Seedance 2.5 reportedly generates 30-second clips in a single pass with up to 50 multimodal references and 4K output. ByteDance has not published full specs yet. Seedance 2.5 will be available on fal alongside Seedance 2.0 and Seedance 2.0 4K.
Seedance 2.5 is ByteDance's next video generation model, and we're all excited about it and its coming to fal.
In this guide, I'll go over everything that is publicly known and also assumed by the industry about Seedance 2.5, including how it'll compare against Seedance 2.0.
TL;DR
Seedance 2.5 has been announced as coming and has been widely reported by media channels.
The two main upgrades over Seedance 2.0 are native 30-second clips and support for up to 50 reference inputs.
ByteDance has not published full specs yet, so treat the details here as early reporting until the model ships.
Seedance 2.5 will be available on fal, and you'll be able to benefit from fal's unified API with no subscription costs and use Seedance 2.5 on a pay-per-use model (typically, pay by second for video generators, which is Seedance 2.0's payment structure).
What is Seedance 2.5?
Seedance 2.5 is the newest model in ByteDance's Seedance video line, which belongs to the company's Seed family of foundation models spanning image, audio (Seed-audio 1.0), and language work (Doubao 2.1 Pro).
The Seedance models are built around motion quality and the camera and scene control that production work needs, with consistency held from one shot to the next.
The jump from Seedance 2.0 is about two numbers:
Clip length, which reportedly moves up to 30 seconds in a single pass.
Reference inputs, which move up to 50.
Native 30-second generation
Seedance 2.5 reportedly generates a full 30 seconds of video in one pass.
That is the part I noticed first, as length has been the main thing holding video models back for some time now.
Today's models tend to top out at a handful of seconds, so anything longer gets assembled from several short clips joined end to end, or with extenders like Veo 3.1 extend video.
From an advertising perspective, that will mean marketers will now be able to direct and create a single 30-second ad from start to finish.
Up to 50 multimodal references
The other change is on the input side: up to 50 references per request, according to CNET.
A reference is anything you give the model to steer the result, from still images and video clips to audio and the specific frames you want it to land on.
CNET reports these can be mixed together, so images, video, and audio all feed one generation.
The practical effect is that you can describe a shot the way you would brief a crew, with the character, the setting, the camera move, and the mood all specified at once.
When the model has that much to work from, the first result tends to land closer to what you actually wanted.
One number to pin down is the baseline this improves on, since fal's current Seedance 2.0 reference endpoint already takes up to nine images, three videos, and three audio clips (although total files across all modalities must not exceed 12).
Will Seedance 2.5 generate in 4K?
Yes, Seedance 2.5 will be able to generate videos in 4K.
CNET says Seedance 2.5 can render those 30-second clips in 4K, which would put it on par with the ultra-HD that Seedance 2.0 4K already offers.
falMODEL APIs
The fastest, cheapest and most reliable way to run genAI models. 1 API, 100s of models
Where will Seedance 2.5 be available when it launches?
Once Seedance 2.5 ships, fal is lined up to host it next to Seedance 2.0 and Seedance 2.0 4K, both running on the platform today.
If you're already using fal and our API, Seedance 2.5 would arrive on the same setup you already use, with nothing new to wire up.
For a model like Seedance 2.5, a few things about fal matter more than usual:
A 30-second 4K clip is heavy to render, and fal's own inference engine is tuned for exactly that kind of load, with good cold starts and GPU capacity that scales up and back down on its own.
A 50-reference, multi-input workflow that stays simple when every model runs behind one API and one key, so you are not juggling separate integrations to feed it images, video, and audio.
Billing is by the second or by the output with no subscription, so heavy bursts of generation do not leave you paying for idle hardware afterwards.
New ByteDance models tend to reach fal quickly, often on launch day, and our platform already runs production workloads for teams like Adobe, Canva, and Shopify.
Editor's opinion: What do the upgrades mean for production workflows?
When you put the two changes together in Seedance 2.5 from 2.0, the model starts to feel less like a prompt box and more like a set of instructions you hand off.
Thirty seconds in one pass will mean a scene can play out without the visible cuts that come from splicing short clips, which will help with anything from a short ad to a longer narrative shot.
Fifty references will also mean the look you have already locked down can travel straight into the generation, so the model has your actual material to work from, and less is left to guesswork.
For teams working from an existing asset library, that is the real win: more of what you have built feeds the output, and you burn fewer attempts getting something usable.
I believe that this model will be loved by advertisers and short-form video creators, since they'll be able to create quality 4K content up to 30 seconds, and be able to add significantly more references than they were able to add previously.
How does Seedance 2.5 compare to Seedance 2.0 and Seedance 2.0 4K?
The Seedance 2.0 models are live on fal today, so they are the clearest reference point for what 2.5 changes.
Here is how the three line up on the details available now, with the Seedance 2.0 figures taken from fal's published model pages and the Seedance 2.5 figures marked as reported.
| Detail | Seedance 2.0 | Seedance 2.0 4K | Seedance 2.5 (reported) |
|---|---|---|---|
| Status | Live on fal | Live on fal | Announced, not yet released |
| Max clip length | Up to 15 seconds | Up to about 15 seconds | Up to 30 seconds |
| Resolution | Up to 720p | Up to 4K | 4K |
| Inputs | Text, image, audio, video | Text, image, audio, video | Text, image, audio, video |
| Reference inputs | Up to 9 images, 3 videos, 3 audio (reference-to-video) | Same family inputs | Up to 50 multimodal (reported) |
| Native audio | Yes | Yes | Expected, couldn't confirm |
Try Seedance 2.5 on fal when it goes live
When 2.5 does land on fal, it'll run through the same API and playground you set up for Seedance 2.0, and other AI video generators like Happy Horse 1.0 and Veo 3.1.
If you haven't signed up for fal yet, you can get started for free.
Recently Added
.webp){kind=small}
Frequently asked questions about Seedance 2.5
What is Seedance 2.5?
Seedance 2.5 is ByteDance's next-generation video model, the follow-up to Seedance 2.0 in the company's Seed family.
The standout reported changes are 30-second clips in a single pass and up to 50 mixed references per request.
When will Seedance 2.5 be released?
A launch is reported for around July, though ByteDance has not put an official date on it.
Until it does, the timing is best treated as a moving target.
What is the maximum Seedance 2.5 clip length?
Coverage points to 30 seconds generated in one pass.
That stands out because most models make shorter clips and reach length by joining them, which can show at the seams.
How many references can Seedance 2.5 use?
Reports put it at up to 50, with images, video, and audio allowed in the same request.
The figure it improves on is still worth confirming against ByteDance's eventual docs.
How is Seedance 2.5 different from Seedance 2.0 and Seedance 2.0 4K?
The 2.0 models are out now and make clips up to 15 seconds, with 2.0 topping out at 720p and 2.0 4K reaching 4K.
Seedance 2.5 reportedly pushes clip length to 30 seconds and references to 50, with 4K expected in the same model.
How much will Seedance 2.5 cost on fal?
No pricing has been announced by our team.
fal bills its video models as you go, by the second or by the output, with no subscription.
