Seedance 2.0 is ByteDance's video generation model: it renders clips of up to about 15 seconds with synchronized dialogue and multiple shots, working from text, images, video and audio. Released in February 2026, it went viral within days. Access runs through Dreamina, the CapCut companion app with trial credits on sign-up, through TikTok's AI Cast, or through a pay-per-second API for developers.
Seedance 2.0 accepts up to 9 images, 3 video clips and 3 audio tracks as references in a single generation. It is a text-to-video model at heart, yet each file gets called by @name inside the prompt (a habit you pick up fast), and the model reads character identity, camera moves or editing rhythm straight from your material.
Sound is not bolted on afterwards: the architecture generates picture and audio together, which is why lip-sync holds up across several languages. You describe three shots in one prompt and the sequence comes out as a single piece, cuts included. Targeted editing and clip extension round things off.
| Reference | Max per generation | What it controls |
|---|---|---|
| Images | 9 | Character identity, sets, composition |
| Video clips | 3 | Motion, framing, camera effects |
| Audio tracks | 3 | Lip-sync, editing rhythm |
| Text | Free-form prompt | Script and shot breakdown |
At launch, Seedance 2.0 topped composite AI video leaderboards ahead of Google DeepMind's Veo 3.1. What the public remembered, though, were the viral clips: a Brad Pitt versus Tom Cruise fight, the Friends cast as otters, Will Smith battling a spaghetti monster.
That realism reignited arguments around face swaps and deepfakes, to the point that ByteDance reportedly paused the global rollout for a while. The family has grown since: in June 2026 the company announced Seedance 2.5 with 30-second shots, while 2.0 picked up 4K output.
Dreamina is the easiest way in: roughly 535 trial credits during the first week, then paid plans. TikTok AI Cast hands out 4 free generations per day, while Doubao keeps its daily quota for accounts based in China.
On the developer side, the API bills per generated second, around $0.3034 at standard quality on fal.ai. These numbers shift with every model update, so check the official pages before reaching for your card.
Partly. TikTok AI Cast allows 4 free generations per day, Dreamina grants trial credits on sign-up, and Doubao's free daily videos require access from China. Beyond those quotas, you buy credits or pay per second through the API.
Announced in June 2026 at the Volcano Engine conference, Seedance 2.5 generates 30-second single shots and takes up to 50 reference files, against roughly 15 seconds and 15 files for 2.0. The older model stays available and gained 4K output at the same event.
Reference control is where Seedance 2.0 pulls ahead: it reads motion, camera work and audio from your own files and handles multi-shot sequences in one pass. Veo 3.1 answers with strong photorealism and tight ecosystem ties on Google's side. For music-driven edits and storyboard workflows, Seedance is the natural fit.
Realistic clips of famous actors and copyrighted characters flooded social feeds right after launch, alarming Hollywood over copyright. For professional work, keep recognizable faces and protected brands out of your prompts: platforms filter and remove that kind of footage.
Verdict: A scribbled storyboard at nine, a sequence with sound before lunch: that is the shortcut on the table here. Indie filmmakers, ad studios and marketing teams get a pocket film set, with classic editing still covering the longer formats.
