Seedance 2.5 is an audio-video generation model from ByteDance, supporting longer scenes, reference control, and editing.
Capabilities
Longer scenes with audio
Generate audio and video together in clips up to 30 seconds. ByteDance also describes the option to extend a generation twice.
Reference-guided direction
Use reference videos to guide framing, camera movement, and the intended action of a scene.
Audio and visual editing
The model supports editing requests that modify audio and visual elements, including green-screen editing.
Production controls
White-model control, camera movement, and performance blocking provide ways to guide scene structure and staging.
Generation modes
Text to video
Start with a written scene description. Specify the subject, setting, action, camera position, and sound. This mode is useful when the scene does not need to begin from an existing image. A clear prompt gives the generation a visual direction and an ordered sequence of events.
Image to video
Use an image as the starting frame and describe what happens next. An optional ending image can guide the closing composition. The prompt can focus on motion: what the subject does, how the camera moves, and what changes between the beginning and the end.
Reference to video
Combine image, video, and audio references with a prompt. Reference inputs can include up to 30 images, 10 videos, and 10 audio clips. Describe the role of each material so that a character image, a scene reference, and a motion example have distinct purposes.
Use cases
Product scenes
A possible brief is a product on a tabletop, a hand entering the frame, and a closing detail shot. Prepare references for the object and setting, then describe the order of the actions. Review small markings and object geometry before using the result.
Narrative studies
Use a short scene to explore a character entrance, a reaction, or a reveal. Decide what the audience should understand at the end, then work backward to the opening composition. This gives each action a purpose within the available running time.
Previsualization
A practical use is comparing alternative treatments of the same scene before committing to a final sequence. Keep the action constant while changing the camera angle or environment. Evaluate which version communicates the scene most clearly.
FAQ
How is a starting frame different from a reference image?
A starting frame defines the opening image for image-to-video generation. A reference image supplies visual material for the prompt to use. Choose a starting frame when the opening composition matters; use references when several materials need to inform the scene.
How should a longer scene be described?
Organize the brief into a beginning, a main action, and an ending. Keep each stage readable and avoid asking for several unrelated actions at once. A useful review question is whether the last image clearly completes the event established at the start.
What should be checked in the result?
Watch the full clip, then inspect transitions and important details. Check subject identity, object interactions, camera continuity, and the relationship between sound and action. When revising, change one part of the brief at a time so that the effect is easier to judge.