ByteDance Launches Seedance 2.5, Generating 30-Second Video Clips Without Stitching

ByteDance began rolling out Seedance 2.5 to the public this week, the latest version of its video generation model, previously announced at the Volcano Engine FORCE conference. The biggest change is native generation of 30-second clips in a single diffusion pass, without stitching together shorter fragments the way the previous version had to.

What's new in the model

Seedance 2.0 could natively generate up to 15 seconds of footage, with longer shots created by joining several such fragments together, which often showed up as visible shifts in pacing, lighting or camera motion between cuts. Seedance 2.5 generates the full 30 seconds as a single coherent clip, which ByteDance says eliminates those artifacts.

The model now accepts up to 50 reference materials at once, combining images, audio tracks and video clips. That's a fourfold increase over Seedance 2.0's limit of 12 inputs, and is meant to help creators maintain consistency of characters, brand and style in more complex scenes with multiple subjects.

A new feature is local clip editing: users can swap out one specific element of a frame, such as a character's outfit, a background element or a product in a scene, without regenerating the entire clip. ByteDance is also touting native 4K resolution, though so far that has only been officially confirmed for Seedance 2.0.

Distribution and business scale

The model is reaching business customers through the Volcano Engine cloud and its Model Ark API, and is eventually meant to arrive in ByteDance's consumer apps as well, meaning CapCut, Dreamina and Doubao. CapCut has more than 400 million monthly active users, giving Seedance potentially the widest distribution reach of any video model on the market.

At the FORCE conference, Volcano Engine also revealed that the Seedance platform has already reached $2 billion in annual recurring revenue from business services, showing how quickly generative video is turning into an actual commercial product rather than just a tech demo.

The shadow of the Hollywood dispute

The launch of 2.5 comes under the shadow of controversy surrounding its predecessor. When Seedance 2.0 debuted in February 2026, realistic clips featuring well-known actors spread rapidly online, including a clip depicting a showdown between Tom Cruise and Brad Pitt. That triggered a wave of legal notices from major film studios, with Disney's lawyers describing it outright this way:

Virtual looting - Disney's lawyers, on Seedance 2.0's use of the company's intellectual property

Under pressure from the Motion Picture Association, ByteDance paused the global rollout of Seedance 2.0 and added filters blocking the generation of recognizable faces of real people as well as protected fictional characters. The US market was left out of the model's first wave of availability as a result. Those legal disputes have still not been formally resolved.

Alongside the launch of 2.5, ByteDance announced a new platform for commercial copyright licensing, meant to let creators and studios legally offer their material for training and content generation in exchange for payment. The program's first partner is Hong Kong director and actor Stephen Chow, a signal that the company wants to build relationships with the film industry rather than fight it in court.

What it means for the Polish market

For Polish creators, advertising agencies and video producers, Seedance 2.5 marks another leap in capability in the generative video segment, alongside OpenAI's Sora, Google's Veo and Kuaishou's Kling. Longer native clips and local editing could meaningfully cut production time for short-form ads and social media content, though copyright questions around the likenesses and brands used remain unresolved at the EU level as well.

Worth noting: starting in August 2026, the EU's AI Act rules on labeling content generated or substantially modified by artificial intelligence come into force. Companies using tools like Seedance to produce material for the European market will need to account for these transparency requirements, regardless of where the model itself was built.

Sources: ByteDance Seedance 2.5 Launches This Week: 30-Second AI Video Carries Copyright Cloud (techtimes.com), ByteDance's Seedance 2.5 breaks the 30-second barrier for AI video generation (the-decoder.com), Seedance 2.5: ByteDance's 30-Second AI Video Model (digitalapplied.com), ByteDance Seedance 2.5 Launches in July (news.aibase.com)