REAL CASE / INPUT → OUTPUT
孔凤春清洁泥膜视频复刻教程
以一条已完成的 26 秒成片为例,逐步说明每一步使用的素材、操作方式和输出结果。
从原片到成片
INPUT SET
准备原始素材
INPUT原片 + 3 类图片



OUTPUT素材职责固定
NEXT参考原片进入剧情与镜头分析;三类图片保留到换图和视频生成阶段。
SOURCE → STORY
看懂原片
INPUT参考原片

OUTPUT6 个必保留节拍
NEXT剧情骨架决定两段分镜各自保留哪些镜头和动作。
STORY → STORYBOARD
拆成两段源分镜
INPUT剧情骨架 + 原片节奏表
OUTPUT两张源分镜


NEXT每张源分镜作为改图的主图,只传递格子、镜头顺序、场景、手位和动作节奏。
STORYBOARD + PRODUCT + PERSON
换人、换产品、去字幕
INPUT每个 Part 的图片顺序
OUTPUT可用于视频生成


Part 1 当前流程改图提示词(完整)
Edit Image A directly into one complete AI-edited 12-panel storyboard for Part 1. Reference roles, in exact input order: - Image A is the source storyboard edit target. It controls only the original 4-column × 3-row canvas, all 12 panel boxes, panel order, camera crop, hand/body placement, indoor bedroom/shelf/wooden-door scene cues, action rhythm, and the existing Shot-label system. - Image A does not control the old host identity or clothing, product identity, product-use object, mud color, subtitles, captions, stickers, or information cards inside the panels. - Image B controls only the exact target product front: a white rounded-square Kongfengchun jar with white lid, green leaf-shaped mark, and real front label `KOPHENIX / 马齿苋净颜清洁泥膜 / PORTULACA CLEANSING MUD FILM`. - Image B does not transfer its white studio background, crop, lighting, or packshot composition. - Image C controls only the open jar and its raised, opaque, milky-white thick clay paste with visible body, ridges, edges, and small peaks. - Image C does not transfer its white studio background, crop, lighting, or packshot composition. - Image D controls only the single male product host: the same face, black hair, age, skin tone, body proportions, and dark short-sleeve shirt in every panel. - Image D does not transfer its tiled bathroom, selfie angle, lighting, towel rack, mirror, or background. Hard role map: - The only person in this Part is one male product host from Image D. - Replace the source host completely, including face, hair, body, and clothing. - Do not introduce a second person or a second version of the host. Hard source-scene rule: - Preserve Image A's original warm indoor bedroom/room scene family, including the blurred shelf, wooden door or wardrobe, neutral wall, and the source camera positions. - The product and identity references are replacement references only; their backgrounds, lighting, framing, and staging must not enter the storyboard. Hard Shot-label rule: - Preserve the black label bars and cyan labels `Shot 01` through `Shot 12` in their original positions. - Keep every label attached to the same panel. Do not delete, rewrite, renumber, reorder, translate, restyle, or move any Shot label. - Remove unwanted text only from inside each panel's video-image area. Shot labels are navigation metadata, not subtitles. Task: - Keep exactly the same source canvas, 12 panel boxes, panel sizes, panel positions, shot order, camera crops, face framing, hand positions, and pacing. - Shot 01–02: close views of the same male host's untreated cheek and nose area, with credible mild oiliness, pores and small closed-comedone texture; a finger points to the problem area. - Shot 03: the host faces camera and applies raised milky-white thick paste to the same cheek/nose area using his fingertip. - Shot 04: the host holds the Kongfengchun white jar near the camera with the front label facing forward and the green mark visible. - Shot 05: one hand supports the open jar while one fingertip lifts a small peak of milky-white thick paste. - Shot 06–09: keep the source face angles and hand positions while the fingertip applies thick, opaque white paste to the nose and cheek problem areas. The paste has visible raised body and edges. - Shot 10–11: the host points to the locally covered nose and cheek areas; keep the source close framing. - Shot 12: the host raises five fingers toward the camera while his face remains partially covered with thick white paste. - Remove all subtitles, dialogue captions, floating Chinese text, ingredient cards, clock stickers, watermarks, and overlays inside the 12 video panels. - In close product panels, preserve legible real product words from Image B rather than blank, smoothed, or invented label marks. Do not: - Do not create a poster, single frame, new collage template, split-panel export, or product packshot. - Do not change the room, scene order, camera crop, body position, panel proportions, or source action rhythm. - Do not squeeze or stretch the person. - Do not create alternate packaging, boxes, extra products, alternate application props, gray/yellow/beige/watery material, or flat lotion-like white smears. - Do not make the pre-wash face already look like a polished after-wash beauty image.
Part 2 当前流程改图提示词(完整)
Edit Image A directly into one complete AI-edited 12-panel storyboard for Part 2. Reference roles, in exact input order: - Image A is the source storyboard edit target. It controls only the original 4-column × 3-row canvas, all 12 panel boxes, panel order, camera crop, hand/body placement, the wash cut and bedroom/shelf/wooden-door scene cues, action rhythm, and the existing Shot-label system. - Image A does not control the old host identity or clothing, product identity, product-use object, mud color, subtitles, captions, stickers, or phone-screen old identity. - Image B controls only the exact target product front: a white rounded-square Kongfengchun jar with white lid, green leaf-shaped mark, and real front label `KOPHENIX / 马齿苋净颜清洁泥膜 / PORTULACA CLEANSING MUD FILM`. - Image B does not transfer its white studio background, crop, lighting, or packshot composition. - Image C controls only the open jar and its raised, opaque, milky-white thick clay paste with visible body, ridges, edges, and small peaks. - Image C does not transfer its white studio background, crop, lighting, or packshot composition. - Image D controls only the single male product host: the same face, black hair, age, skin tone, body proportions, and dark short-sleeve shirt in every live-action panel and in the phone proof image. - Image D does not transfer its tiled bathroom, selfie angle, lighting, towel rack, mirror, or background. Hard role map: - The only live-action person is one male product host from Image D. - The person shown on the phone is an earlier skin record of that same man, not a second identity. - Replace the source host completely, including face, hair, body, and clothing. Hard source-scene rule: - Preserve Image A's source scene changes: Shot 01 is the sink wash action; Shot 02–10 and Shot 12 return to the original warm bedroom/room with blurred shelf and wooden door/wardrobe; Shot 11 remains a brief blurred skincare-bottle B-roll. - The product and identity references are replacement references only; their backgrounds, lighting, framing, and staging must not enter the storyboard. Hard Shot-label rule: - Preserve the black label bars and cyan labels `Shot 01` through `Shot 12` in their original positions. - Keep every label attached to the same panel. Do not delete, rewrite, renumber, reorder, translate, restyle, or move any Shot label. - Remove unwanted text only from inside each panel's video-image area. Shot labels are navigation metadata, not subtitles. Task: - Keep exactly the same source canvas, 12 panel boxes, panel sizes, panel positions, shot order, camera crops, face framing, hand positions, and pacing. - Shot 01: start in the middle of the sink-washing action, with both hands washing the milky-white paste from the host's face; preserve the source hard-cut composition. - Shot 02: close view of the same host checking his clean cheek and nose with one fingertip. The post-wash skin is visibly cleaner and slightly brighter while remaining realistic. - Shot 03: the host wipes one side of his face with a white towel, already in motion. - Shot 04: clean-skin front view in the original room; the host naturally shows the result. - Shot 05: the host holds a phone showing an earlier problem-skin record of the same male identity. - Shot 06: the phone and the Kongfengchun white jar are held beside the host's face; the jar front label faces camera. - Shot 07: tighter phone proof while the host remains partly visible at the edge; keep the source composition. - Shot 08: the host holds the open Kongfengchun jar close to camera; the remaining milky-white thick paste is visibly near the bottom but still has opaque raised paste texture. - Shot 09–10: the host looks at and then presents the same Kongfengchun jar; keep the source hand and face positions. - Shot 11: brief blurred B-roll of several generic skincare bottles, with no readable competing brands. - Shot 12: the host brings the Kongfengchun jar front close to camera for the final product close-up, with the real green mark and readable front product words. - Shot 02–12 must show a clear post-wash progression after the wash proof: the same host's skin becomes visibly cleaner, brighter, smoother, and more even without changing identity, room, camera, or lighting. - Remove all subtitles, dialogue captions, floating Chinese text, stickers, watermarks, phone UI words, competing brand words, and overlays inside the 12 video panels. Do not: - Do not create a poster, single frame, new collage template, split-panel export, or product packshot. - Do not change the source hard cuts, scene order, camera crop, body position, panel proportions, or source action rhythm. - Do not squeeze or stretch the person. - Do not create a second male identity, alternate packaging, boxes, extra products, gray/yellow/beige/watery material, or flat lotion-like paste. - Do not apply the polished post-wash face to the phone's earlier problem-skin record.
NEXT两张替换分镜先进入结构、人物、产品和材质检查。
REPLACED STORYBOARD → APPROVED
质检并锁定分镜
INPUTPart 1 + Part 2 替换分镜

OUTPUT5 项检查通过
NEXT锁定的两张分镜进入口播、参考音频和视频提示词制作。
STORY + SHOTS + SOURCE AUDIO
写口播并切分参考音频
INPUT原片声音 + 剧情骨架 + 分镜
黑头闭口,涂!油皮粉刺,涂!
管它黑头还是油光。孔凤春清洁泥膜。
油皮看好了。四种果酸松动毛孔油脂。
三重矿物泥在外吸附带走。
你就涂在黑头闭口的地方。
涂完之后等五分钟。
直接水洗,脏东西连根拔起。
这个皮肤,女朋友凑再近都不怕。
今天是我打卡这罐泥膜的第三十天。这一罐已经快用空了。
说真的,男生护肤不用那些——瓶瓶罐罐的。
这一罐就够了。
OUTPUT两段可上传音频
NEXT每个 Part 的合格分镜、产品图、人物图、参考音频和台词,一起进入视频生成。
APPROVED ASSETS + PROMPT → VIDEO
分别生成两段视频
INPUT每个 Part 的上传顺序
OUTPUTPart 1 + Part 2
Part 1 实际视频提示词(完整)
参考图角色:
@图片1定义为“分镜板”,只控制 Shot 顺序、场景、景别、动作和硬切节奏;成片不显示分镜网格、边框和 Shot 编号。
@图片2中的产品定义为“泥膜罐”,只锁定白色方圆罐、白盖、绿色叶形标、原生标签和比例;不控制镜头构图和动作。
@图片3中的泥体定义为“乳白厚泥”,锁定颜色、纹理和浓稠质地;不控制镜头构图。
@图片4中的男性定义为“男主播”,锁定脸、黑发、体型和深色短袖;皮肤效果按各Shot执行;不传递参考图背景和构图。
@音频1只参考男主播音色、语速和停顿,不沿用其中的旧产品名;口播只按{}内台词。
生成约13秒、9:16竖屏真实手机短视频。按@图片1的Shot顺序、场景、景别和硬切节奏执行;只补全格间动作,不新增镜头或场景。保持同一男主播和深色短袖。每镜保持可见动作,不生成冻结帧。无字幕、无背景音乐,只保留台词、环境声和动作声;所有Shot场景严格按@图片1执行,不使用@图片4的背景。
0.0–2.2秒|Shot 01–02
画面:Shot 01脸颊黑头闭口极近景→指腹带乳白厚泥真实划过鼻翼并留下白泥→Shot 02下巴油皮粉刺极近景→指腹带乳白厚泥真实划过下巴并留下白泥;四拍依次完成,每拍后硬切。
声音:旁白{黑头闭口,涂!油皮粉刺,涂!}。
音效:<两次短促的泥膜上脸摩擦声>
2.2–4.7秒|Shot 03–04
画面:Shot 03男主播边把乳白厚泥涂到鼻翼和脸颊边说话;硬切Shot 04,手持泥膜罐正面靠近镜头,包装按@图片2,男主播脸和嘴仍可见。
声音:男主播说(Shot 03说前句,Shot 04说完整产品名){管它黑头还是油光。孔凤春清洁泥膜。}。
音效:<厚泥涂抹声>;<罐身轻响>
4.7–7.6秒|Shot 05–07
画面:Shot 05硬切开罐近景,一手托罐,一指挑起乳白厚泥,清楚拉出小泥峰;硬切Shot 06–07,男主播用指腹把乳白厚泥依次涂开在鼻翼和脸颊。
声音:男主播说{油皮看好了。四种果酸松动毛孔油脂。}。
音效:<厚泥分离声>;<指腹涂抹声>
7.6–9.5秒|Shot 08–09
画面:硬切Shot 08–09,男主播继续在另一侧脸颊局部涂抹乳白厚泥。
声音:男主播说{三重矿物泥在外吸附带走。}。
音效:<指腹涂抹声>
9.5–11.4秒|Shot 10–11
画面:Shot 10–11硬切正脸近景,男主播先指鼻翼再指脸颊,指出涂泥位置。
声音:男主播说{你就涂在黑头闭口的地方。}。
11.4–13.0秒|Shot 12
画面:Shot 12硬切男主播正脸,他举起五指表示五分钟,说完后自然放下手。
声音:男主播说{涂完之后等五分钟。}。
Part 2 实际视频提示词(完整)
参考图角色:
@图片1定义为“分镜板”,只控制 Shot 顺序、场景、景别、动作和硬切节奏;成片不显示分镜网格、边框和 Shot 编号。
@图片2中的产品定义为“泥膜罐”,只锁定白色方圆罐、白盖、绿色叶形标、原生标签和比例;不控制镜头构图和动作。
@图片3中的泥体定义为“乳白厚泥”,锁定颜色、纹理和浓稠质地;不控制镜头构图。
@图片4中的男性定义为“男主播”,锁定脸、黑发、体型和深色短袖;洗后皮肤效果按各 Shot 执行,不传递参考图背景和构图。
@音频1只参考男主播音色、语速和停顿,不沿用其中的旧产品名;口播只按{}内台词。
生成约13秒、9:16竖屏真实手机短视频。按@图片1的Shot顺序和原片硬切节奏执行;保持同一男主播和同一深色短袖。Shot 01在水池前洗脸;Shot 02–10和Shot 12回到@图片1的暖色房间;Shot 11只使用@图片1的虚焦静物B-roll。每镜保持可见动作,动作完成后硬切。全片无字幕、无背景音乐,只保留人声、环境声和可见动作对应的音效。
0.0–1.9秒|Shot 01
画面:Shot 01从洗脸中段开始,男主播在水池前双手捧水洗掉乳白厚泥;不补开水龙头等起手动作。
声音:旁白{直接水洗,脏东西连根拔起。}。
音效:<洗脸水声>
1.9–4.4秒|Shot 02–04
画面:Shot 02–04依次硬切:手指检查鼻翼和脸颊→白毛巾擦脸→抬头正脸展示。洗后皮肤明显变得干净透亮、平滑细腻,黑头、闭口、粉刺、油光和粗大毛孔全部消失,肤色均匀,近距离看也呈现完全无瑕疵的状态;保持同一男主播的五官、脸型和身份不变。
声音:男主播说{这个皮肤,女朋友凑再近都不怕。}。
音效:<毛巾擦脸声>
4.4–8.9秒|Shot 05–08
画面:Shot 05–07依次硬切:举手机展示洗前记录→手机与泥膜罐并举→手机记录近景;Shot 08推近近乎见底的开罐泥膜,倾斜罐口展示剩余乳白厚泥。
声音:男主播说{今天是我打卡这罐泥膜的第三十天。这一罐已经快用空了。}。
音效:<手机和罐身轻响>;<罐身轻响>
8.9–10.9秒|Shot 09–10
画面:保持@图片1中的暖色房间和机位。男主播低头看泥膜罐,再抬头看向镜头,把泥膜罐正面贴近镜头;Shot 10结束时产品已经占据画面主体。
声音:男主播说{说真的,男生护肤不用那些——}。
10.9–11.8秒|Shot 11
画面:硬切短促的多瓶护肤品虚焦B-roll,保持@图片1中的原片静物构图和浅景深。
声音:旁白{瓶瓶罐罐的。}。
11.8–13.0秒|Shot 12
画面:硬切回同一男主播和@图片1中的暖色房间,开镜时泥膜罐正面已经贴近镜头,景别与Shot 10一致;男主播稳稳拿住产品直到结束,不重新起手推近。
声音:男主播说{这一罐就够了。}。
音效:<罐身轻响>
NEXT两段生成视频按 Part 1 → Part 2 顺序拼接,再做最终质检。
PART 1 + PART 2 → FINAL
拼接、质检、输出成片
INPUT2 段已生成视频
OUTPUT完整成片

DONE最终输出为一条 720×1280、24fps、带声音、26.19 秒的完整成片。