Four upgrades carry the release
Longer takes that hold a narrative arc, far more reference material per generation, editing that leaves the rest of the frame alone, and native creation in the language you think in.
Longer takes, fuller stories
30-second generation with high-fidelity extension
30-second single-shot output gives you coherent camera expression and a complete emotional arc without manual stitching in post. High-fidelity temporal extension continues a clip into a longer sequence while preserving character, scene and camera consistency — built for ad TVCs, short dramas and brand story films.
More references, richer context
Up to 50 reference assets per generation
Feed in complete sets of characters, scenes and brand assets at once — up to 50 multimodal assets in any combination. High-complexity professional inputs like 3D gray models, product packaging and brand VI are reproduced frame by frame, with structure, materials and camera choreography intact.
Edit with precision
Localized editing and seamless extension
Keep the overall frame, camera and rhythm locked while you add, remove, modify or repaint background, products or characters — down to character age and subtle micro-expressions. Extend forward or backward and bridge between clips. True "one creation, multiple deliveries."
Built for creators everywhere
Native 10+ languages, stronger instruction control
Describe creative intent directly in your native language — no translation layer. Instruction understanding is significantly stronger, so complex camera choreography, emotional turns and multi-layer scene descriptions are followed more precisely.
Multimodal reference control
Blend image, video and audio references in one generation. New in 2.5: an audio-only reference, for high-quality audio-video sync and purely sound-driven visuals.
50 References = 30 Images + 10 Videos + 10 Audio clips
All major formats, ≤ 30 MB each.
mp4 / mov, ≤ 200 MB and 2–30s per clip, ≤ 30s total.
Common formats, ≤ 15 MB and 2–30s per clip, ≤ 30s total.
Uploaded assets are numbered automatically — point to them with @ syntax to lock visual information. Follow each reference immediately with a sentence explaining how it should be used, so bindings stay stable.
For best stability keep distinct subjects within 5, aim for 6–10 reference assets with one clear view each, and keep reference video and audio clips between 5 and 10 seconds.
Where teams are using it
Four commercial contexts from the launch showcase.
Entertainment production
A peak-hour Michelin kitchen scene in one continuous take, character age and expression editing, and 3D scene previsualization.
Educational content
A multilingual museum guide that switches language per gallery, and a stop-motion history of money across five eras.
Advertising and e-commerce
Full 30-second TVCs, a product swapped into an existing viral video, and reusable SKU sequences that keep the product consistent.
Diverse scenarios
Music videos cut to the beat across eight languages, and synthetic training data built by editing existing footage.
Key features
| Dimension | Dreamina-Seedance-2.5 |
|---|---|
| Aspect Ratio | 21:9, 16:9, 4:3, 1:1, 3:4, 9:16; supports Smart Aspect Ratio. For first-frame, first-and-last-frame, extension and editing scenarios, only Smart Aspect Ratio is supported — the output inherits the ratio of the input reference. |
| Resolution | 480p, 720p. Additional resolutions in future releases. |
| Duration | Any integer from 4 to 30 seconds. Supports Smart Duration. For editing scenarios, only Smart Duration is supported — the output inherits the input duration. |
| Frame Rate | 24 FPS |
| Reference Assets | Image, video and audio references. Newly supports audio-only references for high-quality audio-video synchronization. |
| Output Format | MP4; MOV newly supported. |