Seedance 2.5
Coming soonByteDance's next-generation video model with 30-second single-pass generation
Announced at the Volcano Engine FORCE conference on June 23, 2026. Seedance 2.5 isn't available yet. Talk to our sales team about enterprise access, or follow us on X to see when it lands.
All workflows supported natively.
One integration for generation and editing, with no separate endpoints.
Why Seedance 2.5 stands out
30-second single-pass generation
Generates a continuous 30-second clip directly, without stitching shorter segments together. No seams and no post-processing required to hide joins.
Up to 50 multimodal reference inputs
A roughly 4× jump from Seedance 2.0's ~12 inputs. Use image references alongside audio and video inputs in a single generation call to compose scenes with precise control over characters and visual context.
Regional re-draw editing
Edit a specific region or subject within a frame while keeping the rest of the scene consistent. Change one thing without regenerating the entire clip.
20% improved prompt adherence
Instructions land more precisely. Seedance 2.5 follows complex, multi-element prompts with noticeably fewer deviations, reducing the iteration required to get a usable result.
Built-in content filters
Filters blocking generation of recognizable real faces and copyrighted characters carry forward from Seedance 2.0. They apply at the model level across all outputs.
Enterprise-grade pipeline fit
Announced as a global enterprise beta with a public launch targeted for early July 2026. Built for teams running high-volume generation in production.
Built for real workflows
Seedance 2.5 is ByteDance's most capable video model to date. Longer native clips, a much wider reference surface, and in-place editing mean fewer generation cycles and more usable first-pass results, whether you're building a creative tool or running a production pipeline.
Long-form content production
Generate a complete 30-second clip from a single prompt or reference set, with no stitching and no post-processing required to hide joins.
Multi-reference scene composition
Combine up to 50 image references alongside audio and video inputs in one call to compose complex scenes with precise control over characters and visual context.
Regional scene editing
Change a single element in a frame, such as a character or background region, without touching the rest of the clip. Precision editing at the clip level.
Iterative creative direction
Refine outputs across multiple passes using regional re-draw. Adjust one area, keep everything else, and converge on the exact frame composition you need.
Audio-visual sync
Pass audio references alongside visual ones to generate clips where the soundtrack and on-screen action are composed together from the start.
High-volume pipeline generation
Enterprise-grade throughput for teams generating at scale. The same Runware API and infrastructure apply, with a more capable model behind them.
How it compares
Where Seedance 2.5 sits next to other leading video models. Its edge is the combination of 30-second native generation, a 50-input reference surface paired with in-place regional editing, a combination no other model currently offers together.
| Model | Seedance 2.5 | Seedance 2.0 | Gemini Omni | Veo 3.1 | Kling VIDEO O3 |
|---|---|---|---|---|---|
| Max duration | 30s | 15s | 10s | 8s | 15s |
| Input types | Text, image, audio, video | Text, image, audio, video | Text, image, audio, video | Text, image | Text, image, audio |
| Reference inputs | Up to 50 | ~12 | Multi-turn | Limited | Multi-turn |
| Native audio | Yes | Yes | Yes | Yes | Yes |
| Max resolution | 1080p | 1080p | 1080p | 4K | 4K |
| Regional editing | Yes | No | Yes (conversational) | No | Limited |
| Content filters | Real faces & copyrighted characters blocked | Real faces & copyrighted characters blocked | Standard Google SafeSearch | Standard Google SafeSearch | Standard |
| Best fit | Long-form clips, heavy multi-reference composition, regional editing | Audio-visual sync, multi-reference composition | Iterative conversational editing grounded in real-world knowledge | Cinematic generation, high-resolution output | Long-form, character-driven content in 4K |
| Capabilities | 30s single-pass, 50 multimodal refs, regional re-draw, 20% better prompt adherence | Multi-reference composition, audio-visual sync; no regional editing, 15s max | Conversational multi-turn editing, real-world knowledge; shorter clips, fewer refs | Real-world knowledge, 4K output; limited multi-turn editing and reference inputs | Long-form 4K output; limited real-world knowledge and regional editing |
Specs and capabilities are indicative and evolve as each model updates. Seedance 2.5 figures are ByteDance's stated specs from the Volcano Engine FORCE keynote (June 23, 2026) and have not been independently verified.
Frequently asked questions
What is Seedance 2.5?
ByteDance's next-generation video model, announced at the Volcano Engine FORCE conference on June 23, 2026. It generates continuous 30-second clips in a single pass, accepts up to 50 multimodal reference inputs, supports regional re-draw editing, and delivers 20% better prompt adherence compared to Seedance 2.0.
When will Seedance 2.5 be available?
ByteDance announced Seedance 2.5 as a global enterprise beta, with a public launch targeted for early July 2026. It isn't generally available yet. Request access through the form below and we'll let you know as soon as you can start using it on Runware.
How is Seedance 2.5 different from Seedance 2.0?
Four headline upgrades: native 30-second single-pass generation (vs 15s stitched), up to 50 multimodal reference inputs (vs ~12), regional re-draw editing that lets you change one part of a frame without touching the rest, and a claimed 20% improvement in prompt adherence. Content filters from 2.0 carry forward unchanged.
What are the content filters?
Seedance 2.5 carries forward the content filters ByteDance added in 2.0: generation of recognizable real faces and copyrighted characters is blocked at the model level. These filters apply to all outputs and are not configurable.
What does regional re-draw editing mean in practice?
You can target a specific region or subject within a generated frame and re-draw it while keeping everything else in the clip consistent. For example, swap a background or replace an object without regenerating the full clip from scratch.
How does Seedance 2.5 compare to other video models?
The 30-second native duration and 50-input reference surface, combined with regional editing, are not available together in any other current model. Gemini Omni leads on conversational multi-turn editing; Veo 3.1 on real-world knowledge grounding; Kling on 4K output. On Runware, all models are accessible through the same API, so you can pick the right one per task.
How much does Seedance 2.5 cost?
Seedance 2.5 isn't available on the platform yet, so pricing hasn't been published. Once it launches, rates will appear on the Runware pricing page alongside our other video models. Contact us to be notified when it goes live.
Do I need a separate integration for editing vs generation?
No. Generation and multi-reference composition, along with regional editing, will all run through the same Runware API endpoint.
Talk to us about enterprise access
Have questions about Seedance 2.5? Chat to our team about enterprise usage, including volume discounts and dedicated RPM, and we'll follow up shortly.