Seedance 2.5 is ByteDance's newest AI video model. It renders a native 30-second single shot at 4K, accepts up to 50 multimodal references, and lays down synchronized audio in the same pass. Start with the free AI video generator below.
Write a prompt, pick a ratio, resolution and duration, then generate. This AI video generator supports reference images, first and last frame control, and native audio.
Currently running Seedance 2.0 and Seedance 2.0 Fast. The 2.5 model will appear in the model list here as soon as it is available.
Access is still rolling out
The model shipped first on Jimeng AI and Doubao Pro, and ByteDance has not opened a public API yet. We are integrating it now. Until it lands, the AI video generator below runs on Seedance 2.0 and Seedance 2.0 Fast — so you can create today and switch the moment the new model goes live here.
Seedance 2.5 is the successor to Seedance 2.0 in ByteDance's family of AI video models. ByteDance previewed this AI video generator at the Volcano Engine FORCE conference on June 23, 2026, then shipped it on July 31, 2026.
The headline change is length. Most video models cap one generation at five to fifteen seconds, so a thirty-second scene means stitching several clips together and hoping the character survives the cut. A Seedance 2.5 video renders the full thirty seconds in a single pass — which is why ByteDance calls it native rather than extended.
The second change is reference density. One request accepts up to fifty multimodal reference inputs — images, video, audio and text mixed freely — so a character, a product or a brand palette can be pinned across an entire shot instead of drifting scene by scene.
Is Seedance 2.5 out yet? Yes — here is where this AI video generator stands as of August 1, 2026.
ByteDance unveiled the model at the Volcano Engine FORCE conference, alongside a 4K upgrade for Seedance 2.0.
Seedance 2.5 went live on Jimeng AI and Doubao Pro, rolling out gradually rather than to every account at once.
CapCut and Dreamina already expose it. Several other creative tools have announced support without shipping it yet.
ByteDance has not published open endpoints or rates. Teams that want to build on the model are still waiting.
Five capabilities separate this AI video generator from the models that came before it.
The workflow is the same whether you run the 2.5 model or Seedance 2.0 in the AI video generator above.
Name the camera, the subject, the action and the mood. Vague prompts drift; "handheld mini-DV footage, slight hand shake, natural zoom" gives the model something to hold onto for thirty seconds.
Upload a first frame, a first and last frame pair, or a set of reference images. References are what keep a character recognisable from second one to second thirty.
Pick the aspect ratio for wherever the video will be published, then choose resolution and duration. Higher resolution and longer clips cost more credits.
The AI video generator runs each request as an asynchronous job. Watch the back half of the result closely — that is where consistency problems show up first.
This AI video generator rolled out through ByteDance's own apps first, then through partners. Here is the honest picture as of August 1, 2026 — availability changes weekly.
| Platform | Status | Notes |
|---|---|---|
| Jimeng AI / Doubao Pro | Available | ByteDance's own apps. First to get the model, but access is gradual and region-limited. |
| CapCut | Available | Creators are publishing 30-second clips daily. Prompt filtering is strict. |
| Dreamina | Available | BytePlus-side rollout aimed at longer AI video. |
| Other creative tools | Coming soon | Several video platforms have announced support without shipping it yet. |
| Seedance2 | Integrating | Seedance 2.0 and Seedance 2.0 Fast work now, with a documented API. The 2.5 model is being added. |
Seedance 2.0 is not obsolete. As an AI video generator it is faster, cheaper and available everywhere, and for clips under fifteen seconds the quality gap is smaller than the marketing suggests.
| Capability | Seedance 2.5 | Seedance 2.0 |
|---|---|---|
| Max single-pass length | 30 seconds native | Up to 15 seconds |
| Reference inputs | Up to 50, multimodal | Images, first and last frame |
| Resolution | Up to 4K native | Up to 4K |
| Audio | Synchronized in one pass | Native audio |
| Regional editing | Yes, redraw part of a frame | Not available |
| Availability | Rolling out, no public API | Widely available, API ready |
Pick Seedance 2.5 when the shot has to be one continuous thirty-second take, or when many references need locking at once. Pick Seedance 2.0 when you want speed, lower cost, predictable availability, or API access today.
Every page about this AI video generator lists the wins. These are the failure modes creators are actually hitting, and they matter more for your budget than the benchmark numbers do.
What creators are actually building with this AI video generator in its first week.
Every clip below came out of the AI video generator on this page — no stock footage, no borrowed samples.
These are Seedance 2.0 results. We will add Seedance 2.5 examples here once the model is integrated.
Independent creators putting this AI video generator through real tests — worth watching before you commit credits to a long generation.
Videos are published on YouTube by their original creators and embedded here.
ByteDance has not published open rates for the new model yet. What we can tell you is what our AI video generator costs today, running Seedance 2.0 and Seedance 1.5 Pro.
There is no public API from ByteDance yet, which is why every "Seedance 2.5 API pricing" page you find is quoting estimates rather than live rates.
Yes. ByteDance released it on July 31, 2026 on Jimeng AI and Doubao Pro, after announcing it at the Volcano Engine FORCE conference on June 23, 2026. Access is rolling out gradually, so not every account sees it immediately.
July 31, 2026 for ByteDance's own apps. Third-party platforms are adding the model on their own schedules, and a public API has not been released.
Thirty seconds in a single native pass, without stitching. An ultra-long beta mode extending single generations to around 180 seconds has been spotted in ByteDance's own app, but it is not generally available.
There is no unlimited free tier. ByteDance's apps meter it, and third-party platforms charge credits or subscriptions. Free credits and trials are common, so you can test the AI video generator before you commit.
Not a public one yet. Pricing figures circulating online are estimates. Our API currently serves Seedance 2.0, Seedance 2.0 Fast and Seedance 1.5 Pro, and will add the newer model when it becomes available.
No. It is a proprietary ByteDance model. There are no official weights on GitHub or Hugging Face, and repositories claiming to expose a Seedance 2.5 API are unofficial.
Use the 2.5 model for continuous thirty-second takes and heavy reference control. Use Seedance 2.0 for speed, lower cost, wide availability and API access. For clips under fifteen seconds the difference is smaller than it looks.
Long-form consistency is the weakest area. Objects vanish, scale shifts and lighting jumps after the first ten to fifteen seconds. Pin anything important with reference inputs, keep the camera move simple, and consider generating in shorter segments when every beat has to be exact.
Generate with Seedance 2.0 today and get free credits to test the AI video generator. When the newer model is integrated it appears in the same AI video generator and the same API — no migration needed.