Description

Give a complete reference video to a multimodal AI to automatically break down its real shots, timeline, visual changes, and audible sounds, then generate production-ready prompts for video creation. It is not limited to live-action, advertising, or film footage; it also supports product showcases, architecture and scenery, animals, food, sports actions, dance, 2D/3D animation, UI demos, screen recordings, infographics, visual effects, and abstract imagery. Based on the video’s information structure, the assistant can output complete text-to-video prompts, or generate shot-by-shot prompts for the reference image or first frame, continuous motion, necessary end frames, and shot-level stability constraints needed for image-to-video creation. When both the static image and temporal changes matter, it can provide both versions. The generated content preserves information that can be directly verified, including subjects, composition, camera movement, action causality, structural changes, text, dialogue, ambient sound, and music, while tracking each shot through its actual ending state. When text, sound, identity, location, or filming method cannot be clearly confirmed, the assistant avoids filling in details without evidence. It also distinguishes the intended scene from external elements such as player watermarks and control overlays, helping reduce content drift during generation. Suitable for creators and production teams who need to reverse-engineer text-to-video prompts or I2V shot prompts from reference videos, or organize source material for video reconstruction, style reproduction, and generative advertising.