Seedance 2.0 Technical Architecture

Public technical descriptions refer to a unified multimodal audio-video joint generation architecture. Inputs: text plus up to 9 images, 3 video clips, and 3 audio tracks (subject to platform limits). Text drives scene, action, and style; images/videos/audio provide reference for composition, motion, camera, and sound. The @ tag system in prompts lets you assign roles to each asset.

Video: typically 4–15 seconds selectable; resolution up to 2K (2048×1080); aspect ratios often include 16:9, 9:16, 1:1, 4:3, 3:4, 21:9 and adaptive. Audio: native stereo, generated jointly with video (not post-dubbed); lip-sync supported for multiple languages in public reports. Video extension and in-place editing are supported in many workflows.

Third-party technical write-ups describe a dual-branch diffusion transformer that processes visual and audio streams in a single inference, enabling lip-sync, sound effects, and music to be aligned with the picture from the start. Consistency across shots is achieved by reusing the same reference image(s) and referring to them in the prompt.