ByteDance Seedance 2.5: The 30-Second 4K AI Video Model That Accepts 50 Reference Inputs

ByteDance previewed Seedance 2.5 at its Volcano Engine FORCE conference in Beijing on June 23, 2026, as the successor to Seedance 2.0 . The model is currently in global enterprise beta, with a public launch targeted for early July 2026 . Pricing and public API access have not yet been announced .

What Makes Seedance 2.5 Different?

Seedance 2.5 represents a significant leap in AI video generation capabilities. Its headline features include:

• 30-second native single-shot video: generates a seamless clip up to 30 seconds in one pass, not stitched from shorter segments .
• 4K resolution: outputs at native 4K, up from the lower resolutions of Seedance 2.0 .
• 50 multimodal reference inputs: accepts up to 50 images, videos, or audio files as references (up from 12 in Seedance 2.0), giving users much more control over characters, style, and scene composition .
• Native synchronized audio: produces video and audio together in one generation, eliminating a separate dubbing or audio step .
• Localized editing: allows editing specific characters or details within a generated video without regenerating the entire clip .
• 180-second extended mode: Dreamina's official page lists a beta long-video mode that extends output to 180 seconds .
• Improved lip sync and multi-shot continuity: better character consistency across shots and more natural lip movements .

Release Timeline and Availability

Milestone	Date
Announcement & enterprise beta	June 23, 2026
Public launch target	Early July 2026
China launch target	July 2026 (China-first rollout expected)

Seedance 2.5 will be available through ByteDance's consumer ecosystem:

• Dreamina (the primary AI creative app) — the model will be accessible inside Dreamina's "AI Video" tool with a "Seedance 2.5" mode selection .
• CapCut — the public launch is expected on CapCut as well, and users can export from Dreamina directly to CapCut for editing .
• Doubao — ByteDance's other AI assistant will also get the model .
• Volcano Engine API — enterprise and developer access via ByteDance's cloud platform .

As of late June 2026, Seedance 2.5 is not yet generally available to the public — only enterprise beta testers have access . Users can still use Seedance 2.0 on Dreamina in the meantime .

How Seedance 2.5 Stacks Up Against Competitors

Seedance 2.5 enters a rapidly shifting AI video landscape. Independent comparisons rank it as the current leader on native duration, reference control volume, and brand/style consistency .

vs. Kling 3.0 (Kuaishou): Kling 3.0 leads on raw resolution and has a free tier, but Seedance 2.5 beats it on native duration (30 seconds vs. shorter clips), number of reference inputs, and brand consistency .

vs. Veo 3.1 (Google DeepMind): Veo 3.1 has strong developer API access and integrated audio, but Seedance 2.5 offers significantly longer native clips — 30 seconds versus roughly 8 seconds — and far more multimodal references .

vs. Sora 2 (OpenAI): Sora 2 still excels at physics realism and stylization, but OpenAI has deprecated Sora and is shutting it down, making it a dead end for new production work .

vs. Runway Gen-4: Runway leads on temporal consistency and professional-grade motion control, but Seedance 2.5's native 30-second generation and 50-reference input capacity outstrip Runway's current capabilities in length and controllability .

Analyst take: Seedance 2.0 had already displaced Kling v3 on the Artificial Analysis leaderboard within 90 days of launch, and Seedance 2.5 extends that lead by pushing the evaluation axis from single-scene clip quality toward multi-shot narrative coherence .