ByteDance Introduces Seedance 2.5 With Longer AI Videos and Advanced Editing

Seedance 2.5 Doubles Native Video Generation Length

ByteDance has officially launched Seedance 2.5, the latest version of its AI video generation model, introducing major upgrades in video length, editing precision, and multimodal content creation.

The new model can generate 30-second videos in a single pass, doubling the native generation limit of the previous version. Users can also extend generated clips through multiple rounds, enabling the creation of coherent videos lasting several minutes.

Improved Multimodal Creation and Editing

One of the biggest improvements in Seedance 2.5 is its expanded multimodal workflow.

The model supports up to 50 reference assets in a single project, including images, videos, and audio clips. It can combine these references to maintain consistent characters, visual styles, scenes, and audio throughout long-form video generation.

ByteDance also introduced more precise editing tools, including:

• timestamp-based scene editing;

• localized modifications without regenerating the entire video;

• green-screen compositing;

• camera angle and motion control;

• improved subject consistency across multiple scenes.

Better Visual Quality and Professional Applications

According to ByteDance, Seedance 2.5 significantly improves image quality by reducing common AI artifacts such as overly smooth or "plastic-like" textures.

The company says the model produces more realistic lighting, skin details, object textures, and motion while reducing unwanted subtitles and audio artifacts.

Beyond content creation, ByteDance expects the technology to be used in education, industrial manufacturing, robotics, and autonomous driving, where synthetic video data can support AI training and simulation.

Competition in AI Video Continues to Intensify

The launch further strengthens ByteDance's position in the rapidly evolving AI video market.

With native 30-second generation, Seedance 2.5 currently exceeds the generation length offered by several competing models, while continuing ByteDance's push into professional AI video production.

The model is rolling out through Jimeng AI and Doubao Pro, with API access expected to become available soon via the Volcano Ark platform.

Source: https://seed.bytedance.com/en/blog/one-take-creation-flexible-referencing-introducing-seedance-2-5