Seedance 2.5

Overview

Seedance 2.5 is ByteDance Seed's July 31, 2026 update to the Seedance audio-video generation model family. The official Seedance 2.5 page positions it as a next-generation audio-video joint generation model for longer storytelling, tighter reference control, and more reliable editing.

Compared with Seedance 2.0, the clearest shifts are longer duration and more precise multimodal reference and editing control. Seedance 2.5 supports videos up to 30 seconds in a single generation, with two extension rounds available for richer narrative sequences. It also focuses on reference-video understanding, white-model control, green-screen editing, professional camera movement, and performance blocking.

What's New

Up to 30-Second Storytelling

Seedance 2.5 extends Seedance beyond the 15-second multi-shot format associated with 2.0. ByteDance says the new model can create videos up to 30 seconds in one generation, and users can extend the result twice for more complete scenes.

This matters for creators building short ads, product stories, music-driven clips, or previz sequences where a single idea needs a beginning, middle, and payoff. The longer format reduces the need to stitch several independent clips together, though creators should still review continuity, pacing, and subject consistency before using outputs in production.

Stronger Reference Video Understanding

The official page says Seedance 2.5 understands reference videos more precisely, including intention, framing, and cinematic language. That is a broader promise than simple motion transfer: the model is positioned to interpret why a reference works, then apply that visual logic to a new generation.

For creative teams, this should make reference-led workflows more practical. A director can provide a source clip for camera feel, blocking, edit rhythm, or shot language, then use text instructions to steer the final scene rather than describing every motion detail from scratch.

More Reliable Audio-Visual Editing

Seedance 2.5 also expands editing reliability. ByteDance says the model responds to a wider range of audio and visual editing requests, which makes the release more useful after the first generation instead of only at prompt time.

This is important because AI video workflows often fail at revision: a usable clip may need one object changed, a background cleaned up, or an audio-visual beat adjusted without rebuilding the whole output. Seedance 2.5 is positioned as a more editable model for those iterative production loops.

White-Model Control and Green-Screen Editing

The official page calls out white-model control and green-screen editing as advanced production capabilities. These features point Seedance 2.5 toward previsualization, staging, compositing, and professional creative workflows rather than only consumer prompt-to-video use.

White-model control can help creators reason about staging and motion before final styling, while green-screen editing can support downstream compositing. The public page does not disclose detailed interface constraints, so teams should verify exact platform behavior before planning a pipeline around these controls.

Availability & Access

ByteDance officially launched Seedance 2.5 on July 31, 2026 and began rolling it out through Jimeng AI, Doubao Pro, and other platforms. Availability may still vary by account and region, and BytePlus ModelArk API access should be verified before developer integration.

For now, treat access as ByteDance-platform dependent. Seedance 2.0 was surfaced through ByteDance creative products and enterprise channels such as Jimeng AI, Doubao, and Volcano Engine; Seedance 2.5 availability should be verified in the product surface or account region you plan to use.

Pricing & Plans

The public Seedance 2.5 model page does not show pricing, but Volcano Engine's official rate table lists metered Seedance 2.5 pricing for 480p and 720p output. Existing Seedance access has historically depended on the ByteDance product surface and region, so production cost should still be confirmed inside the official platform before planning a campaign budget.

Volcano Engine lists Seedance 2.5 at RMB 70 per million tokens without video input and RMB 42 per million tokens with video input. BytePlus ModelArk API access may still be rolling out, so the safest planning method is to measure cost per usable output rather than cost per generation. Track retry rate, seconds kept, edit rounds avoided, queue time, and post-production work, because a longer or more editable generation may still be cheaper if it replaces several failed shorter clips.

Best For

• Brand and agency teams building 20-30 second ad concepts that need one coherent narrative beat rather than stitched short clips.
• Filmmakers and previz artists testing camera movement, staging, and performance blocking before manual production.
• Social video teams that already use Seedance 2.0 and want longer, more editable generations for campaign assets.
• Product marketers creating reference-driven scenes where framing, motion language, and visual style must follow existing footage.
• Creative operations teams evaluating whether fewer longer generations can replace multi-clip assembly and revision work.

FAQ