ByteDance Launches Seedance 2.5 Amid Ongoing Hollywood Copyright Dispute

ByteDance began the public rollout of Seedance 2.5 in early July, a new version of its video generation model and the first in the family to natively produce 30-second clips without stitching together separately generated segments. The launch coincides with an unresolved legal dispute with major American film studios over the model's previous version.

What's new in the model

The biggest technical change is native generation of 30-second shots in a single pass, with scene and pacing changes happening within the same clip. Previously, creators had to stitch together several shorter segments, which often produced visible jumps in style or character continuity. The second new feature is local region editing: a creator can redraw a single element, such as changing a character's clothing, the background, or a product in a scene, without regenerating the entire piece.

This addresses one of the most costly problems in AI video production, where fixing a single minor error previously meant regenerating the entire clip from scratch, with the risk that the rest of the footage would come out differently than the first time. The model also accepts up to 50 simultaneous references in the form of images, video, and audio files, intended to make it easier to maintain character and style consistency across longer productions.

The shadow of the Hollywood dispute

The Seedance 2.5 launch comes in the shadow of an unresolved legal conflict. When Seedance 2.0 debuted in February 2026, the Motion Picture Association sent ByteDance a cease-and-desist letter, and Disney, Warner Bros, Paramount, Sony, and Netflix each sent their own letters on the matter. The allegation is that the model was trained on copyrighted film and television material without the rights holders' consent.

The studios suspect the training data may have included clips from movies, TV series, and music videos that users uploaded to TikTok's library. ByteDance has not publicly disclosed what material was used to train the model, limiting its communication to descriptions of the model's capabilities rather than the origin of its data. In response to the allegations, the company added content filters and C2PA-compliant watermarks that track material provenance, but it has not settled with any of the studios in court.

Seedance 2.5 is therefore launching without confirmation that the training data or generation safeguards fully address the Motion Picture Association's specific concerns. That means commercial use of the model by production studios and brands currently carries legal uncertainty until the litigation over the previous version is resolved.

Where it stands among rivals

Seedance 2.5 enters a market where Google's Veo 3.1, Kuaishou's Kling 3.0, and Alibaba's upcoming Wan 3.0 are already competing. ByteDance's advantage lies in integration with its own app ecosystem, above all CapCut, which gives the model instant access to hundreds of millions of active users without having to build separate distribution.

For Polish creators, marketing agencies, and production studios using AI video tools, this means another option with a genuinely longer native shot length and cheaper correction edits, but also the need to weigh legal risk from commercial use of training material of unclear origin, especially in productions aimed at markets where copyright enforcement is stricter than in China.

Sources: ByteDance's Seedance 2.5 breaks the 30-second barrier for AI video generation (the-decoder.com), AI Copyright Litigation Risk and the Seedance 2.0 Controversy in Hollywood (millershah.com)