Update — August 7, 2026. Everything this post said was "coming" has now shipped: Seedance 2.5's public API went live (served by FAL, $0.0214 per 1,000 tokens), and Seedance 2.5 is now live on ImagineToVideo — set as the default model, priced in credits by duration and resolution. One open question from release day resolved the other way: the live API outputs 480p/720p only, so there is no 4K or even 1080p tier — Seedance 2.0 keeps the resolution edge. Try it in the generator; the FAQ below has been refreshed to the live facts, and the rest of the article stands as the release-day record.
On July 31, 2026, ByteDance's Seed team officially released Seedance 2.5, its new-generation video creation model. When we previewed Seedance 2.5 after the June 23 FORCE announcement, the model was in enterprise beta with an "early July" target. The launch landed a few weeks later than that target — but it landed. Seedance 2.5 is now live inside ByteDance's own products, and the official spec sheet is finally something we can quote rather than reconstruct from conference reporting.
This post covers what ByteDance actually announced at release, what changed against the June preview coverage, what still is not public — pricing and API access, most importantly — and our concrete plan for bringing Seedance 2.5 to ImagineToVideo.
What ByteDance officially announced
The release announcement leads with a positioning statement: Seedance 2.5 goes beyond longer single-pass generation, aiming to deliver "the journey from idea to finished video" with advancements in long-form storytelling, multimodal reference, and editing. Three highlights carry the release:
1. 30-second audio-video clips in a single pass
Users can generate high-quality, 30-second audio-video clips in a single pass — a complete story in one take, picture and sound together. The official project page adds that a clip can be extended twice for a longer narrative. For context: Seedance 2.0, which you can use on ImagineToVideo today, generates 4–15 seconds per clip with native audio.
2. Up to 30 images, 10 video clips, and 10 audio clips as reference
A single generation accepts up to 30 images, 10 video clips, and 10 audio clips as reference material. That is the precise official breakdown of the "50 multimodal inputs" figure that circulated after the June preview — and notably, the official list is images + video + audio (early reporting said images, video, and text). ByteDance says the model can understand reference video well enough to capture intent, framing, and cinematic language.
3. Timestamp-level editing
Seedance 2.5 offers timestamp-level control for targeted editing of audio and video content — fix a specific moment without regenerating the whole clip. The official page also lists white-model control, green-screen editing, professional camera movement, and performance segmenting. This is more specific than the "region-level editing" phrasing from the June coverage; the official framing is about targeting moments in time.
Still the vendor's word. These are ByteDance's official claims. There is no public API yet, so no third party — including us — has independently benchmarked any of them. We flag every spec on this page accordingly, and we will publish hands-on results once we can actually run the model.
What the release did NOT include
Just as important as the headline features:
- No resolution spec. Neither the release announcement nor the official project page mentions 4K or any output resolution for Seedance 2.5. Some third-party reports claim "native 4K" — we are not repeating that, because ByteDance's own release materials do not say it. (Our June preview flagged the same issue: the FORCE-era 4K upgrade was attributed to Seedance 2.0, not 2.5.) ⚠️ Resolution and frame rate: officially undisclosed.
- No public API. Seedance 2.5 is live inside ByteDance's own products — Jimeng AI and Doubao. API service is described as coming to Volcano Ark, with no date attached. Until that ships, no third-party platform can run the real model.
- No pricing. There is no official rate card. ⚠️ Any per-second price you see quoted for Seedance 2.5 today is speculation.
Seedance 2.0 → 2.5: the release-day comparison
| Dimension | Seedance 2.0 (available on ImagineToVideo today) | Seedance 2.5 (released, official claims) |
|---|---|---|
| Max single-clip duration | 4–15 seconds | 30 seconds in one pass, extendable twice |
| Audio | native synchronized audio | audio-video generated together in the single pass |
| Reference inputs | up to 9 images + 3 video clips | 30 images + 10 video clips + 10 audio clips |
| Editing | video-to-video re-rendering | timestamp-level targeted editing, white-model control, green-screen |
| Resolution | 480p / 720p / 1080p | not officially disclosed ⚠️ |
| Where to use it | ImagineToVideo, FAL, Replicate | Jimeng AI, Doubao; API "coming" to Volcano Ark |
| Pricing | published (credits here, per-second on API hosts) | not published ⚠️ |
The honest summary: the leap is about duration, reference budget, and editing control. Resolution, frame rate, and cost remain open questions that only the API launch will answer.
Where you can use Seedance 2.5 today
Two places, both ByteDance's own: Jimeng AI (the creative suite) and Doubao. Access terms and any free quotas there are set — and changeable — by ByteDance. If you want the honest free-access picture, we keep it updated on our Seedance 2.5 free access page.
For developers: there is no endpoint to call yet. We track the API rollout, what ByteDance has said about Volcano Ark, and what you can prototype against in the meantime on our Seedance 2.5 API status page.
Our integration plan
In the June preview we committed to a four-step plan. The release doesn't change the plan — it starts the clock on it:
- Track the official API. Seedance 2.5 is released, but the Volcano Ark API is still "coming." We evaluate the moment it — or a hosted provider like FAL or Replicate — opens access. This is the gating step, and its timing belongs to ByteDance.
- Benchmark the claims. 30-second single-pass audio-video, the 30/10/10 reference budget, timestamp editing: we test all of it against real prompts before anything goes user-facing, and we publish what we find.
- Calibrate credit pricing. Once the official rate card exists, we map it to credits the same way we did for Seedance 2.0 — no surprises.
- Roll out. Gated release first, then general availability across the platform.
And one thing we already shipped today: Seedance 2.5 is now pre-listed in the ImagineToVideo model picker as Coming Soon. You will see it at the top of the model list in the generator — visible, labeled, and not selectable until it actually works. When it switches on, your existing prompts and reference workflows carry over.
We are deliberately not committing to a date. The dependency is the official API and pricing, and both are ByteDance's to ship.
What to generate while you wait
The Seedance you can run today is a genuinely strong model:
- Seedance 2.0 — 4–15 second clips with native synchronized audio at up to 1080p; text-to-video, image-to-video, and video-to-video; up to 9 reference images and 3 reference videos. Try it free — new users get 1 free generation.
- Seedance 2.0 Fast — the same feature set at a lower credit cost, built for iterating on motion and framing before a final render.
- Multi-scene chaining — while 30-second single takes stay exclusive to 2.5, chaining 4–15 second takes in the multi-scene tool covers longer stories today.
For the full release-day spec sheet with verification status on every claim, see our Seedance 2.5 overview page.
FAQ
Is Seedance 2.5 officially released? Yes — and it is now generally accessible. ByteDance's Seed team officially released Seedance 2.5 on July 31, 2026, and since August 7, 2026 a public API is live (served by FAL), which is what ImagineToVideo runs in production. It also remains inside ByteDance's own products, Jimeng AI and Doubao.
What are Seedance 2.5's headline features? Three headline capabilities, now verifiable on the live API: 30-second audio-video clips generated in a single pass, lip-synced dialogue from quoted lines in the prompt, and reference-driven generation (the API accepts up to 30 images, 10 video clips and 10 audio clips combined; ImagineToVideo currently exposes 9 images + 3 videos).
Does Seedance 2.5 support 4K? No. The live API schema offers exactly two output tiers — 480p and 720p — with no 1080p or 4K option. Third-party "native 4K" claims were never backed by ByteDance's release materials, and the shipped API confirms it: Seedance 2.0 actually retains the resolution edge with its 1080p tier.
How much does Seedance 2.5 cost? API pricing is now published: FAL bills $0.0214 per 1,000 tokens, which works out to roughly $0.47 per second at 720p and $0.22 at 480p with audio, with reference runs that include video billed at a 0.6x multiplier. On ImagineToVideo, generations are priced in credits by duration and resolution, with reference and video-to-video runs billed at a discounted rate.
When will Seedance 2.5 be available on ImagineToVideo? It already is — live since August 7, 2026 and set as the default model in the generator. Text-to-video, image-to-video and video-to-video with references all work, new users get a free generation to try it, and Seedance 2.0 and 2.0 Fast remain available alongside it at lower credit costs.
