Seedance is ByteDance’s first‑generation video foundation model designed for both text‑to‑video (T2V) and image‑to‑video (I2V) generation. The 1.0 release emphasizes smooth, stable motion and precise prompt following, producing cinematic 1080p clips and natively supporting multi‑shot storytelling with consistent subjects, style, and atmosphere.
Key capabilities
- Native multi‑shot narrative generation with strong subject/style consistency.
- High spatiotemporal fluidity and structural stability for complex actions and multi‑agent scenes.
- Dual‑modality prompting (text and image) and accurate interpretation of diverse styles—from photorealism to illustration.
- A unified architecture and post‑training stack (fine‑grained SFT, video‑specific RLHF) with optimizations for fast inference.
The official technical report for Seedance 1.0 was posted in June 2025, detailing the data curation, architecture, training, and acceleration techniques behind the model. Seedance is accessible through ByteDance’s Seed portal with options to try the model and obtain API access, supporting creative workflows from storyboards and ads to multi‑shot narrative shorts.
