参考图驱动的分镜转视频:单品概念短片
以「从文字到画面」的分镜概念为出发点,演示 reference-to-video 工作流的典型用法:先用两张参考图钉住主体外观与场景基调——一张是产品特写(Image1),一张是场景气氛参考(Image2),再通过提示词把静态参考转成 5 秒带镜头运动的短片。 构成上,base 只包含 schema 要求的非媒体字段:reference_image_urls(由 assets 生成并绑定)与 duration;顶层 prompt 会作为作者 input.prompt 真实提交,其中以 Image1、Image2 索引引用两张参考图。可选字段 size 固定为 1280*720(16:9、720P 档),audio 关闭以保持静音素材便于后期配音,watermark 关闭。 适用场景:广告与社交短视频的创意打样、分镜预览、产品演示素材的前期验证。参考图约束外观与构图,可在同一提示词下换参考图做多版本 A/B 迭代;受 schema 限制,单次输出时长为 3–15 秒,长叙事仍需人工剪辑拼接。适合用 Vidu reference-to-video 模型族创作此类参考图驱动的短片段。
风格 Storyboard · 产品短片 · 参考驱动
模型 Vidu Q3
参考输入

参考图

参考图
提示词
Reference-to-video storyboard concept. Use Image1 as the product reference for shape, material and surface finish; use Image2 as the lighting and environment reference. Build a five-second cinematic product shot: the object rests on a matte concrete plinth in a dim studio, a slow dolly-in moves from a three-quarter wide framing to a tight close-up, single soft key light from the upper left rakes across the surface, subtle dust motes drifting in the beam, background falling into gentle falloff. Camera motion is smooth and stabilized, no handheld shake, no morphing of the product geometry, appearance and proportions stay consistent with Image1 throughout. Clean cinematic color grade, shallow depth of field, realistic reflections, calm and restrained advertising tone.
复制完整提示词,改成你自己的生成任务。
标签