Seedance 2.5: a 30-second video in a single segment
On 23 June 2026, at the Volcano Engine FORCE 2026 conference in Beijing, ByteDance officially announced its new video model, Seedance 2.5. Volcano Engine president Tan Dai unveiled it from the stage. The model is currently in a closed enterprise beta, with a public release promised for early July.
What is officially confirmed
- 30 seconds of native video in a single segment. Not a stitch of short clips, but one continuous generation — with scene and tempo changes. For creators, that means a whole scene or a long take in a single pass.
- Up to 50 multimodal references per task. Images and video clips for precise control over style, character consistency and narrative logic. For comparison, version 2.0 allowed up to 9 images, 3 videos and 3 audio clips.
- Native 4K for the Seedance 2.0 line — right now. The update takes effect immediately, without waiting for 2.5, and includes 10-bit color.
- 3D white-box preview. The first video model in the industry with this feature: before the final render you can build a rough 3D previsualization of movement and camera, so you don't burn generations on revisions. It is a step toward a director's tool rather than a slot machine.
Part of a bigger package
Seedance 2.5 wasn't shown alone. At the same conference ByteDance presented a suite that covers the entire content pipeline in one announcement: Seedance 2.5 for video, Seedream 5.0 for images and the multimodal Doubao 2.1. Detailed specs for Seedream 5.0 weren't revealed on stage — we're waiting for the official page.
A curious detail
According to sources, a 2.1 version was originally in the works, but at the last moment the team decided to skip the interim number and ship 2.5 straight away — to pull ahead of competitors.
Price and what's only rumored
No price has been named yet. As usual, a reference point should appear on the Volcano Ark platform along with the release.
Separately, there are rumors that are not officially confirmed — treat them as expectations, not facts:
- a 60-second ceiling (30 was the figure actually announced);
- roughly 50% cheaper pricing;
- a separate lightweight Mini version;
- near-real-time speed;
- carrying a character across generations without re-supplying references.
What it means for creators
The point of Seedance 2.5 isn't just "longer video". Thirty seconds in one segment remove the seams and desync that give away AI stitching, while the 3D previsualization and fifty references turn generation from a lottery into a controllable process: a rough motion-and-camera draft first, the final render second. That shifts AI video toward a working director's tool.
Where to try it
Seedance models already show up in aggregators from our AI tools rating — for example the video aggregator Pollo AI. After the public 2.5 release it will most likely appear there too. While we wait, you can find the right service for your task in our guide "Which AI Tool to Choose in 2026".
All figures and dates are as of the announcement. We'll update this article when the public release ships and pricing is known.