Seedance 2.5: Release Date, Features & Where to Try It First

What Is Seedance 2.5?

Seedance 2.5 is the next version of ByteDance's video generation model, announced on June 23, 2026 at the Volcano Engine FORCE conference in Beijing. If Seedance 2.0 was the model that made AI video feel like real filmmaking (multimodal references, native audio, genuine prompt adherence), 2.5 is the version aimed at finished productions: clips twice as long, four times as many reference inputs, and the ability to edit a shot instead of re-rolling it.

The announcement came alongside a broader ByteDance model refresh: Seedance 2.0 itself picked up 4K output, a new image model called Seedream 5.0 was revealed, and the Seed 2.1 language model family went live. But Seedance 2.5 was the headliner, and it's the one that matters if you make AI videos.

One important caveat before the details: ByteDance has demoed these capabilities live but hasn't published full official specs yet. Some numbers below come from early API ecosystem documentation and could shift by general availability. We've marked what's confirmed and what's reported.

Release Date and Availability

Here's the timeline so far:

• June 23, 2026 — announced at the Volcano Engine FORCE conference, with live demos including localized editing
• Late June 2026 — global enterprise beta
• July 2026 — public rollout through API providers began; early playgrounds are already live on some platforms
• Now — API access is arriving provider by provider; full official specs and pricing are still landing

If you want to be first in line, the provider situation matters more than the official date, which brings us to the practical question.

Where to Try Seedance 2.5 First (and Cheapest)

Seedance models reach most creators through API platforms, not through ByteDance directly. Two things are worth knowing right now:

Kie.ai has confirmed Seedance 2.5 support is coming to its unified Seedance API (which already runs Seedance 2.0 and 2.0 Mini), with free credits for new users to test it at launch. Kie.ai is currently one of the cheapest places to run Seedance 2.0, and it's positioning 2.5 the same way: affordable per-generation pricing, one API key, webhook support. If you want day-one access, create a free Kie.ai account now and you'll have credits ready when the model flips on.

Early Seedance 2.5 playgrounds have also appeared on other API platforms for text-to-video and image-to-video, so limited hands-on testing is already possible while everyone waits for full API documentation.

The Headline Features

Native 30-Second Videos

Seedance 2.0 tops out at 15 seconds. Seedance 2.5 generates 4 to 30 seconds natively, in one pass, with no stitching. That's long enough for a complete story arc: setup, development, payoff. ByteDance also showed a beta long-video mode that extends clips to 3 minutes, though that's further out.

This matters more than it sounds. Stitched clips drift: characters change slightly between segments, lighting shifts, motion resets. A native 30-second generation keeps one consistent world for the entire duration.

50 Multimodal References

This is the biggest structural change. Seedance 2.0 accepts about 12 reference inputs. Seedance 2.5 accepts up to 50, and early API documentation reports the split as:

References can reportedly be tagged with a role, so the model knows whether an input pins a character's identity, a product's look, a brand color palette, a camera style, or a voice. ByteDance has also shown 3D white-model support, meaning you can block out scene layout and camera paths with a rough 3D model before generating.

For anyone who has fought AI video for character consistency, this is the feature to watch. Thirty image slots means a full character turnaround sheet, wardrobe references, and environment plates in a single generation.

Local Editing

The demo that got the most attention: change one region of a finished clip without regenerating the rest. ByteDance showed a lipstick ad where the shade was swapped in place while the actress's performance, the camera move, and the lighting stayed identical.

Until now, a 95%-right generation meant re-rolling and hoping. Local editing means fixing the 5%. It's the difference between a slot machine and an editor. Whether and how API providers expose this is still unconfirmed, but the capability is real and was demonstrated live.

Native Audio, Deeper Integration

Seedance 2.0 already generates synchronized audio. In 2.5, audio is co-processed in the same latent space as the visuals, which in practice means tighter lip-sync and sound effects that land exactly on the action that causes them. Combined with audio reference clips (feed it a song or a voice), soundtrack-driven videos become a native capability instead of a post-production step.

Quality and Control Upgrades

The rest of the changelog, briefly:

• Native 4K output (early API tiers are launching at 480p and 720p first) and 10-bit color for cleaner grading
• Around 20% better prompt adherence, per ByteDance, meaning fewer wasted generations
• Stronger cross-shot character consistency for multi-scene storytelling
• Advanced camera control, including rack focus, crane moves, and whip pans (our camera movements library covers the vocabulary that carries over)
• Licensed film IP program with revenue sharing, letting platforms build experiences around authorized characters

What's Still Unconfirmed

An honest list, because a lot of coverage blurs this line:

• The exact 30/10/10 reference split comes from early ecosystem docs, not an official ByteDance spec sheet
• Final API pricing is unpublished; the per-second estimates above are third-party
• Local editing has only been shown in official demos; no independent hands-on reviews exist yet
• 4K API access appears to be arriving after the initial 480p/720p tiers
• The 3-minute long-video mode is beta with no public timeline

We'll update this post as official specs land.

Frequently Asked Questions

Is Seedance 2.5 available right now? It's rolling out. Enterprise beta started in late June 2026, and public API access is arriving through providers in July. Kie.ai has confirmed support is coming with free trial credits for new users.

How long can Seedance 2.5 videos be? 4 to 30 seconds natively, twice the 15-second ceiling of Seedance 2.0. A beta long-video mode extends to 3 minutes.

What does Seedance 2.5 cost? Billing is per second of generated video and scales with resolution and reference count. Early estimates put a 30-second clip at lower settings around $0.66 to $1.80; official pricing isn't published yet.

Should I wait for 2.5 or use Seedance 2.0 now? Use 2.0 now. It's proven, widely available, and still one of the best AI video models you can run today. Everything you learn about prompting it carries directly into 2.5. Our Seedance 2.5 vs 2.0 comparison breaks down exactly what changes.

Do I need to write prompts differently for 2.5? The fundamentals carry over, but 30-second clips reward time-segmented, multi-beat prompts, and the expanded reference system changes how you structure inputs. See our Seedance 2.5 prompt guide.

Star in a Cinematic AI Video Today

While Seedance 2.5 rolls out, Seedance 2.0 is already generating some of the best AI video on the internet, and you don't need an API key or a prompt guide to use it. Starrd's templates run pre-engineered Seedance 2.0 prompts: pick a scene, upload a photo or two, and get a cinematic video starring you.

Browse Starrd Templates

Red carpet arrivals, World Cup pitch invasions, future babies, and more — one tap each.

Try It