Seedance 2.5 Release Date: Confirmed Timeline and What Shipped

If you are still searching for a Seedance 2.5 release date, the search is over — and a lot of pages have not caught up. Plenty of articles still describe Seedance 2.5 as "coming soon" or "expected in Q3." It shipped.

Seedance 2.5 launched on July 31, 2026. ByteDance first announced it on stage at the Volcano Engine FORCE conference on June 23, 2026, and it is now rolling out across Jimeng AI, the Pro tier of Doubao, and the ModelArk API.

This article is built from primary sources only: ByteDance's own Seed model pages, BytePlus ModelArk documentation (where the model ID and rate card live), and dated reporting from TechNode. Everything you can verify yourself is linked, and the full list is in Sources at the end.

What This Article Solves

Release-date searches for AI models are unusually noisy because pages get written during the rumor window and never updated. So you end up reading a June-dated article that says "expected in July" and cannot tell whether it actually happened.

This page gives you three things those pages do not: the confirmed dates with the receipts, the exact model identifier that proves it exists, and a checklist of where the model is actually reachable right now — not where it was "expected" to appear.

The Confirmed Timeline

The model ID is the detail worth internalizing. ByteDance encodes the release date directly into its identifiers — 260628 is June 28, 2026, the same convention as dreamina-seedance-2-0-260128 (January 28, 2026). If a model has a published ID in the BytePlus pricing docs, it is real and billable. If it does not, it does not exist yet, no matter what a roundup post claims.

What Actually Shipped

The headline feature answers the loudest complaint about the previous generation.

30-second native generation. Seedance 2.0 capped at 15 seconds. ByteDance's official Seedance 2.5 page describes creating videos up to 30 seconds in a single generation, with the option to extend for longer narratives. This is single-run output, not stitched clips — the distinction that matters for shot continuity.

Massively expanded reference input. Per TechNode's launch report, a single input accepts up to 30 images, 10 videos, and 10 audio clips. That is the difference between prompting a look and actually art-directing one.

Editing controls, not just generation. White-model control lets you drive geometry and camera movement with an untextured reference before the model handles surfacing. Green-screen editing ships alongside it. ByteDance also describes improved responsiveness to audio and visual editing requests, and more precise interpretation of reference videos.

Audio-video joint generation. Carried forward and refined from Seedance 2.0, which introduced joint audio-video generation with text, image, audio and video all accepted as references.

Where You Can Use It Today

Four routes, and they became available at different times — which is why "is it out?" got confusing:

1. Jimeng AI — ByteDance's consumer creation app, part of the launch rollout.
2. Doubao Pro — the paid tier of ByteDance's assistant.
3. ModelArk API (international) — model ID dreamina-seedance-2-5-260628, with published per-token rates. A live rate card is the strongest possible proof of availability.
4. Volcano Engine (China) — the domestic platform, where Seedance model IDs typically appear first.

If you just want to see whether current-generation Seedance output fits your style without setting up a Volcano Engine or BytePlus account, you can run Seedance-family text-to-video and image-to-video jobs directly in the AI video generator on this site — credit costs per generation are listed on the pricing page, and there is a dedicated Seedance 2.5 model page with the current capability breakdown.

Why the Version Number Jumped to 2.5

This trips people up, so it is worth being precise about it.

ByteDance was reportedly preparing a 2.1 release and moved to the 2.5 label before shipping. The practical consequence is that a larger-than-usual capability jump — longer clips, deeper reference control, editing — arrived under a half-version number rather than a whole one.

That matters for how you search. A half-version increment sounds incremental, so people scanning release notes skip it. In this case the increment includes the single change most creators had been asking for since the 15-second cap: double the native duration. If you evaluated the family before July 2026 and moved on, 2.5 is the release worth re-testing.

The verification habit below is what keeps you from having to guess about any of this.

How To Verify a Seedance Release Yourself

Five checks, ordered by how fast they settle the question:

1. Search the BytePlus pricing page for the version string. A billable model ID means it shipped.
2. Check the Volcano Engine model list for the Chinese-market identifier.
3. Look for a dedicated page on seed.bytedance.com. Each shipped generation gets one.
4. Check FORCE conference coverage. ByteDance uses this venue for flagship video model reveals — 2.5 was announced there.
5. Cross-check dated tech press such as TechNode, which reported the launch with specifics rather than speculation.

Undated blog posts and social threads fail all five. That is how "Seedance 2.5 is coming in Q3" survived weeks past the actual launch.

FAQ

When was Seedance 2.5 released? July 31, 2026. It was announced at the Volcano Engine FORCE conference on June 23, 2026.

Is Seedance 2.5 available now? Yes — through Jimeng AI, Doubao Pro, the ModelArk API (dreamina-seedance-2-5-260628), and Volcano Engine.

How long can Seedance 2.5 videos be? 30 seconds in a single generation, with the option to extend for longer sequences. Seedance 2.0 capped at 15 seconds.

What is the difference between Seedance 2.5 and 2.0? Duration (30s vs 15s), far larger reference capacity (up to 30 images, 10 videos, 10 audio clips per input), and new editing controls including white-model control and green-screen editing.

Which Seedance model is current? Seedance 2.5, launched July 31, 2026. The 2.0 series (standard, Fast and Mini) remains available as lower-cost tiers.

Does Seedance 2.5 generate audio? Yes. It is an audio-video joint generation model, building on the capability introduced in Seedance 2.0.

Sources

Primary sources for every date and specification above. Model capabilities change; verify before making production commitments.

1. ByteDance Seed — Seedance 2.5 — official capability page: 30-second single generation, extension, white-model control, green-screen editing.
2. ByteDance Seed — Seedance 2.0 — audio-video joint generation and multimodal reference support.
3. ByteDance Seed — Seedance — Seedance 1.0 baseline capabilities and published benchmarks.
4. BytePlus ModelArk — Pricing — the model ID dreamina-seedance-2-5-260628 and its live rate card.
5. BytePlus ModelArk — Video Generation API — API surface for video generation tasks.
6. BytePlus ModelArk — seedance-1.0-pro-fast — resolution, duration and frame-rate conventions across the family.
7. TechNode — ByteDance launches Seedance 2.5 — July 31, 2026 launch date, reference limits, Jimeng AI and Doubao Pro rollout.
8. Volcano Engine model list — Chinese-market Seedance model identifiers.
9. Seedance 1.0 technical report (arXiv 2506.09113) — the foundational paper and its stated deployment path.
10. BytePlus ModelArk product page — platform-level availability and account setup.

Last verified: August 4, 2026, by the seedance3ai.video editorial team, which tracks ByteDance video model releases across BytePlus and Volcano Engine documentation. This page is updated when a new model ID or official announcement appears in the channels listed above.