ByteDance's video model, and one of the few that produces its audio at the same time as its picture rather than having a soundtrack laid over afterwards. It takes a prompt on its own or a still frame to start from, and up to nine reference images to hold a look across shots.
Which plan: Included from Starter.
Audio is generated with the frames, so footsteps and room tone match what is on screen.
Hand it a frame you already generated and it animates that exact image forward.
The widest reference lane of the video models here — enough to hold a character and a set.
Arrange action into beats and give references distinct jobs.
A courier enters the shop, sets a parcel on the counter and waves to the shopkeeper. The camera follows through the doorway, then holds at the counter.
Read the prompting tutorial. Select this model before using Enhance prompt to apply its guidance to your draft.
Sources: Seedance 2.0 reference examples.
| Setting | Details |
|---|---|
| Made by | ByteDance |
| Runs | Text to video, and image to video |
| Sound | Generated with the picture |
| Reference images | Up to 9 |
| Opens at | Starter |
A paper lantern drifting up past a temple roof at blue hour, slow push-in, embers rising, distant wind and a single bell
Open Seedance 2.0 Mini in the studio — Choose the model and review your prompt and quote before generating.
Yes — the audio is produced alongside the picture rather than added afterwards, so it matches the motion on screen.
Yes. Attach a still and the same model runs image to video; with nothing attached it runs text to video.
It opens at the Starter plan. There is a free video lane, just not this model: LTX 2.3 and MiniMax H3 Studio run on our own GPUs and are open to free accounts, which daily credits stretch to two or three short clips a day. Seedance bills us per generation, so it needs a plan.