From Text to Image: Reference-Image-Driven Storyboard Concept to Video

This piece demonstrates how the Vidu Q3 reference-to-video pipeline works: first, two concept reference images are generated with the pipeline (the protagonist's appearance and the scene art), then they are introduced in the order Image1, Image2 via a natural-language prompt, outputting a continuous shot with native audio. The reference_image_urls in the schema supports 1–7 reference images, referenced in the prompt by array index naming, so in the storyboard stage you only need to split characters, props, and scenes into separate reference images, and describe camera movement, action pacing, and sound in text to get a narrative segment with consistent subjects. duration can be set from 1–16 seconds; this piece uses 8 seconds to accommodate the complete rhythm of "establishing shot → push-in → character action"; size uses 1920*1080 landscape and audio is enabled, so dialogue and ambient sound are output directly in sync with the picture. Applicable scenarios: storyboard previsualization for comic dramas/short dramas, concept validation for advertising and e-commerce short videos, and series short-film production requiring multi-subject consistency. Developers can directly refer to this entry's base and assets structure, replace their own character and scene images into reference_image_urls, and reproduce the same type of generation workflow.