ByteDance SeeDance 2.5 adds native 4K and 50-reference multimodal support
ByteDance announced SeeDance 2.5 at Volcengine FORCE with native 4K resolution, support for up to 50 multimodal references including 3D models, and prompt-based editing. The company also retrofitted 4K into SeeDance 2.0 and launched a copyright-licensing platform for creators.
ByteDance announced SeeDance 2.5 at the Volcengine FORCE conference this week, bringing native 4K output, support for up to 50 multimodal references including 3D models, and prompt-based editing. The company also retrofitted native 4K support into SeeDance 2.0 ahead of its public launch, with SeeDance 2.5 shipping in July.
The multimodal reference expansion is the standout feature. Earlier video models typically accept a handful of reference images or a single style guide; SeeDance 2.5 ingests up to 50 references in a single pass—including 3D model files, which enables consistent character rigs and asset libraries in video workflows. Prompt-based editing lets users tweak generated clips without re-running the full pipeline, a capability becoming standard in commercial video-gen tools but rare in open-weight models. ByteDance also updated Seedream 5.0 Pro with layer-by-layer content editing, refreshed Seed-Audio 1.0, and advanced the Seed LLM family to version 2.1, though text models remain less competitive than the visual offerings. The company's new copyright-licensing platform allows rights holders to license their content for remix templates, with revenue split between creator and user—a direct response to IP controversies that have dogged Chinese AI companies. Hong Kong director Stephen Chow signed on as the first partner; daily generation volume using his comedy films has already surpassed 100,000.
