Seedance 2 is a multimodal video generation model whose core feature is a "reference-anything" control approach, enabling fine-grained control over style, subject, motion, camera, and audio, and supporting video extension, editing, and built-in audio generation.
Uses Seedance 2 is mainly used for controllable text-to-video, image-to-video, and video-to-video generation. It is suited for multi-shot sequence creation that requires maintaining consistency of characters, clothing, products, and typography, as well as for replicating complex action rhythms and camera language, and for generating dialogue, ambient sound, and sound effects for videos.
Main capabilities
- Multimodal input: supports text, images, video, and audio.
- Reference anything: can reference style, subject, motion, camera, or sound.
- Consistency enhancement: maintains stability of characters, clothing, products, and typography.
- Motion and camera replication: follows complex action rhythms and camera language.
- Video extension and editing: can extend clips or edit existing material based on references.
- Audio generation: can generate dialogue, ambient sound, and sound effects along with the video.
- Output settings: under supported configurations, can output up to 2K resolution, with selectable aspect ratios and formats suited to the platform.
How to use
- Add prompts and multimodal material: combine a text prompt with up to 9 images, 3 videos (total duration ≤15 seconds), and 3 audio files (total duration ≤15 seconds); set a duration of 4–15 seconds.
- Assign @ references: use @ to mark first-frame/last-frame, style, subject, motion, camera, or audio roles. For simple image-to-video, you can choose First/Last-frame mode; for mixing multiple materials, you can choose Omni-Reference mode.
- Generate, extend, and edit: after previewing the material, generate, extend, or edit shots to maintain consistency of characters, motion, and camera across the sequence.
Prompt best practices
- Input limits and duration: images ≤ 9; videos ≤ 3 (total duration ≤15 seconds); audio ≤ 3 (total duration ≤15 seconds); total number of files ≤ 12; duration 4–15 seconds.
- Choose the appropriate mode: use First/Last-frame when there is only a single image plus a prompt; use Omni-Reference and @ tags when mixing multiple materials.
- Clearly state references: in the prompt, explain what each material controls, for example @image1 as the first frame, @video1 for camera/motion, @audio1 for rhythm; check @ names to avoid confusion.
- Structured shots: bySubject → action → camera → style writing; add shot size and movement (push/pull/follow); use Shot 1/2/3 to indicate multi-shot sequencing; include dialogue tone, ambient sound, and sound effects; short sentences help with lip sync; specify sound sources (wind, crowd, footsteps).
- Extension and editing: specify the extension duration and replacement content, e.g. "extend @video1 by 5 seconds"; when editing, specify the target (character/prop/scene).
Use cases The official guide provides multiple prompt examples, covering: slice-of-life realism (e.g. a girl hanging laundry), street chase action, family reunion continuity, luxury bag product showcase, elevator suspense shot, rotating tablet product showcase, sci-fi POV template, particle title opening, comic panel animation, seasonal childhood opening, time-lapse extension, sunflower scooter extension, cat-and-dog talk show, Yu opera singing scene, one-take rooftop chase, one-take spy tracking, singer replacement edit, shark reveal edit, music beat styling, beat-synced scenery, mirror breakdown emotion, kitchen comparison ad, etc. These examples demonstrate Seedance 2's applicability in advertising, narrative, action, product showcase, and music synchronization directions.
Limitations
- Input materials have clear limits: images ≤ 9, videos ≤ 3 (total duration ≤ 15 seconds), audio ≤ 3 (total duration ≤ 15 seconds), total number of files ≤ 12.
- Generation duration is limited to 4–15 seconds.
- 2K output is only available under supported configurations.
- Specific information such as billing, pricing, and platform support is not currently provided in the materials.
