Seedance 2.5 is an audio-video joint generation model for AI video creation, mainly used to generate video content with longer duration and stronger continuity.
Purpose Seedance 2.5 is used to create cinematic AI videos, supports native 30-second generation, up to 1080p output, and can be controlled by combining image, video, and audio references. It is suitable for scenarios such as advertising, social content, e-commerce, and storytelling.
Main capabilities
- Longer complete scenes: Can create a complete 30-second scene in a single clip, with a smoother narrative flow and fewer splice transitions.
- Stronger reference control: Can combine image, video, audio, and other references to guide character consistency, style, motion, and brand direction; supports up to 50 multimodal inputs.
- More practical editing: Can refine local details while preserving the overall scene composition, motion, lighting, and identity; can modify only the parts of the video that need adjustment instead of regenerating the entire scene.
- Better production planning: Supports the use of structured prompts and 3D white model guidance to shape composition, blocking, camera direction, and the final image.
- R2V reference-guided complex scenes: Can use green-screen footage or white model references to more precisely control character movement, spatial position, and interaction; the model follows structured motion paths rather than relying only on text prompts.
- Long video and continuity: Standard mode can generate videos up to 30 seconds, and the beta long-video mode can extend to 180 seconds; maintains character recognition, lighting, motion, and scene consistency across longer sequences.
- Local region editing: Can replace incorrect objects, modify visual details, or improve characters while preserving lighting, composition, motion, audio, and timeline continuity.
- Multilingual creation and 4K output: Supports major languages such as Chinese, English, Spanish, Indonesian, Malay, Thai, Arabic, Portuguese, Vietnamese, Japanese, and Korean; improved stability can remove random subtitles and unwanted background music, providing a clean 4K video base for subsequent text and music editing.
- Native 30-second + audio generation, conversational in-video editing, and stronger character continuity.
How to use
- Add references: Where supported by the selected model, add reference materials to guide the visual style.
- Prompt: Use a clear instruction to describe the subject, action, camera, and mood.
- Settings: Select an available model, and adjust output settings as needed for the shot, such as duration, aspect ratio, and resolution.
- Generate: Review credit consumption, then after generation play back and refine the result.
In the generator, you can use @ input to reference uploaded image, video, or audio filesPrompts are moderated before generation; requests involving sexual content, violence, hate, illegal activity, or infringing content may be blocked.
Use cases Seedance 2.5 is suited for complete narrative clips of 30 seconds or more, such as product launch films, cinematic story clips, social media content, e-commerce videos, and advertisements. Its reference control and multimodal input capabilities are suited for projects that require character consistency, unified style, or a clear brand direction; structured prompts and 3D white-model guidance are suited for production workflows that require planning composition, blocking, and camera scheduling.
Limitations
- Prompts are moderated before generation, and some content may be blocked.
- Standard mode is up to 30 seconds; the 180-second long-video mode is in beta.
- Whether some capabilities (such as visual reference) are available depends on the selected model.
- Details not explicitly provided in the materials, such as fees, pricing, platform support, and specific usage steps, are not yet available.
