Thirty Seconds in a Single Run
Seedance 2.5 generates up to 30 seconds in a single run, where most AI video generators stop at five or ten and leave you stitching. You can extend twice from there, so one idea can run to roughly a minute and a half without restarting the scene. ByteDance unveiled it at the Volcano Engine FORCE conference in Beijing, and it is rolling out through Jimeng AI and the Pro tier of Doubao.
Native 4K, 10-Bit Colour
Output is native 4K rather than an upscale, at 10-bit colour depth — smoother gradients and enough headroom to grade the footage afterwards instead of fighting banding in the sky. ByteDance also reports roughly 20% better prompt adherence than Seedance 2.0, so the shot you described is closer to the shot you get.
Audio Built in the Same Latent Space
Sound is not bolted on after the render. Seedance 2.5 processes audio inside the same latent space as the picture, so a footstep lands on the frame where the foot lands and a door slams when the door shuts. What you get back is a clip whose effects are already in sync, not a soundtrack you have to nudge into place.
Fifty References, and a 3D Preview First
One request takes up to 50 reference inputs — 30 images, 10 video clips and 10 audio clips — against 12 in the previous Seedance. There is also a 3D white-box preview: block the motion out as a rough low-fidelity animation, check the camera actually does what you meant, then spend the render on the version you want.
What can you create with Seedance 2.5? Get inspired with these ideas:
Short-Form Video
Thirty seconds is a whole TikTok or Reel, not a fragment — shoot the full beat in one run instead of stitching three generations and hoping the cuts match.
AI Films
Build scenes at native 4K with sound effects already locked to the action, then grade the 10-bit footage the way you would grade a real shoot.
Block Out, Then Edit
Rough the shot out as a white-box animation first, check the camera move reads, then render. On the way back it handles green-screen editing, camera movement and performance blocking, so a clip can be adjusted rather than regenerated.
Consistency Across a Sequence
Feed in up to 30 images, 10 clips and 10 audio references at once so a character, a location and a look survive from one shot to the next.
Product Marketing
Turn a product into a 4K spot long enough to actually tell a story, with the packaging and colours pinned by reference frames.
Image to Video
Start from a still and let Seedance 2.5 move it, or hand it reference clips and audio and let it reshape those instead.
Create with Seedance 2.5 in 3 simple steps
Describe What You Want
Write your prompt and attach any reference images, clips or audio you want Seedance to follow.
Seedance 2.5 Generates It
The model renders picture and synced sound together, up to 30 seconds at native 4K.
Download and use
Get your result ready to share, post, or integrate into your projects.
What is Seedance 2.5?
Seedance 2.5 is ByteDance's video generation model, unveiled at the Volcano Engine FORCE conference in Beijing. It produces a 30-second clip in a single run at native 4K, takes up to 50 reference inputs, and generates sound in the same pass as the picture. It is rolling out through Jimeng AI and the Pro tier of Doubao.
How long can a Seedance 2.5 video be?
Up to 30 seconds in a single generation, which is where it separates from AI video generators that top out around five to ten. You can then extend twice, so a single idea can run to roughly a minute and a half without starting the scene over.
Does Seedance 2.5 generate 4K video?
Yes, natively rather than by upscaling a smaller render, and at 10-bit colour depth. That extra depth is what gives you smooth gradients in skies and skin, and enough latitude to colour grade the clip afterwards.
Does Seedance 2.5 make sound as well as video?
It does, and the sound is produced inside the same latent space as the picture rather than added afterwards. That is what keeps effects synchronised to what is happening on screen, so a footstep or an impact lands on the right frame without manual alignment.
How many reference files can Seedance 2.5 take?
Up to 50 in one request: 30 images, 10 video clips and 10 audio clips, alongside style references and 3D white models. The previous Seedance accepted 12, so this is the change that makes character and location consistency across a sequence practical.
When and where can I use Seedance 2.5?
ByteDance announced Seedance 2.5 at its Volcano Engine FORCE conference and is rolling it out through Jimeng AI and the Pro tier of Doubao, with an API expected on the Volcano Engine Ark platform. There is no public date for wider availability yet. Join the early access list above and we will email you the moment it is live on Overchat AI.
