Seedance 2.0 — Vídeo 2K Multi-Shot com Entrada Quad-Modal

The @ reference system lets you assign specific roles to each uploaded file: @face for character likeness, @motion for movement style, @style for visual tone, @audio for soundtrack sync. No other model offers this level of compositional control over multimodal inputs.

Dolly zooms, rack focuses, tracking shots, POV switches, smooth handheld — Seedance 2.0 scored 9/10 for camera control in benchmark testing, the highest among competing models. Each shot in a multi-shot sequence can have its own camera behavior.

ByteDance incorporated physics-aware training that penalizes impossible motion during generation. Cloth drapes and wrinkles naturally, water splashes with correct weight, collisions have impact, and characters shift balance when walking.

The Dual-Branch Diffusion Transformer generates audio and video in parallel, not sequentially. Spoken dialogue syncs to lip movement at the phoneme level, foley effects match on-screen action, and environmental ambience adapts to scene changes.

Characters retain their facial features, clothing, hairstyle, and proportions across all shots in a multi-shot video. The internal reference-locking system ensures the same person looks the same regardless of camera angle or scene change.

Generate up to 6 independently controlled camera cuts in a single generation. Each shot can have its own framing, camera movement, and action — the output looks like an edited sequence, not a raw single-take generation.