跳到主要内容
指南 · Key · 2026-08-09

如何从参考视频整理 AI 视频提示词

对照 35 秒原片,阅读五段完整提示词,再看一个新场景的核对与改编过程。

AI 视频提示词描述你希望模型生成的镜头:主体、动作、环境和镜头选择。参考视频能让这些选择更具体。对编剧和 AI 短剧创作者来说,关键是判断保留什么、修改什么,而不是试图找回看不见的原始提示词。

1.阅读从参考视频到提示词的完整案例

本案例使用《Tears of Steel》原片 00:20–00:55 的节选,包含火箭升空、桥上交谈和转入未来实验室。先播放这段 35 秒素材,再读五段已保存的提示词,最后跟随下方的原片核对和人工改编。素材使用 CC BY 3.0 授权,播放器旁保留影片署名与许可链接。

提示词是 VidBreak 已保存的实际输出,没有为了展示效果而修饰原文。这是一个案例,不是准确率基准测试。后面的改编和丝带练习都是人工编写的文字,尚未通过视频生成实测。

参考原片00:35

Tears of Steel · 35 秒片段
展示的提示词范围:00:04.5–00:14.5

Tears of Steel,(CC) Blender Foundation · CC BY 3.0。原片已截取;本站选取并翻译部分输出供阅读。 来源 · 许可

这里是已保存英文分析的中文翻译。请播放原片核对;完整结果保留英文原文。

真实保存输出摘录 · 中文译文

主体与场景

Thom 在画面左侧侧身倚着金属栏杆;Celia 站在右侧,身体部分朝向他,机械右臂显眼地搁在栏杆上。

动作 · 原片 00:04.5–00:10

Thom 倚着栏杆,转向 Celia 解释时轻轻用双手比画;Celia 听着,疑虑逐渐加深。

镜头 · 本段前半部分

硬切到平视中景双人镜头,沿运河桥栏杆横向构图,带轻微自然机位起伏。

声音 · 本段前半部分

清晰的前景男性对白,伴随轻柔的运河水声和远处城市街道环境声。

查看完整 5 段提示词 →

保存分析中的五个时间段

以下时间从这段 35 秒素材的起点算起,沿用保存分析中的近似分段,不是已核准的切镜点。简短概述由编辑整理;展开任意一段,即可阅读完整英文原始提示词。

时间段 / 原片范围保存的提示词描述了什么
第 1 段
00:00–00:04.5
火箭从工业城市的天际线上方升起。
第 2 段
00:04.5–00:14.5
Thom 与 Celia 在运河桥上交谈,Celia 举起机械手。
第 3 段
00:14.5–00:20
两人的争执继续,包含手势和 Thom 的回应。
第 4 段
00:20–00:27.5
场景转到未来实验室与记忆回放。
第 5 段
00:27.5–00:35
技术人员操作回放设备。

保留实际保存的英文输出,不在原文中修正或补写。下方核对与改编是单独的编辑工作。每段原文内部的时间都从该段的零秒重新计算。

继续看原片核对与修正 ↓

完整原始提示词 1 · 00:00–00:04.5
4.5-second Cinematic live-action sci-fi establishing footage featuring realistic matte textures, atmospheric depth, and sharp architectural silhouette definition. Overcast futuristic industrial cityscape with dark steel launch gantries and residential rooftops framing a vertical rocket ascent under diffuse, cool daylight. Classic European brick gables form the lower foreground and middle ground, framing an expansive industrial launchpad where a rocket ascends centrally in the hazy distance. The sequence conveys an ominous sense of cold separation and technological scale, avoiding vibrant saturated fantasy tones, rapid camera zooms, or abrupt atmospheric shifts. Distant low-frequency industrial rumble and atmospheric wind accompany a clear, resentful female voiceover delivery. Subject: Rocket and Cityscape: Distant multi-stage rocket ascends vertically between dark metallic launch scaffolds set against a sprawling urban horizon of European rooftops. Propelling steadily upward with a concentrated pillar of bright white exhaust expanding into the cloudy atmosphere. 0-2 seconds: Static establishing wide shot positioned from an elevated vantage overlooking the skyline and launch facility. The distant rocket continues its vertical climb as dense white exhaust plumes expand outward between the flanking industrial scaffolds. Distant low industrial hum and gentle atmospheric wind ambience across the skyline. Rocket reaches the middle-upper frame with its bright exhaust column trail centered between the launch towers. 2-4.5 seconds: Continues the static wide shot without camera movement or reframing. The rocket ascends toward the upper edge of the frame into the overcast haze while the exhaust glow illuminates the distant cityscape. Celia (angry): "You're a jerk, Thom." Clear foreground female voiceover delivery over the sustained low-frequency rumble of the distant rocket. Rocket enters the high clouds as the exhaust illumination fades across the distant skyline. Camera and sound requirements: Retain only the evidenced spoken voiceover dialogue in the timeline and forbid any added dialogue or narration. Preserve the subtle low ambient electronic synth drone without introducing melodic score. Retain only the distant industrial rumble and gentle atmospheric city wind acoustics at balanced low levels. Maintain a completely locked-off, stationary wide shot without pans, tilts, shakes, or zoom movements. Maintain a continuous, uncut establishing shot lasting the full beat duration. Forbid any added subtitles, watermarks, title cards, or digital text overlays. Lock the overcast lighting, industrial scaffold geometry, city rooftop silhouettes, and exhaust plume trajectory. Avoid rapid camera moves, stylized futuristic bloom, or oversaturated fantasy color grading. Preserve the cold, solitary mood and underlying interpersonal tension rated at an intensity of 50.
完整原始提示词 2 · 00:04.5–00:14.5
10-second Cinematic live-action footage with natural daylight rendering, realistic skin textures, and shallow depth of field along an urban canal. Amsterdam canal bridge with dark metal railings under soft overcast daylight, surrounded by green foliage, calm canal water, and historic brick buildings in the background. Thom stands on the left in profile leaning against the metal railing, while Celia stands beside him on the right facing partially toward him, her mechanical right arm resting prominently along the rail. The scene captures escalating interpersonal friction and awkward defensiveness, strictly avoiding melodramatic overacting, comical expressions, or artificial CGI lighting flares. Natural outdoor city ambience combined with conversational dialogue and distinct mechanical servo whirring from the prosthetic arm. Person: Thom: Standing on the left side of the bridge frame, a slim young man with dark wavy hair wearing a dark casual jacket over a patterned t-shirt and jeans. Leaning against the bridge railing with both hands, turning his upper body slightly toward Celia as he speaks. Earnest yet evasive facial expression with shifting eyes and an uneasy, conciliatory mouth. Celia: Standing on the right side of the frame, a young woman with reddish-brown hair in a ponytail, dressed in a sporty blue and black vest and sporting a detailed metallic cybernetic right arm. Resting her mechanical hand on the railing before shifting her weight and raising the robotic hand toward Thom. Frustrated, skeptical gaze with narrowed eyes and a tight, confrontational jaw. 0-5.5 seconds: Hard cut to a medium two-shot at eye level, framed horizontally across the canal bridge railing with slight organic camera breathing. Thom leans on the railing, gesturing subtly with his hands as he turns toward Celia to plead his case, while Celia listens with growing skepticism. Thom (pleading): "Look, Celia, we have to follow our passions. You have your robotics, and I just want to be awesome in space." Clear foreground male dialogue accompanied by gentle canal water ripples and distant urban street ambience. Thom finishes speaking with his hands resting on the railing as Celia begins turning directly toward him. 5.5-10 seconds: Continues the medium two-shot with a subtle slow push-in, maintaining eye-level perspective on both characters against the foliage background. Celia turns her torso toward Thom, lifts her metallic cybernetic arm from the rail, and holds the robotic hand directly into his view while delivering her challenge. Celia (confrontational): "Why don't you just admit that you're freaked out by my robot hand?" Crisp female dialogue accompanied by subtle metallic servo clicks and hums as the robotic arm lifts. Celia holds her robotic hand raised toward Thom while Thom recoils slightly with an uncomfortable expression. Camera and sound requirements: Retain only the evidenced spoken dialogue between the characters in the timeline and forbid any added dialogue or narration. Forbid any background musical score, preserving pure naturalistic dialogue and diegetic sound. Preserve natural canal water acoustics, distant city ambience, and subtle mechanical servo Foley from the robotic arm. Maintain a steady, subtle handheld-style medium two-shot with smooth, restrained reframing. Maintain a continuous, uncut two-shot take throughout the full duration of the beat. Forbid any on-screen subtitles, logos, watermarks, or artificial text overlays. Lock character wardrobe, prosthetic arm mechanical construction, bridge railing position, and diffuse daylight color temperature. Avoid exaggerated robotic transformations, fast cinematic pans, or dramatic color grading shifts. Maintain a sustained tone of interpersonal frustration and defensiveness rated at an intensity of 70.
完整原始提示词 3 · 00:14.5–00:20
5.5-second Cinematic live-action shot maintaining crisp daylight rendering, natural urban color saturation, and realistic shallow depth of field. Outdoor Amsterdam canal bridge featuring dark wrought-iron railings under diffuse cool daylight, with lush green tree foliage and historic canal architecture softly blurred in the background. Thom stands in the left middle ground in profile facing right, while Celia stands in the right middle ground facing him with her cybernetic arm held between them. The beat delivers heightened emotional exasperation and awkward romantic confession, strictly avoiding cartoonish body language, exaggerated physical comedy, or high-tech HUD overlays. Rapid, emotionally charged spoken dialogue punctuated by mechanical prosthetic servo hums and soft city ambience. Person: Thom: Standing on the left, a slim young man with dark wavy hair wearing a dark casual jacket and jeans, positioned in profile facing right. Stepping back slightly from the prosthetic arm before throwing both hands into the air with exasperated gestures. Flustered and embarrassed expression with widened eyes, tense brow, and an agitated, open-mouthed confession. Celia: Standing on the right, a young woman with reddish-brown hair in a ponytail wearing a blue and black vest with a metallic cybernetic right arm. Standing firmly upright with her robotic right hand extended steadily toward Thom throughout his confession. Stern, resolute gaze with narrowed eyes and an unyielding, vindicated facial expression. 0-2.2 seconds: Continues medium two-shot framing at eye level, subtly reframing as Thom leans backward. Thom hesitates, stammering and glancing nervously down at Celia's outstretched cybernetic hand while trying to formulate an excuse. Thom (hesitant): "I'm not freaked out by it's..." Hesitant, stammering male speech over subtle background foliage rustle and distant canal ambience. Thom pauses with his head tilted back and his lips parted, poised to break into exasperation. 2.2-5.5 seconds: Maintains eye-level medium two-shot with subtle handheld stabilization as the argument peaks. Thom throws his hands up in exasperation, waving his arms emphatically while shouting his confession, then lowers his shoulders in defeat as Celia remains motionless. Thom (exasperated): "All right, fine! I'm freaked out. I'm having nightmares that I'm being chased by these giant robotic claws." Loud, impassioned male vocal delivery with slight acoustic resonance in the open air, accompanied by faint rustling clothing Foley. Thom lowers his hands to his sides while Celia maintains her unwavering glare and raised mechanical hand. Camera and sound requirements: Retain only the evidenced dialogue spoken by Thom in the timeline and forbid any added dialogue or narration. Forbid any musical score, maintaining strictly diegetic dialogue and environmental ambience. Preserve natural canal breeze, distant city traffic, and subtle mechanical servo noise from the prosthetic arm. Keep a stable medium two-shot with organic eye-level framing tracking character gestures. Maintain an uncut, continuous take throughout the argument's peak. Forbid any on-screen text, subtitles, watermarks, or graphical overlays. Lock character positions, wardrobe, prosthetic hand details, and natural daylight consistency. Avoid unnatural CGI limb effects, stylized jump cuts, or melodramatic camera zooms. Preserve peak defensiveness and raw exasperation scored at an intensity of 75.
完整原始提示词 4 · 00:20–00:27.5
7.5-second Cinematic sci-fi interior footage with high contrast, glowing holographic vector lines, deep shadow tones, and industrial gothic architectural details. Sprawling, dimly lit ruined cathedral converted into a high-tech laboratory with heavy cables, metallic gantries, glowing orange and cyan holographic displays, and diffuse overhead spotlighting. Glowing orange and blue wireframe holographic avatars stand in the close foreground, while a massive circular mechanical core with hanging cables dominates the vast background hall. The atmosphere is somber, monumental, and tinged with technological melancholy, strictly avoiding sterile clean-room visuals, modern office computer equipment, or colorful fantasy glow effects. Resonant metallic room echoes and electronic data hums accompany a reverberant, processed audio playback of the past breakup. Subject: Hologram Memory Avatars: Pair of glowing orange and cyan wireframe human figures standing in the center foreground amidst dark laboratory machinery. Re-enacting the bridge confrontation through floating vector point-cloud silhouettes. 0-2.5 seconds: Hard cut to a close-up tracking shot on the floating glowing wireframe figures amidst dark machinery. The glowing wireframe holographic models of Thom and Celia gesture dynamically in sync with the recorded memory of their past argument. Subtle digital data glitch pulses and deep low-frequency room tone. Holographic figures continue gesturing as the camera begins pulling back smoothly into the wide lab space. 2.5-7.5 seconds: Smooth slow crane pull-back expanding into a wide establishing shot of the massive cathedral laboratory. The camera glides backward across the cavernous dark space, revealing sprawling machinery and cables while centered title text displays prominently. Celia (cold): "Whatever, Thom. We're done." Echoed, processed female voice playback reverberating through the large metallic cathedral space accompanied by a rising synth pad. The camera settles on the wide cavernous laboratory view as the on-screen title text remains centered. Camera and sound requirements: Retain only the evidenced reverberant playback dialogue in the timeline and forbid any added dialogue or narration. Preserve the slow-building, atmospheric ambient synthesizer pad score. Maintain deep cavernous room reverberation, subtle electrical hums, and holographic projection pulses. Execute a smooth, steady crane pull-back from close holographic detail to wide establishing scale. Maintain a smooth, continuous sequence shot transitioning from close holograms to the wide laboratory view. Preserve the prominent bold white capital title text '40 YEARS LATER' centered on screen. Lock the dark gothic laboratory architecture, suspended mechanical core, cable clutter, and orange-blue holographic graphics. Avoid generic clean white sci-fi aesthetics, contemporary computer monitors, or bright sunlight. Maintain a somber, monumental sense of elapsed time and melancholic nostalgia rated at an intensity of 65.
完整原始提示词 5 · 00:27.5–00:35
7.5-second Cinematic sci-fi footage featuring floating translucent holographic user interfaces, shallow depth of field, and industrial practical console textures. Cluttered futuristic laboratory workstation filled with analog dials, patch cables, and glowing green and magenta holographic UI panels under moody directional overhead lighting. The technician sits centered behind an industrial control console in the middle ground, framed by translucent floating graphic interfaces in the foreground and blurred lab machinery in the background. The scene conveys calm technical precision and focused anticipation, strictly avoiding erratic handheld camera shaking, cartoonish interface animations, or modern consumer electronics. Crisp tactile switch Foley, digital interface audio confirmation tones, and clear spoken technical status dialogue. Person: Technician: Seated in the center of the control desk, a young technician with curly brown hair, round metal-framed glasses, and a dark hooded casual jacket. Hands actively manipulating physical dials and tactile toggle switches on the cluttered control console while gazing at the floating holographic readouts. Concentrated, professional facial expression with focused eyes behind glasses and a slight nod of confirmation. 0-2.5 seconds: Cut to a tight close-up on the control console surface and technician's hands adjusting dials beneath floating holographic graphics. The technician's fingers turn analog knobs and flip metallic switches as green and magenta interface elements update in response. Crisp tactile switch clicks, mechanical dial rotations, and subtle digital interface chirps. Console switches settle into locked positions as holographic indicators flash confirmation. 2.5-7.5 seconds: Smooth cut to a medium shot through translucent floating HUD graphics, framing the seated technician looking upward. The technician completes the console calibration, looks up toward the main holographic readout, and clearly announces the status confirmation. Technician (focused): "Robots' memories synced... and locked." Clear spoken dialogue from the technician supported by a sustained low electronic synth drone and positive interface tone. The technician remains seated looking at the console as floating green text 'SIMULATION READY' illuminates the frame. Camera and sound requirements: Retain only the evidenced spoken line from the technician in the timeline and forbid any added dialogue or narration. Preserve the subtle, low electronic pulse building quiet tension without intrusive melodic themes. Preserve tactile console Foley, switch clicks, and electronic interface audio cues at clear spatial levels. Maintain stable, controlled framing with smooth transitions between close console details and medium subject coverage. Execute a clean two-shot progression from close hardware manipulation to medium character status delivery. Preserve floating holographic HUD interface text including 'SIMULATION READY' and system telemetry readouts. Lock the technician's wardrobe, glasses, console dial arrangements, and green-magenta holographic overlay styles. Avoid exaggerated cartoonish computer graphics, high-energy action camera moves, or flat brightly lit laboratory environments. Maintain a calm, focused technical atmosphere rated at an emotional intensity of 60.

2.使用之前,先核对保存分析

本次示范检查了覆盖整段素材的若干画面,包括第 4.7、7、10、12 和 14 秒。这是画面抽查,没有逐帧核准全部内容,也没有独立核验音轨。上方原始提示词保持不变,修正在自己的工作稿中进行。

要素保存分析与原片核对修正或取舍
分段边界保存分析把第二段起点记为 4.5 秒,但 4.7 秒的画面仍是火箭;到第 7 秒才已看到桥上人物。不把 4.5 秒当作已核准的切镜点。本练习采用约 7–14 秒的桥上段落;实际剪辑时另行确认精确切点。
人物与动作7–14 秒的抽帧显示两人在栏杆旁转向彼此,Celia 举起机械手。保留双人互动;下方的新人物、手势和工作间属于主动创作的改动。
镜头原文同时写了镜头呼吸、缓慢推进和连续双人镜头。静态抽帧无法核实完整运镜。新场景只选固定机位的平视双人中景。这是人工简化,不冒充从静帧得出的运镜纠错。
对白与声音原文包含引述对白、水声、城市环境声和机械声,本次画面核对未独立验证这些声音。改编中去掉来源对白和声音清单;需要复用声音细节时先重新听原片。
主观解释原文把情绪强度写为 70。这是生成解释,不是从素材测出的数值。换成看得见的动作与克制表情。

有用的成果是一份能检查、能修改的初稿。无法确认的观察应标记待查或省略,不要把它写成看似精确的制作指令。

3.比较两种人工练习提示词

另用一条系在码头栏杆上的红丝带练习。这不是《Tears of Steel》的场景;插画为概念图,以下文字未经生成模型实测。

红丝带概念插画,并非视频实测成果

已有概念插画 · 人工练习

模糊的初稿

海边美丽而富有情感的红丝带,史诗级电影运镜,惊人细节。

人工文生视频草稿

近景:一条红丝带系在风化的木码头栏杆上。松散的一端被海风抬起,扭转一次,再落回木头旁;绳结始终系在原处。阴天清晨的柔和光线,背景是海面,机位固定。

这段话明确主体、动作和构图,不把全景、特写、暴风雨和日落塞进同一个短镜头。

人工图生视频草稿

丝带松散的一端被微风抬起,扭转一次,再落回栏杆旁;绳结始终系在原处,机位保持固定。

起始图片已经明确主体、光线和构图时,可以用更短的动作描述。查阅实际使用模型的说明:时长、参考图和画幅可能是单独控件,也并非所有模型都支持声音。

4.改写一个场景,不照搬所有细节

从第二段的双人交流出发。下表记录了具体取舍,便于对照原始输出,看清新提示词改了什么。

原始提示词要素新场景的选择原因
Thom、Celia、运河桥与机械手工作台旁的两位同事使用自己的人物和不同环境。
十秒、对白与多个动作五秒内完成一次转身与回应缩小需要修改和测试的动作范围。五秒是人工设定的目标,不是量出的原片时长。
镜头呼吸与缓慢推进固定机位、平视双人中景只保留一个明确的镜头要求。
引述对白与具体来源声音这一稿不规定对白和声音避免把尚未核准的来源声音带入新场景。
精确情绪分数及大量排除项表情克制,两人始终在画内描述可见行为,只保留少量约束。

人工改写练习 · 未进行视频生成测试

五秒场景。固定机位、平视中景双人镜头:柔和日光下,两位同事站在安静的工作台旁。左侧的人转向另一人,抬起一只手;右侧的人与其对视,短暂停顿。两人始终留在画面中,表情克制,保持同一个连续场景。

新人物、工作间、时长和动作取舍均为人工改动。这一稿尚未在视频生成器中运行。请按实际模型支持的时长与输入设置执行测试并记录结果;仅凭这段文字,不能认定模型兼容或能够精确复现。

5.剪辑计划与一次生成分开

参考视频可能从全景切到特写,再切到反应镜头。把它记成分镜计划,按工具限制分别生成或拍摄,最后剪到一起。十分钟的详细还原是规划材料,不意味着模型会在一次请求里按每个时间点执行。

6.一次只调整一个问题

看到的问题先尝试怎样修改
想要固定机位,镜头却在动。去掉冲突运镜,明确机位固定。
丝带从栏杆上脱落。减小动作,并重申绳结保持系住。
几乎没有动作。明确松散端及动作,检查起始图是否容许运动。
镜头中途换了环境。去掉转场,本镜头只留一个场景。
动作正确,但外观不合适。保留动作文字,只改一项光线或风格细节。

这些是实验,不是保证有效的修复。把提示词、模型、设置和结果放在一起,才能知道改了什么。

7.把这张练习表用在下一个场景

参考视频及选取范围:

观察到的主体与环境:

要保留的一个动作:

构图与运镜:

需要生成或留作剪辑备注的声音:

必须保持一致的细节:

新人物/环境(人工改动):

需要核实的问题:

实测后的模型、设置与结果:

8.从自己的参考视频提取提示词

用 VidBreak 准备文字初稿,再对照原片检查。支持的公开链接可只读预览前 15 秒;上传或完整任务需要登录,AI 提示词可选择最多 10 分钟的范围,开始前先看预计积分。

需要了解如何导入参考视频、选择范围或核对提取结果?先看从视频提取提示词的完整操作指南,再回到本页完善提示词结构和措辞。

用自己的参考视频试试视频转提示词

想更细地研究镜头,可继续阅读逐镜头分析指南或提示词成果使用说明。

逐镜头分析 · 成果使用说明

常见问题

每条提示词都应该包含声音吗?

只有目标模型支持所需声音时才加入;否则保留作剪辑备注或另行制作。

提示词越长越好吗?

不是。增加能消除歧义的细节即可;冲突动作和多次场景切换反而让一个镜头变得模糊。

可以把整份还原直接贴给生成模型吗?

把它当成规划材料,拆成模型支持的镜头或片段,逐条按目标模型限制调整。

这些示例已经通过 AI 视频生成测试了吗?

本页包含 VidBreak 对《Tears of Steel》的已保存分析,以及另行编写的提示词练习。保存的分析展示了 VidBreak 的文字输出;人工练习不是生成视频实测。当前页面没有证明这些草稿在某个生成器中的实际效果,也不保证精确复现。请按目标模型的要求调整,并测试你自己的结果。