Quick Answer
Seedance 2.5 is ByteDance’s new flagship AI video model, officially launched on July 31, 2026, with its public developer API opening on August 7. The model generates up to 30-second single-shot clips with native audio and supports up to 50 multimodal reference inputs. CapCut is rolling the model out internationally, though US availability has not been confirmed.
Key Takeaways
- Seedance 2.5 launched on July 31, 2026 and its public developer API opened on August 7, 2026, making it accessible through third-party platforms including CapCut on a rolling basis.
- The model generates up to 30 seconds of continuous, single-shot video with native audio co-processed in the same pass, doubling the 15-second cap of its predecessor, Seedance 2.0.
- Up to 50 multimodal reference inputs per generation (images, video clips, audio, and style guides) let creators lock character appearance, product identity, and scene atmosphere across an entire clip.
- A beta Long Video Mode extends total output to 180 seconds through multi-round generation, though that feature carries an explicit beta label and consistency results may vary.
- US availability on CapCut has not been officially confirmed, unresolved copyright disputes from the Seedance 2.0 rollout remain in the background, and no official public pricing for Seedance 2.5 has been published as of August 10, 2026.
ByteDance’s Seedance 2.5 is the company’s most capable AI video model to date (by ByteDance’s own description), and its public API opened on August 7, 2026, marking a meaningful step beyond the enterprise-only beta period that followed its June announcement. The model sits inside CapCut’s Dreamina interface and is rolling out to subscriber accounts across multiple regions. For US-based creators, the practical situation is more cautious: CapCut’s own rollout page lists Europe, Asia, the Middle East, and South America as the current expansion regions, and ByteDance has not announced a confirmed US launch date for Seedance 2.5.
What is Seedance 2.5, and how does it differ from Seedance 2.0?
Seedance 2.5 is ByteDance’s flagship video model, officially announced on June 23, 2026, at the Volcano Engine FORCE conference in Beijing. ByteDance skipped versions 2.1 through 2.4, framing 2.5 as a generational jump rather than a point release. That framing is reflected in the most significant change: clip duration.
Seedance 2.5 is notable for its ability to generate videos of up to 30 seconds, double the length of the previous model’s 15 seconds. The mechanism behind that extension matters practically. Earlier AI video models, including Seedance 2.0, assembled longer clips by stitching several shorter generations together. Stitching creates visible inconsistencies at the seam points: lighting can shift, character appearance can drift, and audio synchronization can break. Seedance 2.5 generates a full 30-second clip natively in a single pass, eliminating the stitching problem that causes character drift and lighting inconsistency between clips.
Seedance 2.0 introduced a unified multimodal audio-video joint generation architecture, producing video and audio together by one model rather than dubbed afterward, across four input modalities. Seedance 2.5 carries that architecture forward across the longer clip duration. Audio is co-processed in the same latent space as video, not generated separately and synced after. This matters for dialogue-heavy content and any clip where lip sync or ambient sound needs to track precisely with on-screen action.
What are the confirmed technical specifications for Seedance 2.5?
Confirmed capabilities per ByteDance’s official launch post include 30-second high-quality audio-video clips in one pass with multi-round extension, and reference input of up to 30 images, 10 videos, and 10 audio clips per request. ByteDance’s product pages and several third-party API guides describe a broader 50-reference ceiling across all input types combined. Seedance 2.5 supports native generation at up to 30 seconds in standard mode as a single continuous clip without stitching. The beta Long Video Mode extends output to 180 seconds via multi-round generation, maintaining character, scene, lighting, and audio consistency.
Resolution claims vary between ByteDance’s consumer-facing product pages and what the public API currently delivers. CapCut’s official product page states that Seedance 2.5 is live on CapCut and describes native 30-second 4K video generation with up to 50 multimodal references. At the API level, the picture is different. Resolution actually moved in the other direction from the initial announcement: Seedance 2.0 tops out at 1080p, while Seedance 2.5 currently caps at 720p through the API, meaning ByteDance’s original 4K announcement did not make it into the shipped model. The Seedance 2.5 API accepts every integer duration from 4 to 30 seconds, generates at 480p or 720p with 720p as the default, and does not support 1080p. Creators using the model through CapCut’s consumer interface and those accessing it through the developer API should verify the resolution option available to them before committing to a production workflow.
Seedance 2.5 supports prompt input and generation in 11 languages, including English, Chinese, Spanish, Japanese, Korean, and Arabic. Seedance 2.5 ships with 3 generation modes: text-to-video, which produces a clip from a written prompt alone; image-to-video, which animates a still image into a moving scene; and reference-to-video.
What is the reference system, and why does it matter for creators?
The reference system is one of the most practically useful aspects of Seedance 2.5 for commercial and content production work. Seedance 2.5 is built around reference control: up to 50 multimodal references in a single generation let a creator constrain character, product, palette, camera style, and voice simultaneously. The predecessor model, Seedance 2.0, accepted roughly 9 images and 3 video clips per generation. Moving to 50 reference assets per request means a creator can provide a character reference image, a product photo, a brand color-palette guide, an audio sample, and multiple scene references within the same generation job.
A region-level editing feature compounds this capability. Once a master clip is generated with the full reference set, teams can swap specific elements, such as a product variant, a background locale, or a seasonal overlay, without re-running the full generation. For high-SKU consumer brands or teams localizing content across markets, that workflow is substantially faster than what exists today.
That said, the reference budget is a ceiling, not a guarantee of quality. More reference files do not automatically produce a better result. First-pass usability depends on choosing the right route and giving character, product, scene, motion, and audio references clear roles. Creators new to the model should start with a focused reference set and expand only once the core generation is behaving as expected.
How does Seedance 2.5 compare to Veo 3.1 and Kling 3.0?
No single model leads every category in the current AI video field, and Seedance 2.5 is no exception. The table below summarizes how the 3 leading models compare across the dimensions most relevant to content creators as of August 2026. All capability claims are sourced from vendor documentation and third-party guides; independent head-to-head benchmarks for Seedance 2.5 specifically are limited at this stage.
| Dimension | Seedance 2.5 | Google Veo 3.1 | Kling 3.0 |
|---|---|---|---|
| Max clip length (native) | 30 seconds | 8 seconds (extendable) | Up to 3 minutes |
| Resolution (consumer) | 4K (CapCut product page); 720p (API) | Up to 4K | Up to 4K |
| Native audio | Yes, co-processed | Yes, audio-native pipeline | Available (weaker integration) |
| Reference inputs | Up to 50 | Limited | Up to 4 images |
| Region/local editing | Yes (Intelligent Edit Mode) | No | Limited |
| US consumer availability | Not confirmed | Yes (Google AI Pro/Ultra) | Yes |
| Best fit | Long-form, heavily referenced brand or narrative content | Hero clips with strict audio-visual quality requirements | High-volume social content at lower cost per clip |
Seedance 2.5 leads on duration and reference scale. Kling 3.0 holds a resolution and cost advantage. Veo 3.1 has an established position on official developer access and audio quality. There is no honest universal winner among the three models. Seedance 2.5 is the most relevant choice to test for a continuous 30-second, heavily referenced scene. The practical approach for most production teams is to select the model based on the specific job, rather than committing exclusively to one platform.
Where and how can you access Seedance 2.5 today?
ByteDance released Seedance 2.5 as a model and creator product on July 31, 2026, then opened the public developer API on August 7, 2026. Seedance 2.5 is available through multiple third-party API platforms from August 7 onward for text-to-video, image-to-video, and reference-to-video workflows.
For consumer access through CapCut, the rollout is regional and tied to subscriber accounts. Seedance 2.5 launched globally on July 31, 2026, debuting first on Dreamina and rolling out to CapCut across Europe, Asia, the Middle East, and South America for subscriber accounts, with broader availability continuing to expand. Seedance 2.5 appears as a selectable model in CapCut’s video editor as it becomes available in your region. If the model does not appear in your account’s model selector, it has not reached your region yet. Checking the selector directly is more reliable than relying on landing-page descriptions, which can describe a planned feature before it is selectable.
For developers, access is available through BytePlus ModelArk or Volcano Engine directly, or through a unified API provider that gives developers a single key and endpoint without requiring a separate ByteDance account. The model is billed per second of generated video. ByteDance has not yet published a specific per-second rate for Seedance 2.5 through its own platforms, so developers should check the Seedance 2.5 API page for current, confirmed pricing rather than relying on third-party estimates. One third-party provider, PiAPI, listed 720p generation at $0.385 per second as of August 7, 2026, but that rate reflects one intermediary’s pricing and may not reflect BytePlus’s direct rates.
What is the US availability situation, and what are the copyright concerns?
US availability for Seedance 2.5 on CapCut has not been officially confirmed as of August 10, 2026. This is not a new complication. The Seedance 2.5 announcement arrived following ByteDance’s voluntary pause of the global rollout of its predecessor, Seedance 2.0, after cease-and-desist letters from every major Hollywood studio over alleged copyright infringement. Those legal disputes remain unresolved.
Copyright filters are active in the current model. Following the Motion Picture Association’s formal cease-and-desist action against Seedance 2.0 in February 2026, ByteDance added content filters that block generation of recognizable real faces and copyrighted characters. Those filters are expected to carry forward into Seedance 2.5. All content generated by Seedance includes an invisible watermark that identifies it as AI-generated. This watermark persists even if the content is shared or altered off-platform. CapCut also embeds C2PA Content Credentials, an industry-standard technology that helps other platforms and individuals verify whether content was created with AI.
Creators using Seedance 2.5, on any platform, should understand that the underlying copyright litigation involving Seedance 2.0 has not concluded. US availability timing for the newest model remains uncertain, and previous copyright disputes around Seedance 2.0 may affect how quickly it becomes available in some regions. Using any AI video model to recreate recognizable intellectual property, fictional characters, or real individuals’ likenesses carries legal risk that the platform’s own filters do not fully eliminate. If you are using AI-generated video in a commercial context, consult with a legal professional before publishing content that draws on protected material.
What should US creators do right now?
US-based creators who want to use Seedance 2.5 have 3 practical paths available today. First, monitor the CapCut model selector directly. The model will appear as a selectable option once it reaches your account; it will not require a separate download. Second, access the model through a third-party API provider that has already integrated the August 7 release, if you are comfortable with a developer workflow. Third, continue working with Seedance 2.0, which remains available through CapCut’s Dreamina interface in the US and carries the same native audio-video joint generation architecture that made it competitive on third-party benchmarks.
Shortly after its release in February 2026, Seedance 2.0 took first place in the video generation benchmark of Artificial Analysis, an AI model performance analysis firm. That baseline matters: Seedance 2.5 builds on the same architecture and extends it, so the predecessor is a reasonable stand-in while the newer model’s regional rollout continues. Creators who depend on US legal certainty for commercial work should also track the unresolved MPA dispute before committing Seedance-generated content to a production pipeline. The copyright situation is an active legal matter, not a resolved one, and the outcome could affect how the model operates in the US market.
For broader context on AI tool risks and data practices, see our coverage of OpenAI pausing Astra development over a critical cyberattack risk and Google bringing Gemini AI to K-12 students, which examines how AI platform rollout decisions affect end users. For a broader look at the competitive AI landscape, Alibaba’s Qwen3.8-Max launch illustrates how Chinese AI labs are competing across multiple model categories simultaneously. The ongoing race in AI video sits inside a wider pattern of rapid model releases: ChatGPT’s approach toward one billion weekly users shows the scale at which AI tools are now operating.
FAQ
Is Seedance 2.5 available in the US on CapCut?
Seedance 2.5 is not confirmed as available in the US on CapCut as of August 10, 2026. CapCut’s official rollout page lists Europe, Asia, the Middle East, and South America as the current expansion regions for subscriber accounts. US availability has not been announced, and unresolved copyright disputes from the Seedance 2.0 rollout continue to affect the timeline.
What resolution does Seedance 2.5 actually output?
The resolution situation differs between ByteDance’s announced targets and what shipped: Seedance 2.5 currently caps at 720p through the public API, and the original 4K announcement did not make it into the shipped model. CapCut’s own product pages describe 4K output through the consumer interface, so the resolution available to you will depend on whether you are using the consumer product or the developer API. Verify the resolution option before committing to a production workflow.
How much does Seedance 2.5 cost to use?
ByteDance has not published official per-second pricing for Seedance 2.5 through its own platforms as of August 10, 2026. The model is billed per second of generated video, but a specific confirmed rate has not been disclosed. Through CapCut’s consumer interface, access requires a paid CapCut Pro subscription, which has been priced at approximately $7.99 per month or $74.99 per year for the Pro tier (pricing varies by region and plan; confirm current pricing on CapCut’s official pricing page before subscribing).
Can Seedance 2.5 generate videos of real people or copyrighted characters?
ByteDance added content filters following the Motion Picture Association’s formal cease-and-desist action over Seedance 2.0, which block generation of recognizable real faces and copyrighted characters. Those filters reduce but do not eliminate the risk of generating protected content. Using AI video tools to reproduce real individuals’ likenesses or recognizable fictional characters for commercial purposes carries legal exposure that platform filters do not resolve. Consult a legal professional before publishing such content commercially.
How does the beta Long Video Mode work in Seedance 2.5?
The beta Long Video Mode extends output to 180 seconds through multi-round generation, maintaining character, scene, lighting, and audio consistency. This feature is explicitly labeled as a beta, which means it is subject to change and may produce less consistent results than the standard 30-second single-pass mode. Treat extended outputs as a workflow tool for longer narratives rather than a production-ready feature until ByteDance removes the beta designation.