracking, reframing, handheld movement, or cinematic camera motion. All scene changes are instantaneous direct hard cuts. No crossfades, dissolves, wipes, animated transitions, camera travel through doorways, or architectural transformation. The same adult man and adult woman appear throughout. Preserve their exact clothing, height relationship, hair, body proportions, and handbag from the references. SHOT 1 — 0 TO 4 SECONDS — USE [Image 1] Begin directly from the exact corridor composition shown in [Image 1]. The couple continue walking away from the fixed camera in the existing direction. The woman remains slightly ahead on the right and gently leads the man by the hand. They take only two or three natural walking steps. Her long hair and handbag move slightly with her walking. The man follows at the same pace. They do not turn, stop, look at the camera, or reach another location during this shot. At exactly the end of this shot, make an instantaneous hard cut. SHOT 2 — 4 TO 7 SECONDS — USE [Image 2] Cut directly to the exact elevator-lobby camera position, framing, and composition shown in [Image 2]. The couple stand outside the closed elevator doors in the same left-right arrangement. They remain close together and continue holding hands. Only subtle natural waiting motion: slight breathing, a small shift of body weight, and minimal movement of the woman’s hair and handbag. The elevator doors remain closed throughout this shot. The couple do not turn toward the camera and do not begin walking. At the end of this shot, make an instantaneous hard cut to the camera inside the elevator. SHOT 3 — 7 TO 11 SECONDS — USE [Image 3] Cut directly to the exact fixed elevator-interior CCTV view shown in [Image 3]. Begin with the elevator doors partially open and the couple still completely outside the threshold, exactly as shown in [Image 3]. The elevator doors continue opening to their full width at a normal mechanical speed. The man enters first from the left, and the woman follows from the right while carrying her handbag in her right hand. They take only one or two natural steps into the elevator. Their hands may remain joined. Both people finish just inside the elevator with their arms lowered. No embrace, no raised arms, no kiss, and no turning toward the camera in this shot. At the end of this shot, make a direct jump cut to a slightly later moment recorded by the same elevator camera. SHOT 4 — 11 TO 15 SECONDS — USE [Image 4] Cut directly to the exact fixed elevator composition and later physical state shown in [Image 4]. The elevator doors are now closed. The man has his back toward the camera. The woman already has her arms around his shoulders exactly as shown in [Image 4]. Preserve the handbag’s visible position and do not create extra arms, hands, fingers, straps, or bags. Only minimal natural motion: slight breathing, a very small body sway, and subtle movement of the woman’s hair. Do not develop the action further. No kiss and no additional intimate action. During only the final fraction of a second, create one very brief analog signal tear, then cut immediately to a completely black frame. FINAL RESTRICTIONS No split screen, no collage, no four-panel layout, and no simultaneous display of the reference images. No duplicated people, no extra people, no identity changes, no clothing changes, no disappearing handbag, and no altered room geometry. Do not introduce readable captions, subtitles, logos, camera labels, dates, or timestamp text.

用于该视频的四张参考图片。 我在 Krea AI 上用 Seedance 2.0 一次生成了整个序列。它比我预期的更贴合四镜头 CCTV 结构,尽管镜头切换和时间跳跃仍然需要一些清理。 那一次生成耗尽了我剩下的 Krea 积分,所以我让网页版 ChatGPT 完成了 MP4,并添加了时间戳和信号故障效果。 那么……你们都在哪里用 Seedance 又不会破产? 完整提示词如下 ↓ 创建一个连续的 15 秒、4:3 横版画质视频,以同一晚、同一间日本情人酒店的四个独立原始模拟 CCTV 录像片段呈现。 将 [Image 1]、[Image 2]、[Image 3] 和 [Image 4] 视为一个按时间顺序排列的监控分镜。 每张图片都是其所分配镜头的摄像机位置、镜头畸变、房间几何结构、灯光、构图、人物位置、服装、发型、身体比例、手提包及初始物理状态的唯一依据。 四张图片代表四个独立的录制镜头。不得对位置之间进行形变、混合、插值或变换。 全局摄像机与剪辑规则 每个镜头来自不同的固定天花板角落安保摄像机,但 Shot 3 和 Shot 4 除外,这两个镜头使用同一台固定电梯摄像机,在不同时刻拍摄。 每台摄像机始终完全锁定。不得进行平移、俯仰、变焦、跟踪、重新构图、手持运动或电影化摄像机运动。 所有场景切换均为瞬时的直接硬切。 不得使用交叉淡入淡出、溶解、擦除、动画转场、穿过门口的摄像机移动或建筑结构变换。 同一位成年男性和成年女性贯穿全片出现。需保留参考图中他们精确的服装、身高关系、发型、身体比例和手提包。 SHOT 1 — 0 至 4 秒 — 使用 [Image 1] 直接从 [Image 1] 所示的精确走廊构图开始。 这对情侣继续沿现有方向背对固定摄像机行走。女子始终略微领先在右侧,并轻轻牵着男子的手。他们只自然地走两到三步。 女子的长发和手提包随着行走轻微摆动。男子以相同的速度跟随。 在本镜头期间,他们不会转身、停下、看摄像机,也不会到达另一个位置。 在本镜头正好结束时,进行瞬时硬切。 SHOT 2 — 4 至 7 秒 — 使用 [Image 2] 直接切到 [Image 2] 所示的精确电梯厅摄像机位置、构图和取景。 这对情侣站在关闭的电梯门外,保持相同的左右排列。他们紧靠在一起并继续牵手。 只有细微的自然等待动作:轻微的呼吸、轻微的重心转移,以及女子的头发和手提包的最小幅运动。 电梯门在本镜头全程保持关闭。这对情侣不会转向摄像机,也不会开始行走。 在本镜头结束时,瞬时硬切到电梯内部的摄像机。 SHOT 3 — 7 至 11 秒 — 使用 [Image 3] 直接切到 [Image 3] 所示的精确固定电梯内部 CCTV 视角。 开始时电梯门部分打开,情侣两人仍完全站在门槛外,与 [Image 3] 所示完全一致。 电梯门以正常的机械速度继续打开至全宽。男子先从左侧进入,女子随后从右侧进入,右手拎着她的手提包。 他们只自然地向电梯内走一到两步。他们的手可以保持相握。两个人都刚好停在电梯内部,手臂自然下垂。 本镜头中不得出现拥抱、抬手、亲吻或转向摄像机的动作。 在本镜头结束时,直接跳切到由同一台电梯摄像机拍摄的稍后时刻。 SHOT 4 — 11 至 15 秒 — 使用 [Image 4] 直接切到 [Image 4] 所示的精确固定电梯构图及稍后的物理状态。 电梯门此时已关闭。男子背对摄像机。女子的双臂已经环绕在男子的肩膀上,与 [Image 4] 所示完全一致。 保留手提包可见的位置,不得额外生成手臂、手、手指、背带或包。 只有最微小的自然动作:轻微的呼吸、非常小幅的身体晃动,以及女子头发的细微运动。不要进一步发展动作。不得出现亲吻及任何额外的亲密动作。 仅在最后一瞬间,产生一次非常短暂的模拟信号撕裂,然后立即切到一个完全黑色的画面。 最终限制 不得使用分屏、拼贴、四宫格布局,也不得同时显示参考图片。 不得出现重复人物、额外人物、身份变更、服装变更、手提包消失,也不得改变房间几何结构。 不得引入可读的字幕、说明文字、标志、摄像机标签、日期或时间戳文字。