What problem does it solve? Seedance 2.5 uses a different prompt dialect than Seedance 2.0, and generic prompts waste credits on failed generations. This Skill routes the four generation modes correctly, writes explicit reference roles with exclusions, and stages long videos so beats are not dropped. ## Core Features & Use Cases - Mode routing and parameter surface: Picks between t2v, omni_reference, video_edit, and video_extension before writing, and enforces the Higgsfield limits (480p/720p, 4-30s duration, extension_mode rules, video_edit billing by source length). - Reference-role grammar: Declares what each @Image/@Video/@Audio material contributes and what to exclude, with fidelity grades, a five-step multi-reference workflow, and first/last-frame and multi-keyframe control written in prose. - Long-video staging and playbooks: Stages 30-second videos into end-state beats, splits incompatible jobs into separate prompts, and provides templates for video editing, forward/backward extension, storyboard grids, blockouts, and seamless transitions. - Use Case: A user supplies a character sheet, a location still, and a 6-second plate and asks to replace the actor. The Skill routes this to the omni_reference v2v lane, matches duration to the source, and writes the performance-inheritance clause. ## Quick Start Write a Seedance 2.5 prompt for a 15-second video using my three reference images, with the first image as the opening frame.
