ByteDance officially launched its next-generation video generation model, Seedance 2.5, on July 31, after multiple delays. The model doubles single-generation video duration from the previous generation's 15 seconds to 30 seconds, and achieves major breakthroughs in multimodal reference capabilities and refined editing control. Seedance 2.5 continues the unified multimodal audio-video joint generation architecture of Seedance 2.0, with enhancements in long-form narrative, multi-asset fusion, and precision editing. The model is currently rolling out on platforms including Jimeng AI and Doubao Pro, with its API service set to open soon via Volcano Ark. It supports multi-round extension, allowing users to create coherent videos lasting several minutes with a unified audio-visual language, optimizing shot transitions and scene continuity. In the industry landscape, Seedance 2.5's 30-second single shot leads competitors, compared to Kling 3.0's 15 seconds, Veo 3.1's 8 seconds, and Sora 2's discontinuation. Seedance 2.5 also optimizes image quality, audio quality, and motion quality, addressing "plastic-like" artifacts through fine control over textures, skin texture, eye details, lighting, and color saturation, while reducing uncontrolled subtitles and background music. The model supports up to 50 full-modality reference inputs in a single pass.