What is Seedance?

Seedance is a closed video generation model family developed by ByteDance Seed, the research arm of ByteDance, with Seedance 2.0 launched on Feb 12, 2026 and Seedance 2.5 on Jul 31, 2026. It generates clips from text, images, audio and video references and is served only through ByteDance products and the Volcengine and BytePlus cloud APIs.

Volcengine's model list documents Seedance endpoints for enterprise customers; consumer access runs through Dreamina and CapCut.

What can Seedance 2.0 and 2.5 do?

Seedance 2.0 introduced audio-visual joint generation and multimodal referencing: a single request can combine up to nine images, three video clips and three audio clips as references, and the model produces synchronized sound with the picture. Text, image and video inputs steer motion, style and identity.

Seedance 2.5, announced Jul 31, 2026, extends this toward what ByteDance calls one-take creation, with 30-second long-form clips and more flexible referencing. Both versions are closed weights; developers reach them through Volcengine or BytePlus endpoints rather than downloads.

Does Seedance allow NSFW output?

Seedance is served exclusively behind ByteDance platforms, and every route applies the platform's moderation. The launch blog for Seedance 2.0 adds a specific rule for likeness: uploading a real person's portrait as a subject reference requires identity verification or prior legal authorization.

Sexual content policies are governed by the terms of the app or API through which Seedance is accessed, and the model pages we read on Sep 3, 2026 do not publish a separate NSFW clause. In practice the combination of platform moderation and the likeness check makes Seedance unsuitable for the workflows that unfiltered open models enable.

“If you wish to use real human portraits as subject references for video generation, identity verification or prior legal authorization is required.”

— ByteDance Seed, Seedance 2.0 launch blog, accessed Sep 3, 2026

How is Seedance used in tools we review?

None of the tools in our rankings publicly names Seedance as a backbone, and given the identity check at the platform layer we would not expect a third-party service to route intimate content through it. We list the model here because readers compare closed, moderated video models against the open ones that unfiltered tools rely on.

The comparison matters for how we rate: a service that depends on a moderated API inherits its provider's consent controls, while one built on open weights has to prove its own. The video tools hub explains which reviewed services offer clip output at all.

Where is Seedance available?

Consumers meet Seedance inside Dreamina, CapCut and the Doubao assistant; enterprises call it through Volcengine in China and BytePlus internationally. The Volcengine model documentation lists the Seedance endpoints, parameters and pricing for API customers.

There is no open checkpoint, no Hugging Face repository and no local inference path, so reference-to-video with Seedance always leaves a server-side record with ByteDance. See image-to-video for the underlying input mode.

Sources

1. ByteDance Seed — Seedance 2.0 official launch — launch Feb 12, 2026; identity-verification rule, accessed Sep 3, 2026.
2. ByteDance Seed — Seedance 2.0 model page — multimodal referencing limits, accessed Sep 3, 2026.
3. ByteDance Seed — Introducing Seedance 2.5 — Jul 31, 2026; 30-second clips, accessed Sep 3, 2026.
4. ByteDance Seed — model list — Seedance releases by date, accessed Sep 3, 2026.
5. Volcengine — Seedance model documentation — API endpoints and parameters, accessed Sep 3, 2026.

Reference entry by the Clothoff AI Editorial Team; not legal advice. The access date marks the last verification.