ByteDance unveiled Seedance 2.5 on June 23, 2026 at its Volcano Engine FORCE conference in Beijing. The headline: a single, continuous 30-second clip generated at native 4K from one prompt — roughly double the length of Seedance 2.0, without stitching shots together.
An enterprise beta is live now, with a public launch targeted for early July 2026. Here's everything that's been announced, how it compares to Seedance 2.0, and — honestly — which numbers are still preview claims rather than benchmarked results.
A note on accuracy: most figures below are ByteDance's own preview claims. The model isn't fully public yet, so nothing here has been independently benchmarked. We'll update this post as 2.5 ships and real tests land.
What's new in Seedance 2.5
30-second clips in a single pass
Seedance 2.0 capped each generation at around 15 seconds. Seedance 2.5 generates a continuous 30-second shot natively — a complete scene, ad, or story beat that holds from the first frame to the last in one pass, instead of being assembled from shorter clips. For anyone fighting cross-shot consistency today, native length is the upgrade that matters most.
Native 4K, 10-bit color
The model renders at 4K natively rather than upscaling from a lower resolution — a distinction that matters for real production pipelines. It also supports 10-bit color depth for smoother gradients and more headroom in post for color grading.
Up to 50 reference inputs
This is the standout jump. Seedance 2.5 accepts up to 50 multimodal reference inputs — images, audio clips, 3D white models, and style references — up from 12 in Seedance 2.0. More references means tighter control over character identity, products, and visual style across a long clip.
Native audio sync
Audio is now co-processed in the same latent space as the visual signal, so onscreen actions and their sound effects are synchronized natively rather than aligned after the fact.
3D white-box preview
A new white-box preview function lets you generate a low-fidelity 3D animation to check blocking and camera before committing to a full-quality render — a real time-saver when each final render is expensive.
~20% better prompt adherence
ByteDance claims 20% better prompt adherence, meaning fewer generations before a usable result. The company also targeted near-real-time, streaming-style generation where you watch the video build as it renders.
Seedance 2.5 vs Seedance 2.0
| Seedance 2.0 | Seedance 2.5 (preview) | |
|---|---|---|
| Clip length | ~15s | 30s, single pass |
| Resolution | upscaled | native 4K, 10-bit |
| Reference inputs | 12 | up to 50 |
| Audio | separate step | native in-latent sync |
| Prompt adherence | baseline | +20% (ByteDance) |
| Preview | — | 3D white-box |
ByteDance skipped the intermediate version numbers — jumping straight from 2.0 to 2.5 — to signal what leadership framed as a "generational leap." CEO Liang Rubo said that "climbing the AI summit is the company's top priority, with its model-as-a-service business evolving into a foundational operation."
Release date and availability
- Enterprise beta: live now (since the June 23 announcement).
- Public launch: targeted for early July 2026.
- U.S. / global: no firm timeline. Broader rollout reportedly hinges on resolving copyright disputes with major studios.
So if you're reading this before July, you can't generate on 2.5 yet unless you're in the enterprise beta.
What you can make today (on Seedance 2.0)
While 2.5 rolls out, Seedance 2.0 already produces cinematic, reference-driven, multi-shot results — and the techniques carry straight over. A few real prompts from the Scenic gallery that preview where 2.5 is heading:
an FPV rooftop skateboarding sequence
a two-character demon-hunting sequence
a night-rain cycling commercial
Browse the full Scenic gallery for 4,000+ copy-ready Seedance prompts, and read how to write Seedance prompts so your prompts are ready the day 2.5 opens up.
Frequently asked questions
When is Seedance 2.5 released? It's in enterprise beta now (announced June 23, 2026), with a public launch targeted for early July 2026. There's no confirmed U.S. timeline.
How long are Seedance 2.5 clips? Up to 30 seconds in a single continuous generation — about double Seedance 2.0's ~15 seconds.
Is Seedance 2.5 really 4K? ByteDance says it renders at native 4K (not upscaled) with 10-bit color. That's a preview claim until the model is publicly benchmarked.
How many reference inputs does it support? Up to 50 multimodal inputs (images, audio, 3D models, style references), up from 12 in Seedance 2.0.
Can I use Seedance 2.5 now? Only via the enterprise beta until the early-July public launch. In the meantime, Seedance 2.0 prompts on Scenic are free to browse and copy.
See the demos
ByteDance's Seedance 2.5 preview clips are circulating on X — see them in this thread from @xiaohu to watch the 30-second 4K output for yourself.
