ByteDance launches Seedance 2.5: 30-second 4K AI video from 50 inputs

At the Volcano Engine FORCE conference in Beijing on Tuesday, ByteDance introduced Seedance 2.5, an advanced AI video generation model capable of producing 30-second clips in native 4K resolution from a single text prompt. The company bypassed four intermediate versions, moving directly from the previous iteration to underscore what it calls a generational leap in capability. An enterprise beta is now available for select partners.

The new model accepts up to 50 distinct inputs , including text, images, and reference videos , to generate coherent, high-resolution footage. This marks a significant upgrade from earlier systems that struggled with consistency over longer durations or required multiple separate generations to achieve similar results.

ByteDance’s decision to skip several version numbers reflects the scale of improvement in Seedance 2.5. The model maintains temporal coherence across full 30-second sequences while preserving 4K detail that was previously unattainable in AI-generated video at this length. Early demonstrations show smooth motion, realistic lighting, and minimal artifacts even in complex scenes.

The release positions ByteDance to compete directly with other major players in the rapidly evolving AI video generation market, including OpenAI’s Sora and Meta’s Make-A-Video. By focusing on native resolution rather than upscaling, Seedance 2.5 aims to meet professional production standards from the start.

Enterprise access will allow businesses to test the model for applications in advertising, content creation, and virtual production. Pricing and broader availability have not yet been announced, but the company signaled that a public release could follow based on beta feedback.