What’s new

Up to 30 seconds in one generation, twice as long as 2.0, and you can keep extending it. Thirty seconds is enough for several connected shots that take a small story from setup to ending.
Far more references: up to 30 images, 10 videos and 10 audio clips per generation, so characters, products, locations and sound can all be matched.
Edit one part only: point at a moment and change it instead of regenerating the whole clip. ByteDance also lists green-screen, clay-model and reference-video camera editing.
Smoother shot changes: according to ByteDance, cuts and scene changes hold together better, and materials, skin, eyes and lighting look more real.

What it’s good for

Complete short pieces under 30 seconds that need a beginning, middle and end: ads, product films, short-drama scenes.
Films where a character or product must look the same throughout: hand it reference images, video and audio together.
A clip that is almost right: fix one detail with a local edit instead of starting over.

Prompting tips

State the format first: duration, aspect ratio and number of shots, e.g. “30 seconds, 16:9, four connected shots”.
Write the timing in beats: 0–6 s setup, 6–18 s development, 18–26 s climax, 26–30 s closing frame.
Name shot size and camera move separately: “medium close-up, slow clockwise orbit”, not a move buried in style words.
Tie each reference to a person or object: say which reference image is whom and that they must stay consistent.
Attach sounds to actions: “rain, a distant tram, a suitcase latch” rather than “nice sound”.
Only list the exclusions that matter: “no face changes, no wardrobe changes, no extra people”.

30 seconds, 16:9, four connected shots. The girl in reference image 1 keeps the same face and the same yellow raincoat throughout.
A seaside town at dawn. 0–6 s: wide shot, she walks her bicycle down a cobbled slope. 6–18 s: medium tracking shot, she parks under the lighthouse and looks out to sea.
18–26 s: close-up, the sea wind lifts her hood and she smiles. 26–30 s: extreme wide closing frame on the lighthouse and the whole bay.
Sound: waves, a bicycle chain, distant gulls. No face changes, no extra people.

Known limits

ByteDance says the physical plausibility of complex motion and stability when several subjects interact still need work.
First-and-last-frame mode and multi-reference mode cannot be combined: with first and last frames set, you cannot add reference images, videos or audio.

Using it in Nomi

On the generation canvas, add a video card, pick “Seedance 2.5” in the model picker and choose a mode (Text to video, First frame to video, First and last frame, Multi-reference). Add your references, write the prompt and generate. In the storyboard you can pick it per shot. The first time, connect any one of the providers above under Settings → Models.