Seedance 1.5 pro

What it is

Seedance 1.5 pro is an audio-video generation model announced by Volcano Engine at its FORCE conference on 18 December 2025. It was presented on the same occasion as the Doubao Large Model 1.8^[1]. A language model works on text; this one works on footage, producing a video clip together with the sound that belongs to it.

Sound and picture generated together

The most distinctive thing about the model is a choice made at the level of architecture. Seedance 1.5 pro uses a native joint audio-video generation architecture, so picture and sound are produced in a single process rather than in two separate stages^[1]. The sound side is not one flat track either: the model generates ambient sound, background music and human voices together with the picture^[1].

A common alternative is to finish a silent clip first and add dialogue, music and effects afterwards. When the two halves are made separately, small mismatches creep in — a line that lands slightly late, a mouth that closes a fraction after the voice stops. Seedance 1.5 pro is built to avoid that split; Volcano Engine says it achieves millisecond-level audio-video synchronisation^[1].

Dialogue is not limited to a single speaker. The model supports multi-person, multilingual conversation with accurate lip sync^[1]. The languages it covers include English and smaller languages, and on the Chinese side dialects such as Sichuanese and Cantonese^[1]. For a creator working in a regional language, that means people can appear speaking the way they actually speak, instead of being dubbed over afterwards.

Drafting before committing

Volcano Engine also described a forthcoming feature called Draft^[1]. A user can preview a low-resolution draft that closely matches the final video before committing to a full generation^[1]. The point is like making a sketch before a painting: framing, timing and performance can be judged while changes are still cheap.

Volcano Engine said the feature raises overall efficiency by 65% and cuts wasted creation cost by 60%^[1]. Both figures are the company's own.

How to use it

At launch, individual users could try the model in the Doubao and Jimeng AI apps^[1]. Enterprise access through an application programming interface (API) on Volcano Engine opened on 23 December 2025^[1][2]. The two routes put the same model in front of very different users: people experimenting inside a phone app, and companies calling the model from their own code and wiring it into production systems.

Who it is aimed at

Volcano Engine pitched the model at video creators. In its framing, the model could carry film-grade narrative tension and capture motion detail and character emotion^[1]. That says something about the kind of footage it is meant to produce — not only short decorative clips, but scenes with pacing and performance in them.

Where it sits

The release came on the same day as the Doubao Large Model 1.8^[1]. Announcing a general large model and a media generation model together suggests two lines being pushed at once: one that handles language and reasoning, one that handles image and sound. Seedance 1.5 pro's contribution to the second line is the decision to treat sound as part of generation rather than as a step that comes after the picture is finished^[1]. Whether that holds up across long, dialogue-heavy scenes is the kind of question that only use by creators will settle.