Models and parameters
Choose a model for the task, then check duration, references, aspect ratio and resolution. Limits from one version do not automatically apply to another.
Seedance 2.5 reference
| Parameter | Documented capability or limit |
|---|---|
| Generated duration | 4–30 seconds |
| Image references | Up to 30 images, each up to 4K |
| Video references | Up to 10, totaling no more than 30 seconds |
| Audio references | Up to 10, totaling no more than 30 seconds |
| Combined references | Up to 50 images, videos and audio files in total |
| Source video for precise editing | Prefer 20 seconds or less for stability; this is a recommendation, not a separate hard cap |
| Seedance 2.5 shot prompt | Up to 15,000 characters |
| Storyboard planning cap | 15, 20 or 30 seconds |
| Features | Multimodal generation, precise video editing and video extension |
The planning cap controls shot splitting; it does not require every shot to have that duration. Check that actions and dialogue can fit naturally within the generation duration you select.
Maximum versus recommended input
| Task | Recommended input | When adding more |
|---|---|---|
| Define subjects with images | 1–8 subjects | Consistency may decline with 9–12 subjects; assign a clear purpose to each |
| Define subjects with video or audio | 1–5 subjects, ideally 5–10 seconds per clip | 6–10 subjects may interfere with one another; total-duration limits still apply |
| Provide precise-edit images | 1–5 images | 6–8 images are not necessarily more stable; identify the master and target references |
Subject count is distinct from file count. When using more than five subjects that also need multiple views, separate the views into individual images rather than a multi-view collage. Recommended clip lengths do not replace the totals: all video references together and all audio references together must each remain within 30 seconds. Assign identity, position, movement and voice to explicit references.
Aspect ratio and resolution
Project ratios include 16:9, 9:16, 4:3, 3:4 and 21:9. Resolution depends on the model and feature; the studio may offer options such as 480p, 720p, 1080p or 4K. Upscaled output resolution is not the same as native generation resolution. Inspect the actual output before delivery.
The 18 August 2026 update added a 1080p option for Seedance 2.5. Your model list may also include Seedance 2.0, Mini or Wan variants. Use the parameters available to your account rather than applying 2.5 limits to every model.
Languages
The Seedance 2.5 guide lists Chinese, English, Spanish, Indonesian, Malay, Thai, Arabic, Portuguese, Vietnamese, Japanese and Korean. Specify the regional accent, speaker and emotion in the prompt, then audition a test shot.
Remaster language choices are model-dependent; see Start a remaster project. Interface language, generated dialogue language and remaster target language are separate settings.
Before switching models
Shots planned for 20 or 30 seconds cannot be directly generated with Seedance 2.0 using the same assumptions. Review shot length, reference limits and feature support first.
Match the model to the task
| Model or family | Documented role |
|---|---|
| Seedance 2.5 | Multimodal video generation, precise editing and extension in either direction; 4–30 seconds per generation |
| Seedance 2.0 / 2.0 Fast / 2.0 mini | Video-generation variants; 2.5 duration and reference limits do not automatically apply |
| Wan 3.0 | A video model option introduced in the September 2026 updates |
| Seedream 5.0 Pro | Image generation, selected separately from video generation |
| Seed Audio 1.0 | Intelligent voice design from character descriptions, with audition audio |
The names identify tasks, not a promise that every entry offers every model. Check the selector for the current task. The recorded image-prompt limit is 1,200 characters; do not confuse it with the 15,000-character Seedance 2.5 video-prompt limit.