如何用一张照片制作「首尔的一天」AI 视频:Seedance 提示词与真实案例

从 Pixocto Gallery 的首尔生活影像中拆解人物一致性、2000 年代 DV 质感、30 秒分镜与环境声音,并提供可直接修改的 Seedance 提示词。

如何用一张照片制作「首尔的一天」AI 视频:Seedance 提示词与真实案例

一条好看的「首尔的一天」AI 视频,不是把人物放到南山塔或明洞,再加一层复古滤镜。真正让它像旧录像的,是一些很具体的东西:安静的住宅巷、普通得不能再普通的小事、朋友随手拍摄的机位,以及自动对焦和曝光偶尔犯下的错误。

下面的方法整理自 Pixocto Gallery 中多条 Seedance 成片。你可以上传一张人物照片,复制文中的 Prompt,再替换人物、季节和事件,制作自己的 2000 年代首尔生活录像。

快速答案:这种视频由什么组成?

核心公式是:

  • 固定人物
  • 安静的首尔生活场景
  • 4~6 个微小事件
  • 早期 DV 摄像机缺陷
  • 自然环境声
  • 清楚的时间轴
  • 人物、服装和物体一致性约束
  • 突然切黑的结尾

它看起来像在记录一整天,但真正有效的不是事件数量,而是“有人刚好拿起摄像机,拍下了一段无关紧要的生活”这种感觉。

上传一张照片,开始制作你的「首尔的一天」→

先看一个最接近的 Gallery 案例

首尔老街区夏夜的 2000 年代 DV 家庭录像

查看完整视频和原始 Prompt

这条视频没有宫殿、购物街或宏大的城市全景。故事只是一个女孩在老街区里听见自行车铃、发现一片落叶、试着把叶子放在车座上,最后笑着走回家。

它成立的原因不是剧情复杂,而是每个细节都在帮助观众相信这是一段真实录像:

  • 地点是水泥巷道、旧公寓、盆栽、自行车和电线杆组成的生活区。
  • 摄影机有手持抖动、失焦、曝光变化和偶然变焦。
  • 人物的反应很小,没有为镜头表演。
  • 声音只有脚步、虫鸣、风、树叶和远处交通声。
  • 结尾不是淡出,而是录像中途突然停止。

第一步:选择适合的参考照片

如果希望视频里的主角是自己,优先选择一张清楚、自然、没有重度滤镜的单人照片。

  • 脸部无遮挡,正面或轻微侧脸都可以。
  • 最好能看到上半身或全身,让模型理解服装和身体比例。
  • 日常穿搭比礼服、舞台装和复杂配饰更适合这种题材。
  • 背景不必在首尔,Prompt 会重新建立场景。
  • 不要使用多人合照、过暗照片或已经严重变形的 AI 图片。

在 Prompt 开头明确告诉模型把照片作为人物依据,并列出不能变化的部分:脸、发型、肤色、身体比例、服装和配饰。

Use the uploaded photo as the exact character reference. Preserve the same face, hairstyle, skin tone, body proportions, outfit, accessories, and identity throughout the entire video. No character redesign, wardrobe change, facial change, or identity drift.

在 Seedance 工作台测试人物一致性 →

第二步:选择“生活里的首尔”,不是“景点里的首尔”

韩国青年平凡夏日的一天

查看完整视频和原始 Prompt

Gallery 中这条高热度案例把场景限定在首尔旧住宅区:低层住宅、楼顶平台、外置楼梯、晾衣绳、盆栽、自行车、架空电线和水泥墙。人物坐在楼顶喝饮料、走过小巷、投丢一个篮球,又在突如其来的夏雨里跑回屋檐下。

这类地点比地标更容易产生“记忆感”。如果画面里同时出现南山塔、明洞、景福宫、街头小吃和霓虹商业街,它会更像一条旅行宣传片。

这个首尔旅行视频案例正好展示了另一种方向:竖屏、HDR、顺滑跟拍、城市地标和商业级调色。它适合旅行短片,却不是本文要复现的旧家庭录像。

住宅区版本可以这样写:

Location: A quiet older residential neighborhood in Seoul during a warm summer afternoon. Narrow concrete alleys, low-rise apartments, rooftop terraces, external staircases, potted plants, laundry lines, parked bicycles, utility poles, overhead wires, mature trees, old walls, and a tiny neighborhood shop. Lived-in, peaceful, and authentic. No tourist landmarks, crowds, advertisements, recognizable brands, or polished commercial activity.

第三步:不要把画面做得太漂亮

“复古滤镜”只改变颜色,DV 家庭录像感还来自摄影行为本身。需要同时描述以下特征:

要素建议写法应避免的效果
运镜朋友手持拍摄,轻微抖动,构图偶尔偏掉稳定器、无人机、丝滑环绕
对焦自动对焦短暂寻找,再落到人物脸上全程锐利、电影级焦点控制
曝光走过阳光和阴影时轻微跳动HDR、精致棚拍布光
颜色褪色、低对比、白平衡不完全准确商业调色、高饱和霓虹
画质柔软细节、轻微数字噪点和压缩过度锐化、干净的现代数码画面
机位反应摄影者偶尔慢半拍或不小心切掉主体每个动作都被完美预判

可以直接使用这段镜头风格:

Camera style: Raw footage recorded by a friend using a cheap early-2000s consumer DV camcorder. Natural handheld shake, imperfect framing, autofocus hunting, slight exposure shifts between sunlight and shade, faded colors, soft contrast, mild digital noise, natural motion blur, imperfect white balance, accidental zooms, and subtle camera-handling noise. The camera operator occasionally reacts late or briefly loses focus. No stabilization, gimbal, drone, slow motion, cinematic choreography, dramatic lighting, HDR, or modern commercial color grading.

第四步:用微小事件代替剧情大纲

这类视频最容易犯的错误,是在 30 秒里安排太多地点和戏剧冲突。模型需要同时保持人物、物体、环境和动作连续;每多一个复杂事件,失真的概率都会增加。

更稳的事件通常具有三个特点:一个动作、一个小反应、一个可以听见的声音。例如:

  • 饮料瓶太凉,人物把它贴在脸上,然后笑了一下。
  • 一片叶子落到头发上,朋友指出来,人物才发现。
  • 投篮没有投进,人物摇摇头,把球留在原处。
  • 自行车铃被风吹响,人物回头寻找声音来源。
  • 瓶盖拧不开,第三次终于打开,人物朝镜头露出得意的笑。

雨天首尔公寓厨房里的开瓶小事

查看这个 15 秒室内案例

这条室内视频从头到尾只有“打开一个太紧的玻璃罐”这一件事。尝试、失败、再尝试、成功,已经足够构成一条完整短片。场景越小,人物反应和环境声音越容易显得真实。

15 秒快速版 Prompt

第一次尝试时,不必马上生成完整的一天。先用一个地点和一个小事件测试人物一致性、镜头质感与声音。

Create a 15-second, 16:9 ultra-realistic early-2000s DV home video using the uploaded photo as the exact character reference.

SUBJECT
Preserve the same face, hairstyle, skin tone, body proportions, casual outfit, and identity throughout. Natural skin texture, relaxed expression, no beauty filter, no identity drift or wardrobe change.

LOCATION
A quiet older Seoul residential alley on a warm late-summer afternoon: narrow concrete lane, low-rise apartments, potted plants, bicycles, utility poles, overhead wires, leafy trees, and old walls. No landmarks, crowds, advertisements, or recognizable brands.

CAMERA
Friend-recorded footage from a cheap early-2000s consumer DV camcorder. Handheld shake, imperfect framing, autofocus hunting, faded colors, soft contrast, slight exposure shifts, mild digital noise, natural motion blur, imperfect white balance, and one awkward accidental zoom. No stabilization, gimbal, drone, slow motion, cinematic lighting, HDR, or modern commercial color grading.

00:00–00:04 — The subject walks slowly down the lane holding a cold bottled drink. A breeze moves their hair and clothes. The camera briefly struggles to focus.

00:04–00:08 — A dry leaf lands on the subject's shoulder. They do not notice at first. The camera operator quietly laughs.

00:08–00:12 — The subject notices the leaf, removes it, looks toward the camera with a mildly embarrassed smile, and laughs naturally.

00:12–00:15 — The subject continues walking, turns back for a tiny wave, then looks away. The recording abruptly cuts to black mid-motion. No fade-out.

AUDIO
Natural location sound only: footsteps, distant traffic, summer insects, leaves, wind, bicycle bell, neighborhood voices, and subtle camera-handling noise. No music, narration, subtitles, captions, logos, watermarks, or artificial sound effects.

REALISM
Believable walking, hand movement, hair, clothing, leaf physics, and foot contact with the ground. No distorted hands, extra fingers, duplicated people, teleportation, floating objects, CGI appearance, exaggerated acting, or sudden transformations.

复制快速版 Prompt,先生成一条 15 秒视频 →

30 秒完整版 Prompt

确认人物和风格稳定后,再扩展为 30 秒。下面这版使用六段连续的小事,适合 Seedance 2.5。

Create a 30-second, 1080p, 16:9 ultra-realistic early-2000s DV home video showing an ordinary summer afternoon in the life of the person in the uploaded reference photo. The footage should feel spontaneous, intimate, imperfect, and genuinely observed rather than performed.

MAIN SUBJECT
Use the uploaded photo as the exact character reference. Preserve the same face, hairstyle, skin tone, body proportions, casual outfit, shoes, bag, accessories, and identity from beginning to end. Natural skin texture, minimal styling, relaxed expressions. No character redesign, wardrobe change, facial change, or identity drift.

LOCATION
A quiet older residential neighborhood in Seoul: narrow concrete alleys, low-rise homes, rooftop terraces, external staircases, potted plants, laundry lines, parked bicycles, utility poles, overhead wires, mature trees, old walls, and a tiny neighborhood shop. Authentic and lived-in. No crowds, tourist landmarks, advertisements, recognizable brands, or polished commercial activity.

CAMERA STYLE
Raw personal footage recorded by a friend using a cheap early-2000s consumer DV camcorder. Natural handheld shake, imperfect framing, autofocus hunting, exposure shifts between sunlight and shade, faded colors, soft contrast, mild digital noise, natural motion blur, imperfect white balance, accidental zooms, and subtle camera-handling noise. The camera operator sometimes reacts late or briefly loses focus. No stabilization, gimbal, drone, slow motion, cinematic choreography, dramatic lighting, HDR, or modern commercial color grading.

00:00–00:05 — LEAVING HOME
The subject steps out of an old apartment, checks their keys, adjusts their bag, notices the camera, gives a small smile, and starts walking down the lane. The camera takes a moment to find focus.

00:05–00:10 — THE COLD DRINK
At a tiny neighborhood shop, the subject buys a cold bottled drink. Outside, they press the bottle briefly against their cheek because of the summer heat and laugh at the camera operator's awkward zoom.

00:10–00:15 — A SMALL INTERRUPTION
While walking through a quieter alley, a bicycle passes. The subject steps aside naturally, waits, and continues. A breeze moves their hair and clothes. They glance back but do not pose.

00:15–00:20 — THE LEAF
The subject sits on a low wall beside a small apartment garden. A dry leaf lands in their hair. The camera operator points it out. The subject removes it and gives the camera an embarrassed, amused look.

00:20–00:25 — SUMMER RAIN
A light summer shower suddenly begins. The subject looks toward the sky, laughs, protects the drink with one hand, and walks quickly toward a covered entrance. Show believable rain, wet concrete, soft reflections, damp hair, and realistic foot contact.

00:25–00:30 — GOODBYE
Under the apartment entrance, the subject catches their breath, wipes rain from their forehead, gives a small genuine smile, and says, “See you tomorrow.” They walk inside. The camera remains pointed at the closing door for one second, then abruptly cuts to black mid-motion. No fade-out.

AUDIO
Natural environmental audio only: footsteps on concrete, distant traffic, birds, summer insects, leaves, bicycle wheels, shop refrigerator hum, bottle opening, light rain, water dripping from rooftops, quiet neighborhood voices, and subtle camera-handling noise. Dialogue should sound casual and local. No music, narration, soundtrack, subtitles, captions, logos, watermarks, or artificial sound effects.

CONTINUITY AND PHYSICAL REALISM
Maintain believable anatomy, movement, object permanence, clothing, hair, rain interaction, reflections, and environmental continuity. Hands and fingers remain natural. The drink stays a separate physical object and never intersects with the face or hand. No duplicated people, distorted anatomy, teleportation, disappearing objects, floating objects, sudden transformations, or CGI appearance.

FINAL FEEL
Like a forgotten personal recording of an ordinary summer day. Not a commercial, fashion film, music video, or travel advertisement. Quiet, warm, nostalgic, spontaneous, and deeply human. The charm comes from small moments and the feeling that the camera happened to be there.

使用 Seedance 2.5 制作完整 30 秒版本 →

有朋友或伴侣时,怎样保持互动自然?

首尔老城区与朋友的夏日 DV 生活记录

查看完整视频和原始 Prompt

多人视频不要只写“他们开心互动”。应该给每段互动一个可以执行的小动作,例如递饮料、指出头发上的叶子、躲开经过的自行车、挥手道别。

另一个韩系情侣 Vlog 案例把照相亭、便利店、街机和牵手压进 30 秒,信息量更大,也更考验人物一致性。第一次做多人版本时,建议把地点控制在一到两个,并反复写明两人的外貌、服装和关系不变。

Keep the same two people throughout. Preserve both faces, hairstyles, body proportions, outfits, accessories, and relationship continuity. Their chemistry should feel familiar and unstaged. Use small physical interactions—passing a drink, briefly blocking the camera, pointing out a leaf, sharing a laugh—rather than exaggerated romantic poses. No face swapping, duplicated people, wardrobe changes, or sudden changes in height and age.

可替换模板:把首尔改成任何城市

不要只替换城市名称。真正决定地域感的是住宅形态、街道物件、天气、声音和人物会做的小事。

Create a [15/30]-second [ASPECT RATIO] ultra-realistic early-2000s DV home video using the uploaded photo as the exact character reference.

Subject: [AGE / APPEARANCE / OUTFIT]. Preserve the same identity, face, hairstyle, proportions, outfit, and accessories throughout.

Location: A quiet older residential neighborhood in [CITY] during [SEASON / TIME / WEATHER]. Include [LOCAL HOUSING], [STREET MATERIAL], [PLANTS], [UTILITY DETAILS], [SMALL LOCAL SHOP OR DOMESTIC SPACE], and [DISTANT AMBIENT ACTIVITY]. No tourist landmarks, crowds, recognizable brands, or advertisements.

Camera: Raw friend-recorded early-2000s consumer DV footage—handheld shake, imperfect framing, autofocus hunting, exposure shifts, faded colors, soft contrast, mild digital noise, imperfect white balance, accidental zooms, and natural camera mistakes. No stabilization, gimbal, drone, HDR, or polished cinematic grading.

Timeline:
00:00–[TIME] — [ONE SIMPLE ACTION].
[TIME]–[TIME] — [SMALL INTERRUPTION OR DISCOVERY].
[TIME]–[TIME] — [NATURAL REACTION].
[TIME]–[END] — [QUIET GOODBYE OR WALKING AWAY], followed by an abrupt cut to black.

Audio: Natural location sound only—[CITY-SPECIFIC AMBIENCE], footsteps, distant traffic, wind, birds or insects, room tone, and subtle camera-handling noise. No music, narration, subtitles, logos, or artificial sound effects.

Realism: Consistent identity and environment, natural anatomy, believable object physics, subtle acting, and no teleportation, duplication, distortion, or sudden transformations.

替换城市和事件,生成自己的版本 →

例如,把首尔改成台北时,不只是把 Seoul 改成 Taipei,还可以加入骑楼、旧公寓、机车声、饮料店冰箱声和潮湿夏夜;改成东京时,则可以换成低层住宅、狭窄街道、自行车棚、电车远声和便利店门铃。

常见失败原因

1. 提示词同时要求“旧录像”和“电影大片”

cinematic lightingHDRperfect compositionsmooth gimbal movement 会与 DV 家庭录像方向冲突。想要旧录像感,就要明确排除这些效果。

2. 30 秒塞进太多地点

公寓、地铁、明洞、宫殿、市场、汉江全部出现,会让视频变成景点合集,也增加人物漂移和场景跳变。优先让故事发生在同一街区。

3. 只描述情绪,不描述动作

“她度过了快乐的一天”无法指导镜头。把它改成可以被拍摄的动作:“她喝了一口太酸的果汁,皱起脸,然后朝镜头笑。”

4. 把所有瑕疵都去掉

失焦、慢半拍、构图偏移和曝光变化不是错误,而是这类视频的真实感来源。画面越干净,越容易显得像现代广告。

5. 忽略物体连续性

如果人物一直拿着饮料、纸袋或雨伞,要在一致性段落里单独锁定它。否则物体可能换手、变形或突然消失。

6. 给自然录像配满背景音乐

虫鸣、脚步、冰箱嗡鸣、瓶盖声和雨落在水泥地上的声音,比配乐更能建立空间。可以先生成自然声音,发布到社交平台时再决定是否添加音乐。

在 Pixocto 中生成

  1. 打开视频生成工作台,选择 Seedance。
  2. 上传一张清晰的单人参考照片。
  3. 第一次先用 15 秒快速版确认人物和视觉方向。
  4. 选择 16:9 可以更接近家庭录像;面向 Reels、TikTok 或 Shorts 时也可以改成 9:16。
  5. 需要完整的一日叙事时,使用支持 30 秒的 Seedance 2.5。
  6. 开启音频生成,保留自然环境声。
  7. 如果人物已经稳定,再替换地点、天气和微小事件,不要每次重写全部约束。

这类视频不依靠宏大剧情。让人物做一件普通的小事,让摄影机偶尔跟不上,让声音里保留风、脚步和远处的生活,一段可信的记忆就开始出现了。

现在上传照片,用 Seedance 开始生成 →