MiniMax H3 牛仔写真 45 秒连续短片教程：一张图，三段生成，完整衔接

这套案例的核心思路不是一次让 H3 生成 45 秒，而是把一个完整的 45 秒牛仔写真短片拆成 3 个独立的 15 秒段落。每一段生成时都重新输入同一张 Picture 1，因此人物身份、服装和视觉基础始终由同一张参考图锁定，再通过三个 15 秒提示词分别控制不同阶段的动作、摆拍与镜头语言。

H3 的 Ref2VA 结构本身就是 subject_definitions → summary → retention_analysis → detailed_description → overall_soundscape → non_diegetic_music 六个部分，并要求镜头按播放顺序组织。(GitHub)

这套 45 秒设计可以理解成：

第 1 段：人物建立 + 玩枪动作秀 → 第 2 段：写真摆拍 + 身体线条 → 第 3 段：近距离写真 + 局部特写 + 最终定格

三个片段虽然分别生成，但实际上属于同一支完整视频。

使用方式：每次只上传同一张人物参考图作为 Picture 1。第一次使用提示词 1，生成 0–15 秒。第二次仍然上传同一张 Picture 1，使用提示词 2，生成 15–30 秒。第三次仍然上传同一张 Picture 1，使用提示词 3，生成 30–45 秒。后期按照顺序拼接即可。第一、二段提示词中的人物锁定方式和六段式结构来自当前教程素材。

第一段：0–15 秒

这一段承担的是角色出场和第一轮动作高潮。

不是舞蹈，而是具有明显动作节奏的牛仔时尚摆拍：

全身建立 → 拔枪 → 手腕动作 → 低机位强化 → 收枪 → 最终姿势

subject_definitions: <Subject 1> is the adult female cowboy shown in <Picture 1>. Her exact facial identity, hairstyle, body proportions, cowboy outfit, accessories, boots, hat, color palette, and overall visual appearance come exclusively from <Picture 1>. She is the only human character in the entire video. <Picture 1> is the sole visual identity reference for <Subject 1>. summary: [reference generation] Create a 15-second cinematic western fashion-action introduction for <Subject 1>. The clip is designed as a stylish character performance and pose showcase rather than a narrative western scene. The visual rhythm progresses from stillness to controlled firearm-handling action, reaches one clear physical climax, then settles into a strong fashion pose that naturally becomes the starting body position for the next 15-second dance segment. Preserve the character and visual identity from <Picture 1> throughout. retention_analysis: <Subject 1> (appears throughout): fully_preserved - preserve her exact facial identity, hairstyle, body proportions, cowboy clothing, hat, boots, accessories, skin appearance, and overall character design from <Picture 1>. <Picture 1> (appearance reference): fully_preserved - use only as the identity and appearance reference for <Subject 1>; do not reinterpret, replace, merge, duplicate, or redesign the character. detailed_description: The target video is a premium live-action western fashion editorial with a bold, controlled, confident attitude. The cinematography combines western cinema composition with modern high-fashion action photography: deliberate framing, restrained camera movement, strong body silhouettes, realistic skin and fabric texture, warm late-afternoon sunlight, slightly muted western colors, subtle film grain, and natural contrast. The camera movement must always serve the character's pose and action rather than compete with it. <Subject 1> remains the only person throughout the entire 15 seconds. [Shot 1] The video opens with a medium-wide full-body western fashion composition of <Subject 1> standing firmly in the environment established by <Picture 1>. Her complete cowboy silhouette, hat, boots, outfit, and holster are clearly readable. She begins almost completely still, with her weight settled onto one leg. After a brief pause, she lowers her chin slightly, touches the brim of her cowboy hat with one hand, then lets that hand descend naturally toward the holster. Her expression remains calm, confident, and slightly defiant. The camera performs a slow Truck Right with small amplitude, keeping her full figure stable in frame. The movement is smooth and restrained, creating an elegant character introduction rather than an action shot. [Shot 2] At 00:03.000, without changing the environment, the camera cuts to a medium cowboy-style composition framing her from approximately the head to mid-thigh, keeping her face, gun hand, holster, and upper body clearly visible. The camera remains almost static. <Subject 1> suddenly draws the revolver from the holster with one clean, decisive motion. She immediately rotates her wrist once in a controlled flourish and catches the grip securely, keeping the firearm pointed safely downward and away from the camera. She pauses for a fraction of a second in the completed pose. Her free hand briefly touches the brim of her hat. The contrast between the quick hand action and the nearly locked camera creates the main visual impact of this beat. [Shot 3] At 00:06.000, the shot cuts to a medium-full three-quarter view from slightly below eye level. <Subject 1> takes one deliberate step forward. As the boot lands, she lowers the revolver beside her thigh, settles her shoulder downward, slightly shifts her hip, and raises her eyes toward the camera. The camera performs one short Arc Shot with small amplitude around her front-left side at normal speed, revealing the change from side profile to a powerful three-quarter fashion pose. This is the single major camera movement of the action climax. The motion must remain smooth and controlled, never becoming a circular spin around the character. [Shot 4] At 00:09.000, the shot changes to a tighter medium portrait-fashion composition. <Subject 1> calmly returns the revolver toward the holster and completes the holstering action. Once the gun is secure, she slowly adjusts the brim of her hat and turns her face slightly toward the camera. Her expression becomes cooler and more self-assured. The camera performs a very slow Push In with small amplitude while maintaining a stable horizon and shallow depth of field. The camera movement is intentionally slower than the previous action, allowing the energy to settle into a premium fashion-photo moment. Her skin remains natural and realistic, with subtle texture rather than artificial smoothing. [Shot 5] At 00:12.000, the camera transitions to a medium-full composition that clearly shows her legs, hips, shoulders, and complete upper-body silhouette. <Subject 1> finishes the holstering gesture, plants one foot slightly forward, lowers her center of gravity, and angles one shoulder forward. Her free hand rests naturally near the waist while the other remains close to the holster. She keeps her eyes toward the camera. The camera begins a very slow Pull Out with small amplitude, gradually giving her more space in the frame. During the final second, she makes only a subtle rhythmic weight shift through the shoulders and hips, stopping on a strong asymmetrical pose that is physically suitable as the opening position of the next dance segment. Do not have her walk away, turn her back, leave the frame, raise the weapon, or perform another large action. End on this stable pose with the body already carrying a slight forward rhythmic readiness, creating a clean visual and physical handoff into the following dance video. Throughout the entire sequence, keep exactly one woman, one consistent identity, one consistent outfit, and one consistent firearm. The firearm must never duplicate, disappear, teleport between hands, change shape, or become detached from the character. The firearm is used only as a controlled fashion-action prop: no firing, no aiming at the viewer, no aiming at another person, no combat, no gunfight, and no violent action. Avoid exaggerated gun spinning, stunt choreography, cartoon western gestures, excessive slow motion, random dancing, runway walking, excessive camera rotation, drone shots, 360-degree camera movement, rapid zooms, or unnecessary cuts. The final frame must remain visually compatible with the beginning of the next dance segment. overall_soundscape: Natural western outdoor ambience with a light dry wind, subtle movement of fabric and leather, quiet boot contact with the ground, restrained holster and firearm handling sounds, and a subtle environmental presence. The firearm handling sounds should remain realistic and understated. No dialogue and no vocal performance. non_diegetic_music: A modern western fashion beat at approximately 96 BPM, built from dry electric guitar, muted acoustic guitar accents, restrained bass, tight low percussion, and sparse rhythmic hits. The rhythm begins sparse during the opening pose, adds a sharp accent on the firearm draw, becomes strongest during the forward step and three-quarter pose, then simplifies during the final three seconds. The final beat should leave a clear rhythmic opening for the first beat of the following dance segment.

第二段：15–30 秒

第二段开始从“动作秀”进入真正的写真摆拍。

这一段的重点不是重复拔枪，而是：

全身 → 中景 → 呼吸和身体线条 → 低机位 → 姿势变化

镜头开始明显靠近人物。

subject_definitions: <Subject 1> is the adult female cowboy shown in <Picture 1>. Her exact identity and all visual characteristics are fully anchored to <Picture 1>. She is the only human subject in the video. <Picture 1> is the sole visual reference for <Subject 1>. summary: [reference generation] Create a 15-second cinematic western fashion posing sequence for <Subject 1>. The sequence is entirely pose-driven rather than dance choreography. Build a continuous chain of confident, seductive, controlled fashion poses with smooth transitions between full-body, medium, close-up, side-profile, and low-angle compositions. Camera movement should remain fluid and elegant, with each change of framing motivated by the previous pose. retention_analysis: <Subject 1> (appears throughout): fully_preserved - preserve the identity and all visual characteristics from <Picture 1>. <Picture 1> (sole visual reference): fully_preserved - maintain the reference exactly throughout the complete sequence. detailed_description: The target video is a cinematic fashion posing sequence with continuous pose transitions and deliberately varied camera language. The performance is confident, controlled, slightly seductive, and editorial. The subject never performs dance choreography. Every movement is a deliberate fashion pose, weight shift, gesture, turn, or controlled body-line transition. [Shot 1] A medium-wide full-body composition begins from a slightly low angle. <Subject 1> stands with one leg carrying most of her weight and the opposite leg extended slightly outward. She slowly shifts her weight from one leg to the other, tilts her pelvis subtly, raises one arm toward the hat, then lowers the arm while turning her torso three-quarters toward camera. The camera performs a slow Truck Right with small amplitude, keeping her complete body readable. Near the end of the shot, she pauses in the three-quarter pose. [Shot 2] At 00:03.000, the camera cuts to a medium shot from slightly below eye level. <Subject 1> begins with one shoulder angled toward camera, slowly rolls that shoulder backward, lifts her chin, then brings one hand across the waist while the other hand lightly touches the hat brim. She gradually rotates her upper body toward camera without changing her position. The camera performs a very slow Push In with small amplitude, ending closer as she holds the pose and turns only her eyes toward the lens. [Shot 3] At 00:06.000, the shot cuts to a side-profile close-up focused on the upper torso, shoulders, neck, and face. <Subject 1> arches her upper back slightly, draws one shoulder backward, takes a slow visible breath, then releases the shoulders into a relaxed asymmetrical pose. She gently slides one hand along the waistline and returns it to a resting position. The camera performs a smooth lateral Tracking Shot parallel to her body, moving slightly from rear three-quarter toward side profile. The final moment holds on the subtle breathing motion and head turn. [Shot 4] At 00:09.000, the camera changes to a low-angle medium-full composition. <Subject 1> lowers into a controlled crouched fashion pose, placing one hand lightly against the ground while the opposite arm remains lifted near the upper body. She turns her torso diagonally, shifts the hips backward, then slowly raises her face toward camera. The camera performs a short Pedestal Up combined with a small Push In, following the rise of her torso without rotating around her. She pauses at the strongest crouched pose. [Shot 5] At 00:12.000, the camera transitions into a full-body three-quarter composition. <Subject 1> rises smoothly from the crouched pose, extends one leg forward, straightens the torso, rotates the shoulders toward camera, and finishes with one hand near the waist and the other near the hat. The camera performs a controlled Pull Out with small amplitude while moving slightly sideways, revealing the complete final silhouette. She makes one final slow weight shift through the hips and shoulders, then stops completely in a strong asymmetrical pose. Avoid dance steps, repetitive body shaking, random hip movement, random arm waving, exaggerated posing, abrupt camera jumps, rapid zooms, 360-degree rotations, or disconnected pose changes. All pose transitions must be physically continuous and visually motivated. overall_soundscape: Subtle synchronized clothing, hand, and foot movement sounds only. No dialogue. non_diegetic_music: A restrained modern western fashion rhythm around 96 BPM with clear beat accents supporting each pose transition and camera cut. Keep the rhythm sparse and controlled rather than dance-oriented.

第三段：30–45 秒

第三段是整个 45 秒的写真高潮段。

这里继续使用同一张 Picture 1，但镜头语言进一步收紧：

全身 → 脸部特写 → 侧面上半身 → 腿部/靴子 → 全身最终定格

与第二段相比，第三段更强调局部身体语言与镜头之间的配合。

subject_definitions: <Subject 1> is the adult female cowboy shown in <Picture 1>. Her exact identity and all visual characteristics are fully anchored to <Picture 1>. She is the only human subject in the video. <Picture 1> is the sole visual reference for <Subject 1>. summary: [reference generation] Create a separate 15-second cinematic western fashion posing sequence using <Picture 1> as the sole reference. This sequence continues the visual language of the previous segment but becomes more intimate and camera-driven, using tighter framing, side-profile body details, face close-ups, lower-body detail shots, controlled weight shifts, torso turns, breathing, and deliberate pose changes. It remains fashion posing, not dance choreography. retention_analysis: <Subject 1> (appears throughout): fully_preserved - preserve the identity and all visual characteristics from <Picture 1>. <Picture 1> (sole visual reference): fully_preserved - maintain the reference exactly throughout the complete sequence. detailed_description: The target video is a cinematic fashion posing sequence built around intimate framing, controlled body language, and smooth camera transitions. The subject performs confident and subtly seductive editorial poses rather than dance movements. Motion is deliberate and economical, with every pose changing through a clear body pivot, weight transfer, head movement, hand gesture, or controlled breathing. [Shot 1] The sequence opens with a full-body three-quarter composition from slightly below eye level. <Subject 1> begins with her weight on one leg, then slowly steps the other foot outward and turns her hips slightly while keeping the upper body facing camera. She raises one hand toward the hat and pauses, then lowers it while shifting one shoulder forward. The camera performs a smooth Tracking Shot moving diagonally toward her front-left side, ending in a stronger three-quarter composition. [Shot 2] At 00:03.000, the camera cuts to a tight facial close-up from slightly below eye level. <Subject 1> barely moves her body. She slowly lowers her chin, raises her eyes toward the camera, then turns her face a few degrees to one side before returning her gaze directly to the lens. One hand briefly enters the frame to touch the hat brim. The camera performs an extremely slow Push In with small amplitude, stopping at the strongest eye-contact moment. [Shot 3] At 00:06.000, the camera cuts to a side-profile upper-body close-up, framing the shoulder line, upper torso, waist, and face edge. <Subject 1> slowly turns her upper torso away from camera and then back toward it, creating a controlled change in the shoulder and waist line. Her breathing remains subtly visible through a natural rise and fall of the upper torso. One hand travels slowly from the waist upward and stops near the shoulder. The camera performs a smooth side Tracking Shot in the same direction as her torso rotation, maintaining consistent relative distance and creating a fluid sense of motion rather than a static insert. [Shot 4] At 00:09.000, the camera cuts to a lower-body detail composition from a low three-quarter angle, framing the legs and boots. <Subject 1> shifts weight onto the rear leg, extends the front leg, slowly pivots the front boot inward, then transfers the weight forward again. Her hips follow the weight transfer naturally while the upper body remains mostly outside the frame. The camera performs a slow Truck Left combined with a subtle Pedestal Up, following the change in stance and ending as the front foot settles firmly. [Shot 5] At 00:12.000, the shot transitions back to a medium-full three-quarter composition. <Subject 1> returns to an upright pose, turns her torso through a smooth half-turn into a front three-quarter position, adjusts the hat brim with one hand, and lets the other hand settle near the waist. She then makes one final controlled shoulder rotation followed by a small hip weight shift and holds the finished pose. The camera performs a slow Pull Out with medium amplitude, revealing the complete pose while maintaining a stable horizon. The final 1.5 seconds are held almost completely still except for subtle breathing and a minimal head movement toward camera. Do not turn any movement into dance choreography. Avoid repetitive hip shaking, continuous body waving, random arm movements, bouncing, exaggerated chest movement, exaggerated slow motion, rapid cuts, spinning cameras, 360-degree movement, or abrupt pose resets. Every cut must introduce a deliberately different framing or body detail, while every transition within a shot remains smooth and physically connected. overall_soundscape: Minimal physical movement sounds synchronized with the pose changes. No dialogue. non_diegetic_music: A restrained western fashion rhythm around 96 BPM with distinct beat accents at each framing transition. Keep the arrangement sparse enough for the camera cuts and pose changes to remain visually dominant.

最终得到的 45 秒结构

00–15s 人物建立 → 拔枪动作 → 手腕动作 → 低机位高潮 → 收枪 → 定格 15–30s 全身摆拍 → 中景 → 肩部/腰部姿态 → 侧面呼吸特写 → 低机位蹲姿 → 全身收尾 30–45s 全身 → 眼神特写 → 侧面上半身 → 腿部/靴子特写 → 全身半转 → 最终写真定格

这样写教程时，可以把它概括成一句：

“同一张人物参考图，连续三次进入 H3 Ref2VA，每次生成 15 秒，通过统一的人物锚点保持角色连续，再用不同的动作阶段、构图和运镜把三个 15 秒片段拼成一个完整的 45 秒牛仔写真短片。”

这种拆分方式也符合 H3 Ref2VA 按参考资产定义、保留关系和逐镜头时间线组织提示词的方式。(GitHub)