ByteDance releases Seedance 2.5: XCMG, XPeng, etc. will be the first to integrate.

Today, ByteDance officially released its next-generation video generation model, Seedance 2.5.

The new model brings comprehensive upgrades in video generation duration, multimodal understanding, partial editing precision, and multilingual support.

For enterprise users, Volcano Engine will soon launch the Seedance 2.5 API service, further driving the adoption of video generation capabilities across industries such as industrial manufacturing, automotive, and robotics.

Official sources state that multiple companies, including Xugong Group, XPeng Motors, AgiBot, Lingchu Intelligence, Weifen Zhifei, Qiongche Intelligence, and Xspark AI (Wujie Zhihang), have reached cooperation intentions with Volcano Engine and will be among the first to integrate Seedance 2.5.

Currently, some companies have already begun exploring real-world scenarios based on the Seedance series models.

Among them, Xugong Group is applying it to industrial operation training and SOP workflow video generation, reducing the time and cost traditionally required for live-action filming and 3D animation production.

XPeng Motors, meanwhile, has integrated Seedance capabilities into its internal one-stop AI design platform, using it to assist visual creation and solution presentation during the product design phase.

Compared with Seedance 2.0, released earlier this year, Seedance 2.5 delivers upgrades across four core capabilities.

First, single-generation video duration has been extended to 30 seconds, with support for high-fidelity temporal extension, maintaining consistency in characters, scenes, and camera style across longer video content.

Second, the new model supports up to 50 multimodal reference materials and adds reference capabilities such as white-box models and green screens, enabling greater controllability for complex projects and industrialized video production.

In terms of editing capabilities, Seedance 2.5 introduces a video partial editing function that allows precise adjustments to background, products, people, and other local elements while preserving the overall visual rhythm and content structure.

Additionally, the new model natively supports more than 10 languages and further enhances its ability to understand and execute complex instructions, meeting the demands of batch production for multilingual, multi-version content.

With the API service going live, Seedance 2.5 is expected to further lower the barrier to enterprise video content production and push generative AI from creative production toward industrialized, large-scale applications.

Seedance 2.5 adds white-box model reference capability, able to render existing white-box assets into photorealistic car assembly sequences.