ByteDance unveils Seedance 2.5, a 30-second 4K AI video model that accepts 50 reference inputs
The TL;DR
ByteDance announced Seedance 2.5 at its Beijing conference, producing 30-second native 4K video from up to 50 multimodal displays.
ByteDance introduced Seedance 2.5 on Tuesday at its Volcano Engine FORCE conference in Beijing, a video production model that produces 30-second clips in native 4K resolution from a single instant. The company skipped four intermediate versions entirely, jumping straight from its predecessor to show what it described as a leap in productivity.
The enterprise beta is already live, and the public launch is targeted for early July. CEO Liang Rubo told the conference that the rise of the AI conference is a priority for the company, as its business model-as-a-service transforms into a core business supported by long-term investment.
The improvement of the article is a reference dimension: the model accepts up to 50 multimodal inputs, including images, audio clips, white 3D models, and style references, from the previous 12. Those inputs give Seedance 2.5 much more granular control over style, movement, and texture than text input alone.
TNW City Coworking Space – Where your best work happens
A workplace designed for growth, collaboration, and endless networking opportunities at the heart of technology.
The model produces in 4K natively rather than upscaling from a lower resolution, a key difference in professional production pipelines. It supports 10-bit color depth for smoother gradients and more room for post-production color grading. ByteDance also wants to catch up 20 percent faster, which means several generations before a usable result.
Audio is now co-processed in the same subtle environment as visual signals, producing native synchronization between on-screen actions and their accompanying sounds. The new 3D white box preview function allows creators to create low-fidelity animations before committing to a full-quality rendering. Together, the features position the model as a production tool rather than a new generator.
The announcement comes three months after ByteDance was forced to add watermarking and IP guardrails to Seedance 2.0 following cease and desist letters from Disney, Warner Bros Discovery, Paramount, and Netflix. The viral fake of Tom Cruise fighting Brad Pitt on a rooftop has drawn a formal complaint from the Motion Picture Association and a reprimand from SAG-AFTRA.
ByteDance stopped the global rollout in mid-March and didn’t relaunch with CapCut until late March, with face-blocking filters, C2PA watermarks, and copyrighted character detection in place. No timeline has been given for making the new model available in the United States.
The competitive landscape has changed dramatically since February. OpenAI shut down Sora in March after the video tool peaked at about a million users and was reportedly costing about a million dollars a day to operate, making just over $2 million in revenue.
Google’s Veo 3.1 filled much of the vacuum, providing native 4K output, audio reproduction, and up to three reference images for style control. But ByteDance’s new model far exceeds Veo’s reference input capacity, accepting 50 of Veo’s three inputs, a critical gap in professional workflows.
The AI video production market is rapidly fragmenting, with Chinese models moving faster in production tools than their Western competitors. Third-party platforms such as Reallusion’s AI Studio are already building professional pipelines around the predecessor model, and the fourth-generation Runway tool is out of the Artificial Analysis top 10.
Whether the new model can reach global markets without dominating the copyright wars that halted its predecessor remains a key question. ByteDance has a model, distribution with 400 million active CapCut users, and direct integration from generation to editing to sharing. What he does not have yet to fix with Hollywood, and all the features that make the model better and suggest the foundations of that conflict that has not yet been resolved.
