Article
What Is Seedance 2.5?
Seedance 2.5 is ByteDance Seed's newest short-form multimodal AI video model, available natively on HiArt as the default video model. It keeps the multimodal workflow from Seedance 2.0 — text, images, reference videos, and audio in one generation — then pushes duration to 30 seconds and expands omni-reference to up to 50 media files.
If you already know Seedance 2.0's
@ reference system, Seedance 2.5 feels familiar: upload assets, assign each one a role in natural language, and generate a clip with synced dialogue, ambient sound, or music. The jump is capacity — longer storytelling and far more reference inputs in a single render.What changed vs Seedance 2.0
Capability | Seedance 2.5 | Seedance 2.0 |
|---|---|---|
Max duration | 30 seconds | 15 seconds |
Omni-reference | 30 images, 10 videos, 10 audio (50 total) | 9 images, 3 videos, 3 audio (12 total) |
Resolution | 480p / 720p | Up to 1080p |
Modes | Text-to-video, first/last frame, omni-reference | Same three modes |
Audio | Native synced audio + lip sync | Native synced audio + lip sync |
The tradeoff is clear: 2.5 maximizes length and reference capacity; 2.0 maximizes resolution. For a side-by-side decision, see Seedance 2.5 vs 2.0 or the Seedance 2.5 model page.
What makes it different
- Up to 30-second clips — twice the max duration of Seedance 2.0, enough for multi-beat ads and short narratives
- Expanded omni-reference — combine up to 30 images, 10 videos, and 10 audio files in one generation
- Built-in synced audio — dialogue, ambient sound, and music with lip sync support
- Director-level control — replicate camera moves, choreography, VFX, and beat-matched motion from reference media
- Canvas-native on HiArt — start a still in any image model, then animate it in Seedance 2.5 on the same canvas
Three generation modes
Seedance 2.5 supports the same three workflows as 2.0. Pick the one that matches your starting assets.
Text-to-video
Describe the scene, action, and camera in plain language. No uploads required — ideal for quick concept tests and longer mood boards up to 30 seconds.
First / last frame
Upload one image as the opening frame, or two images as the start and end points. The model generates the motion between them — useful for product reveals, logo animations, and controlled transitions across a longer timeline.
Omni-reference
The most powerful mode. Upload reference images for character look or scene style, reference videos for camera movement or choreography, and reference audio for rhythm or mood. Use
@Image1, @Video1, @Audio1 syntax in your prompt to tell the model what each file is for — now with room for many more assets than Seedance 2.0.Input limits (omni-reference)
Input | Limit | Format | Max size |
|---|---|---|---|
Images | ≤ 30 | jpeg, png, webp, bmp, tiff, gif | 30 MB each |
Videos | ≤ 10 | mp4, mov | 50 MB each |
Audio | ≤ 10 | mp3, wav | 15 MB each |
Total | ≤ 50 files | — | — |
Output specs
- Duration: 4–30 seconds (selectable)
- Resolution: 480p or 720p
- Audio: native co-generation with lip sync, sound effects, and background music
- Aspect ratio: 16:9, 9:16, 4:3, 3:4, 1:1, 21:9, or adaptive
When to choose Seedance 2.5
- Longer short-form ads — 20–30 second product stories with synced voiceover
- Music videos & lip-synced content — many audio and visual references in one pass
- Character-driven scenes — keep identity consistent across a large image reference set
- Multi-shot narratives — more beats than a 15-second Seedance 2.0 clip can hold
- Reference-rich prototyping — combine look, motion, and sound references without splitting into multiple renders
Prefer maximum resolution finals under 15 seconds? Use Seedance 2.0 instead. Prefer fast motion previews before a long render? Iterate on Seedance 2.0 Fast, then move to 2.5 for the longer cut.
Things to know before you start
- Uploaded images and videos cannot contain realistic human faces — the platform blocks such content for compliance.
- Reference audio cannot be used alone; pair it with at least one reference image or video.
- First/last frame mode and omni-reference mode cannot be mixed in a single generation.
- Output tops out at 720p — plan a separate 1080p pass on Seedance 2.0 if you need max resolution.
- Results depend heavily on prompt quality — especially how clearly you assign roles to each
@reference when you upload many files.
Ready to generate? Open a video node on HiArt canvas, select Seedance 2.5, and start with a simple text-to-video prompt. When you are ready to go deeper, read our Seedance 2.5 prompt writing guide for the
@ reference system, 30-second timing beats, and copy-paste templates.