ByteDance: seedance-2.5

Seedance 2.5 is ByteDance Seed team's video generation model, announced on June 23, 2026 at the Volcano Engine FORCE conference and officially launched on July 31, 2026. It produces a single continuous shot of up to 30 seconds natively in one pass, without stitching segments. A unified joint audio-video generation architecture processes visual and audio signals in the same latent space, yielding ambient sound, foley and score that stay in sync with the generated picture. Multi-turn extension expands content to several minutes while keeping a consistent audiovisual language. The model accepts text prompts alongside up to 50 reference inputs per generation: 30 images, 10 videos and 10 audio clips. These references can guide subject appearance, scene style, motion and voice characteristics, through white-model reference, motion reference and creative reference modes. Timestamp-level control lets creators specify actions, camera changes, rhythm and edit instructions at precise points, while region-level editing replaces a subject, background or product without re-rendering the whole clip. Green-screen, viewpoint and reference editing are also included. Seedance 2.5 suits short dramas that need a complete narrative arc, advertising films and social media ad units, product trailers and demos, and complete onboarding videos. It also serves animated content, film and television production, corporate communications and training, and industrial, embodied-intelligence and autonomous-driving work. Its strengths are native 30-second single-shot generation that avoids character drift and lighting inconsistency from stitching, plus strong micro-level rendering quality and high controllability for brand and product fidelity.