写好提示词、选对模式与模型，并从真实 showcase 范例一键复刻开始。
aiMakeVideo 遵循自然语言。按需叠加这些要素——不必每次写全，但更清晰的简报通常更好控片与控图。
画面里是谁/什么——人物、产品、动物或场景主角。若要多轮迭代，先把身份特征写稳。
随时间发生什么：走、倾倒、转身、环绕、布料被风吹起。静图则写姿态与微动作，而不是时间轴。
景别、镜头感与运镜——推进、环绕、手持跟随、锁死三脚架。一次清晰运镜胜过三种互相打架的运动。
主光方向与质感——柔光美妆、硬朗正午、金色时刻、霓虹轮廓光、影棚柔光箱。
整体观感：商业写实、编辑美妆、电影惊悚、安静生活流。需要时写明避免项（无字幕、无路人）。
[主体] + [动作] + [镜头] + [光线] + [环境] + [氛围]。视频可选时间节拍（0–2 秒 / 2–5 秒）。
从你已有的素材出发。之后换模式也没问题——Recreate 与 Create 同属一个工作室。
16:9Photoreal live-action fantasy escape through an active volcanic forge cavern, crisp high-resolution cinema detail, ember heat haze kept optically sharp, clean HDR contrast, one wide lens and one uninterrupted controlled handheld pursuit shot. The smith-mage is a muscular North African woman, late 20s, shaved sides with dense curls, glowing ember freckles, soot-streaked bronze armor over ash linen, one living molten hammer in her right hand. Keep one identical smith-mage throughout. Exactly 5 distinct lava constructs emerge; construct 5 is a towering slag titan with a chain-anvil. Never reuse a construct. Every hammer blow hits exactly one target. Show 5 separate contacts, cooled-glass shatters, freezes into dead stone. No multi-target smash. Each next construct blocks, hurls slag, or collapses a bridge. Rotate overhead forge strike, lateral quench-smash, vault over magma rivulet, ground-spike kick, and final titan knee break. Every relocation: micro-anticipation; orange-white forge flash hides her then reveals her already striking the next construct; camera keeps physical follow. No transit smear, no whip pan, no motion blur. Keep skin, armor, lava crust, particles sharp. No dialogue or subtitles. Furnace roar, 5 quench-impact accents, rock collapses, flash cracks, industrial percussion sting.
16:9Photoreal live-action modern wuxia duel atop a speeding maglev at golden hour, crisp high-resolution cinema detail, wind-whipped coat fabric, clean HDR contrast, one 40mm-equivalent lens and one uninterrupted controlled handheld tracking shot along the train roof. The duelist is a lean East Asian man, early 30s, sharp jaw, windblown black hair tied low, charcoal tailored coat over combat layers, one carbon-fiber jian in his right hand. Keep one identical duelist throughout. Exactly 6 distinct masked rivals climb onto the roof; rival 6 is a twin-blade captain. Never reuse a rival. Every cut lands on exactly one opponent. Show 6 separate contacts and individual falls off-frame. No multi-target slash. Each next rival lunges, binds, or forces the roof edge. Rotate descending cut, rising cut, bind-and-displace, reverse-grip draw, spinning deflection, and final disarm. Every relocation: micro-anticipation; pale steel flash hides him then reveals him already committed beside the next rival; camera keeps rolling with train landscape visible. No transit smear, no whip pan, no motion blur. Camera stays close with sidesteps and arcs toward each flash destination without snapping. No dialogue or subtitles. Wind roar, rail hum, 6 blade accents, dry flash cracks, accelerating taiko-electronic sting.
1:1Use this image as the first frame. Keep the exact same young Black woman — face, curly black hair, body proportions, pearl jewelry, and the living-room background (sofa, coffee table, kitchen doorway) locked to the reference. Photoreal image-to-video, continuous one take, no cuts. 0–0.8s: She stands centered, facing the camera, soft smile, both arms open slightly to the sides, champagne satin slip dress as in the reference, warm indoor light. 0.8–3s: First full spin left-to-right (clockwise from her left). As she turns through her back and comes around facing camera again, the dress morphs into a different elegant gown — new color and silhouette (e.g. deep emerald silk with soft draping), face and hair unchanged. 3–5.5s: Second continuous left-to-right spin. On the next face-front beat, wardrobe changes again to another elegant dress (e.g. midnight blue satin with a subtle flare), same jewelry, same room. 5.5–8s: Third left-to-right spin. Final face-front reveal into a third elegant look (e.g. soft blush chiffon floor-length), she finishes facing camera with an open-arm pose and a bright smile. Fabric swirls with realistic motion blur on turns; identity rock-solid. Three continuous spins, three elegant dress changes only at the face-front moments. No face morph, no extra people, no text, no logos, photoreal cinematic, 8 seconds.
3:4Same young East Asian woman as the reference: wet black hair, clear plastic hooded raincoat with raindrops, pale dewy skin, cinematic rainy neon city night. Lock face, hairstyle, raincoat, and red-blue neon bokeh to the reference. Handheld medium shot, shallow depth of field, wet asphalt reflections. 0–1.2s: She stares into the camera with a cold, blank look — flat eyes, no smile, rain ticking on the plastic hood. 1.2–2s: Suddenly her expression flips into a wicked, seductive smirk, eyes sharp and playful. 2–3.5s: She turns away so her back faces the camera, then runs a few light steps forward away from camera into the neon street; raincoat flutters. 3.5–5s: She turns to face the camera again and walks backward toward the lens a few steps, eyes locked on camera. 5–6s: Both hands rise; she points both index fingers at the camera / screen, bright teasing smile. 6–6.8s: Short open laugh, one shoulder shake, still facing camera. 6.8–8s: She spins and sprints out of frame to the side; camera holds on rainy neon street. Natural continuous one-take motion, no cuts, no face morph, no extra people, photoreal cinematic, 8 seconds.
3:4Beauty portrait in a translucent raincoat, soft neon bokeh, rain droplets as highlights, porcelain skin, romantic noir elegance, photoreal
1:1Intimate still life of elegant hands holding a glowing tea bowl, warm side light, linen and wood textures, wabi-sabi beauty, photoreal detail
3:4For Image1, keep exact likeness, transform into a soft-lit beauty studio portrait, creamy backdrop, natural skin texture, refined catchlights, high-end beauty campaign, no makeup plastic look
Keep pose and identity. Restyle wardrobe into luxurious silk evening wear, elegant jewelry, fashion-editorial lighting, couture beauty finish, accurate face
aiMakeVideo 是多模型工作室。按任务选——速度、运动质量或静帧细节——再用同一提示词迭代。
Kling、Seedance、Veo 在运动表现、速度与风格上各有侧重。可先用更快/更便宜档位定构图，再提高分辨率或换模型。
Nano Banana、GPT Image 等适合美妆、产品与编辑静帧。更高质量档与更多参考图会消耗更多积分。
首页 showcase 卡片带有真实可上线提示词。点击「复刻」即可写入 Create，再改文案、模型与比例。
精致美妆头像、蝴蝶光、自然肤质、象牙背景、85mm 气质。可用文生图，再用图生视频做轻微推进。
3:4Ultra refined beauty headshot of a young woman, soft butterfly lighting, luminous skin with natural pores, subtle catchlights, clean ivory backdrop, high-fashion beauty still, 85mm
磨砂玻璃上的精华瓶与水珠、腮红粉渐变、奢侈品广告静帧。很适合图生视频：慢环绕或轻推进，锁定瓶身设计。
1:1Premium skincare serum bottle with water droplets on frosted glass, soft gradient backdrop in blush pink, elegant reflections, luxury cosmetic advertising still, ultra sharp
写实追逐语言：主体身份、雨夜霓虹、手持跟随、清晰冲突节拍。从文生视频开始；若要干净成片可写「无对白/无字幕」。
16:9Photoreal live-action cyberpunk chase at rainy night, crisp high-resolution cinema detail, wet asphalt reflections, clean HDR contrast, one 35mm-equivalent lens and one uninterrupted controlled handheld pursuit shot. The courier is a lean East Asian man in his mid-20s, cropped black hair with a cyan underlight edge, thin scar along the jawline, matte black riding jacket, fingerless gloves, compact magnetic satchel locked to his chest. Keep one identical courier throughout. He flees exactly 4 distinct armored drone-riders; each has a unique chassis silhouette and melee tool; rider 4 is a heavy enforcer with a telescoping stun lance. Never reuse a rider. Every near-miss and strike is one-to-one. Show 4 separate contact events: sparks, shattered plating, wipeouts. No multi-target hits. Each next rider actively cuts him off, rams, or grapples. Rotate vault, slide-under, wall-run, and final lance parry without adjacent repetition. Every relocation: micro-anticipation; a hard cyan strobe fully hides him and he vanishes; camera keeps physical follow with city visible; a second local strobe reveals him already in the next escape or attack pose. No transit smear, no whip pan, no motion blur. Camera stays close with smooth arcs and pivots toward each strobe destination without snapping. No dialogue or subtitles. Night city ambience, tire hiss, 4 distinct impact accents, dry strobe cracks, accelerating synth hit.
参考图优先：同一套装与海边悬崖，再写节拍（风→走→旋转→微笑）。典型图生视频写法——锁相似度，写运动。
3:4Same young woman as the reference: long wavy blonde hair, beige linen shirt and wide pants, gold jewelry, barefoot on a sunlit cliff above turquoise sea. Lock face, outfit, and coastal background to the reference. Handheld medium-wide shot, bright golden daylight, windy. 0–2s: She stands on the rock edge, hair and linen fabric whipping in strong wind, soft side glance. 2–4.5s: She turns toward camera, steps forward along the ridge, fabric flares, cinematic backlight. 4.5–6.5s: Arms open wide, spins once in place — linen swirls, hair arcs, sea glitter behind. 6.5–8s: Faces camera with a bright smile, wind still strong; camera slowly pushes in. Continuous one take, photoreal, no face morph, no extra people, 8 seconds.
每次生成前都会估算积分。时长更长、分辨率更高、模型更贵、参考图更多，消耗都会上升。
按渠道选比例——16:9 偏横版、9:16 竖屏、1:1 信息流。先对齐最终裁切，避免运动浪费在画外。
短片更省积分也更好写。先用短时长锁定构图，确认观感后再加长。
打开 Create 使用共享工作室信息流，或直接进入工具页。需要强起点时，从 showcase 点「复刻」即可。
先用同一套清晰公式。再按需要微调——有的模型更吃短节拍，有的能接受更细的运镜描述。用对比出片代替每次推倒重写。
已有满意画面就用图生视频或图生图；想法只在文字里，就用文生视频或文生图，之后再把静帧做成运动。
在首页 showcase 点击「复刻」，简报会进入 Create（或当前工具页创作器）。修改后选模型、设比例，再生成。
