Seedance 2.5 is ByteDance’s next-generation AI video model, previewed on June 23, 2026 at the company’s Volcano Engine FORCE conference and set for public launch in early July. Its headline claim is one that the whole AI video field has been circling: a single, continuous 30-second clip generated directly, in native 4K, with synchronized audio produced in the same pass rather than bolted on afterward. If that holds up in general release, it moves the practical ceiling for AI video from a few seconds of silent footage to a half-minute of finished, sound-included shots.
The short version: Seedance 2.5 pushes on the three things that have made AI video hard to use in real production, which are length, control, and audio. It generates longer clips without stitching shorter ones together, accepts up to 50 reference inputs so you can hold characters and style consistent, and generates matching audio natively. It undercuts the Western leaders on price, at least based on its predecessor. And it arrives at a moment when the competitive field is reshuffling, with one major rival already shut down. This piece covers what Seedance 2.5 is, the features that matter, how it stacks up against Google’s Veo and Kuaishou’s Kling, and the availability and pricing picture, which is still partly unconfirmed.
What Seedance 2.5 is
Seedance 2.5 is a text-to-video and image-to-video model from ByteDance, the company behind TikTok, built on the Seedance 2.0 architecture. It is aimed squarely at production use rather than novelty clips, with an emphasis on longer durations, tighter creative control, and native audio. ByteDance previewed it at its Volcano Engine developer conference in late June, positioned it against the leading Western video models, and put it into enterprise beta with a public rollout targeted for early July 2026.
It ships with three generation modes: text-to-video, image-to-video, and a motion-reference mode that lets you drive the motion of a generated clip from a reference. That last mode, combined with the large number of reference inputs it accepts, is a signal about who the model is for: people trying to produce specific, controlled shots, not just prompt-and-pray one-offs.
What’s new: the 30-second barrier and native audio
Two features define this release. The first is duration. Most AI video models generate short clips, typically five to ten seconds, and getting anything longer has meant generating multiple clips and stitching them together, which introduces visible seams and consistency drift. Seedance 2.5 generates a single continuous 30-second clip natively, without stitching. For anyone who has tried to assemble a coherent scene out of AI-generated fragments, removing the stitching step is a meaningful jump in usability.
The second is audio. Seedance 2.5 generates synchronized audio in the same pass as the video, within the same latent space, rather than requiring a separate audio-generation or manual sound-design step. Native, in-sync audio has been one of the clearest dividing lines between AI video that looks like a tech demo and AI video you could actually drop into a timeline. Generating both together, rather than matching sound to picture after the fact, is the harder and more useful approach.
The other upgrades
Beyond length and audio, Seedance 2.5 adds several features aimed at control and quality:
- Up to 50 multimodal references. You can feed the model as many as 50 reference inputs in a single generation, which is how you keep a character, product, or visual style consistent across a shot. Reference-driven consistency is one of the main things separating usable AI video from unpredictable output.
- Native 4K, 10-bit color. The model renders at 4K with 10-bit color depth, which matters for any professional pipeline where the footage has to hold up on a real display or through grading.
- 3D white-model pre-visualization. A pre-visualization step using 3D white models gives creators a way to block out a shot before committing to a full render, closer to how real production planning works.
- Roughly 20 percent better prompt adherence. ByteDance reports about a 20 percent improvement in how closely the output follows the prompt compared with the prior generation, along with more controllable editing tools.
Taken together, these are production features, not demo features. The pitch is that Seedance 2.5 is meant to slot into a real creative workflow.
How it compares
Seedance 2.5 enters a field that has consolidated around a few serious players, and the competitive picture shifted notably in 2026 when OpenAI’s Sora 2 was shut down in April, removing one of the most recognizable names from active competition. That leaves Google’s Veo and Kuaishou’s Kling as the main Western-market reference points, alongside ByteDance.
The tradeoffs break down roughly like this. Seedance 2.5’s strengths are duration, the number of references it accepts, and brand or character consistency, which is where the 50-reference support pays off. Google’s Veo 3.1 leads on official developer access and has strong native audio of its own. Kling 3.0 leads on raw resolution and offers a free tier, which lowers the barrier to trying it. On price, the Western leaders are considerably more expensive per minute than ByteDance’s line has been, though exact Seedance 2.5 pricing is not yet public.
No single model wins outright. The honest read is that Seedance 2.5 is competing on length, control, and cost, while Veo competes on ecosystem and developer access and Kling competes on resolution and accessibility. Which one fits depends on whether you most need long controlled shots, tight platform integration, or the cheapest path to high-resolution clips.
Pricing and availability
This is where to be careful, because the most important commercial detail is still unconfirmed. As of the late-June preview, ByteDance had not published Seedance 2.5 pricing, and has said it will come with the early-July launch. For context, independent analysis put the previous-generation Seedance 2.0 at roughly 9 dollars per minute of 1080p video, against roughly 24 dollars per minute for Google’s Veo 3.1 and around 20 dollars per minute for Kling 3.0 Pro. If Seedance 2.5 lands anywhere near its predecessor’s pricing, it will be substantially cheaper than the Western leaders, but treat that as an expectation rather than a confirmed figure until ByteDance publishes the numbers.
On availability, Seedance 2.5 is in enterprise beta now, with the public launch targeted for early July 2026. If you are evaluating it, the practical move is to wait for the general release, confirm the actual pricing and any regional availability limits, and test it on your own material before committing, since demo reels and real production footage are very different tests.
What it means for creators and builders
For creative teams and developers building on AI video, Seedance 2.5 is worth watching closely because it targets the exact features that have kept AI video out of serious pipelines: clip length, controllable consistency, and native audio. A model that reliably produces a coherent 30-second shot with matching sound, held consistent by references, is closer to a production tool than most of what has come before.
The caveats are the usual ones for a just-previewed model. General-release quality can differ from a curated preview, the pricing is not yet set, and consistency and physics on real prompts are the things to test rather than assume. The reasonable posture is interested and ready to evaluate: this looks like a genuine step forward on the metrics that matter for real work, and the early-July launch will show whether the general release lives up to the preview.
Frequently Asked Questions
What is Seedance 2.5?
Seedance 2.5 is ByteDance’s next-generation AI video model, previewed on June 23, 2026 and set for public launch in early July. It generates a single continuous 30-second clip in native 4K with synchronized audio in the same pass, accepts up to 50 reference inputs, and offers text-to-video, image-to-video, and motion-reference modes. It is built on the Seedance 2.0 architecture.
What is new compared with previous AI video models?
Two things stand out: it generates a continuous 30-second clip natively without stitching shorter clips together, and it generates synchronized audio in the same pass as the video rather than adding sound afterward. It also supports up to 50 references for consistency, renders 4K with 10-bit color, adds a 3D white-model pre-visualization step, and reports roughly 20 percent better prompt adherence.
When is Seedance 2.5 available?
It was previewed June 23, 2026 and is in enterprise beta, with a public launch targeted for early July 2026. If you want to use it in production, wait for the general release and confirm current availability and any regional limits.
How much does Seedance 2.5 cost?
ByteDance has not published Seedance 2.5 pricing as of the late-June preview; it is expected with the early-July launch. For context, the previous-generation Seedance 2.0 was estimated at roughly 9 dollars per minute of 1080p, versus about 24 dollars per minute for Veo 3.1 and around 20 dollars per minute for Kling 3.0 Pro. Treat any Seedance 2.5 figure as unconfirmed until ByteDance publishes it.
How does it compare with Google Veo and Kling?
Seedance 2.5 leads on clip duration, the number of references it accepts, and brand or character consistency. Google’s Veo 3.1 leads on official developer access and native audio. Kling 3.0 leads on raw resolution and has a free tier. On price, ByteDance’s line has been considerably cheaper per minute, though Seedance 2.5’s exact pricing is not yet public. The best choice depends on whether you prioritize length and control, ecosystem, or cheap high resolution.
What happened to Sora?
OpenAI’s Sora 2 was shut down in April 2026, removing one of the most recognizable AI video models from active competition. That leaves Google’s Veo, Kuaishou’s Kling, and ByteDance’s Seedance line as the main reference points in the current market.
Why does native audio matter?
Generating synchronized audio in the same pass as the video, rather than matching sound to picture afterward, is both harder and more useful. In-sync native audio is one of the clearest lines between AI video that looks like a tech demo and AI video you could actually drop into an editing timeline, because it removes a manual sound-design step and keeps sound and picture coherent.
Who is Seedance 2.5 for?
It is aimed at production use: creative teams and developers who need longer, controllable, consistent shots with audio, rather than short novelty clips. The large reference support, motion-reference mode, and 3D pre-visualization all point at people trying to produce specific, planned shots. As with any just-previewed model, evaluate it on your own material after the general release before committing.
