AI 商品视频生成

整份需求写的都是编排和节奏。四种口味、几何阵列、水果绕轨飞入、每一刀都落在拍子上——全部按拍子逐句写死，饼干本身只靠一张参考图撑住。这里没有一句在描述感受；每一句描述的都是一帧画面。

A bright, high-energy commercial for a fruit-filled sandwich biscuit, floating against seamless colour backdrops that change with the flavour. The biscuit is reference Image 1: hold its shape, crimped edge and dimple pattern in every shot. Studio macro light, no hands, no packaging, no on-screen text. Cut on the beat of an upbeat electronic pop track. 0-2s: coral-pink backdrop. One biscuit hangs dead centre and turns slowly on its own axis while whole and halved strawberries tumble and orbit around it, some drifting past the lens close enough to blur. 2-3s: cut on the beat to mint green. Four green apples float together into a tight cluster in the middle of frame, then split apart to reveal the biscuit hanging behind them. 3-5s: the apples clear and two dark grapes swing in either side of the biscuit; a full bunch of grapes then fills the frame and the backdrop turns lilac. 5-7s: the biscuit alone on lilac, a slow light sweep crossing the backdrop behind it, then three oranges bounce into frame around it. 7-8s: warm-orange backdrop. One orange bursts open in slow motion, pulp and juice spraying outward in macro. 8-9s: the frame settles back on the single biscuit, floating on peach. 9-11s: pull out to a large ordered grid of the biscuits laid flat on pink, evenly spaced and all facing the same way; the camera drifts across the grid. 11-14.5s: one continuous slow push in through the grid, the biscuits growing until the lens is inside the crumb — extreme macro across the dimpled surface. 14.5-15s: lift to a side-on hero of one biscuit, the strawberry filling glossy between the two halves, and hold. Audio: upbeat electronic pop with a four-on-the-floor kick, a crisp whoosh on every colour change, a wet burst on the orange, a dry crumb crunch as the macro lands. No voiceover.

商品片不一定要拍一个人在用它。这条是有包袱的短故事，商品是那个笑点：水果才是主角，蜥蜴是你看到第八秒还没划走的原因。十五秒，按拍子排好，全文印在下面。

A 3D animated commercial in the register of a high-end animated short, with a streak of exaggerated humour. Bright translucent colour, soft desert light, delicate skin texture, dreamy macro depth of field. The desert horned lizard is reference Image 1: keep its markings, horns and expression identical in every shot. 0-2.5s: a desert scorched by the sun, the air shimmering with heat. Wide shot, horizon high, dunes running back to nothing. The lizard drags itself across the scalding sand, tongue slightly out, swaying every couple of steps as though it is about to evaporate. 2.5-6.5s: cut in close on its face, then closer still. Eyes glazed, breath shallow, grains of sand stuck to its snout. The camera stays down at sand level and moves with it. 6.5-7.5s: its nose twitches and it stops. A cool, plump, dew-covered grapefruit sits half-buried in the sand just ahead, catching the light with a hard sparkle. 7.5-9.5s: hold on the reaction — the eyes go wide, the grapefruit soft in the background behind its head. 9.5-11s: it scrambles forward and throws both front legs around the fruit. 11-14.5s: it lies against the grapefruit hugging it, cheek pressed to the rind, blissful. Push into macro on its hands gripping the peel and the beads of dew running down. 14.5-15s: it bites. The rind splits and the pulp bursts open in extreme slow-motion macro, juice thrown toward the lens. End there. Audio: dry desert wind and heat haze through the opening, a soft sparkle as the fruit is spotted, a rising musical swell under the scramble, a loud wet crack on the bite. No voiceover, no on-screen text.

最不好看、却最能出单的那一条。四步安装演示，每一步都钉在自己那张参考图上，最后一个镜头里的机器和第一个镜头是同一台。这就是那条能把退货挡下来的视频——每一拍都标了时间码，节奏才会照着回来。

A 15-second setup tutorial for a capsule coffee machine on a bright kitchen counter. Hands only, no faces. Every step is pinned to its own reference image: Image 1 the machine body and water tank, Image 2 the drip tray, Image 3 the used-capsule box, Image 4 the tank at its MAX line. Clean daylight, real hands at real speed, no on-screen text. 0-1.5s, step one, the water tank. Medium shot from slightly above, the back of the machine. Two hands lower the transparent tank onto the rear of the body and press it straight down until it seats. Voiceover: "Start with the water tank. Line it up with the slot at the back and push down until it clicks." 1.5-5s, step two, the drip tray. Close-up at counter height, front of the machine. One hand slides the black drip tray horizontally into the rails at the base, backs it out to show the fit, then pushes it in flush. Voiceover: "Next, the drip tray. Slide it into the rails at the bottom until it sits flush with the body." 5-8.5s, step three, the capsule box. Close-up from slightly low, the cavity above the tray. Two hands align the white used-capsule box with the recess and push it in level with the tray front. Voiceover: "Now the used-capsule box. Push it in until it is level with the tray. Spent capsules drop in here on their own." 8.5-15s, step four, the first fill. Cut to a tighter side view of the tank on the machine, the MAX line dead centre of frame. A hand lifts the lid and pours clear water from a glass jug in a steady stream; the level climbs, the camera holds while it stops just under the MAX line, and the lid closes. Voiceover: "Fill the tank with fresh water, stopping just under the MAX line, then close the lid. That is the setup done." Audio: quiet kitchen room tone, a firm click as the tank seats, a plastic slide on the tray, water running into the tank. One calm, unhurried voiceover line per step, no music.

这一页最短的提示词，也是最有用的一招：给它一支你喜欢节奏的广告，再给它你的商品，然后说运镜和切点从片子里拿，商品从图里拿。你抄的不是素材，是剪辑。

Take the camera movement and the editing rhythm from reference Video 1, and take the product from reference Image 1: a pair of over-ear headphones. Build an advertising short around them in scenes that suit the product's tone — modern, spare, a little cold. Keep the switching points between product close-ups and wide scene shots where the reference clip puts them, and match its motion speed and transition style.