Seedance 2.0

Seedance 2.0 produces 12-15 second video clips from a prompt, an image, an existing video, or a reference subject, and is ByteDance's most capable video model in the family for holding a specific look, subject, or identity consistent across a generation. It's the workhorse model behind most of AutorunX's default video generation, chosen as the default specifically because it balances quality, control, and cost across a wide range of shot types.

Beyond generating new clips from scratch, it can also edit or extend existing video through its V2V mode, and preserve a specific subject's identity across reference-driven generations — useful whenever a shot needs to feature a particular person, character, or product consistently.

Seedance 2.0 is ByteDance's professional-tier video generation model, built to take multimodal inputs — text, a reference image, an existing video clip, or a reference subject — and produce a coherent 12-15 second clip from them. The 2.0 Identity variant is tuned specifically for reference-to-video generation that preserves a subject's identity (face, build, distinguishing features) across the output, rather than treating the reference as loose stylistic guidance.

The model family spans several variants tuned for different trade-offs: 1.0 Fast and 2.0 Fast prioritize speed and lower cost per clip, 1.5 Pro and 2.0 target higher fidelity and stronger prompt adherence, and 2.0 Identity adds the identity-preservation behavior for reference-to-video and subject-to-video (S2V) work. This lets a single family cover both quick iteration passes and final, production-quality renders.

Resolution scales from 480p up to 4K depending on the request and variant, and beyond straight text-to-video and image-to-video, Seedance 2.0 supports video-to-video (V2V) editing of an existing clip and subject-to-video generation, where a reference subject drives the appearance of the generated output.