script-to-video-seedance
Seedance 2.0 Fast pipeline for motion-heavy video production via Phoenix Cinema Studio API. Specializes in: real-world physics, action / martial arts / sports / chase scenes, joint audio-video generation with phoneme-level lip-sync, multimodal reference (≤9 images + ≤3 videos + ≤3 audio), native Chinese prompts, long single takes (≤15s). Default: `seedance-2.0-fast` @ 480p for cost efficiency. Parses a screenplay into characters, locations, and scenes. Generates character reference images and location establishing shots, registers them via Ark compliance (single-step, no review modal), produces video clips where motion and physics carry the scene. Each clip is one `/generate-video` call, but Seedance can **render multiple internal shots within a single generation** when the prompt describes the cuts — so a single 15s clip can contain 2-3 shot-level beats (WS → MS → CU) that the model stitches together natively. Cross-clip consistency comes from reusing the SAME registered `reference_image_urls` across eve