August 7, 2026 • 5 Min Read

Seedance 2.5 is live on Melius: 30-second video with audio

Alex Chen, Software Engineer
Seedance 2.5 is live on Melius: 30-second video with audio
Alex Chen, Software Engineer

Seedance 2.5 is live on Melius. ByteDance's new video model generates up to 30 seconds in a single take, with audio produced alongside the picture instead of layered on after, and it accepts up to 50 reference files across images, video, and sound. It is in the video model picker now, on the same credit pool as everything else.

We wrote about what Seedance 2.5 was promising back in July. This is the shorter, more useful version: what actually shipped, what you can run today, and what is still to come.

Seedance is now more than half of every video on Melius

Context for why a Seedance release matters here more than most. Over the last 30 days, the Seedance family accounted for 54.6% of every video generation on Melius. When we last published this number it was 26.7%. Nobody is routed to it by default. People run it next to Kling, Veo, Sora, and Wan on the same canvas and keep picking it.

Seedance's share of video generations on Melius
0%20%40%60%MarchAprilMayJuneAugust
Share of finished video jobs on Melius run on a Seedance model. The climb through July ran entirely on Seedance 2.0.

What is new in Seedance 2.5

  • 30 seconds in one take. Double Seedance 2.0's ceiling, and long enough to be a finished shot rather than a loop you have to extend. Melius exposes every length from 4 to 30 seconds.
  • Up to 50 reference files. Thirty images, ten videos, and ten audio clips, capped at 50 in total. This is the feature that holds a character or a product on-model across a long shot.
  • Video-to-video with references. Feed in existing footage plus reference images and a prompt, and restyle or re-shoot it rather than starting from a blank frame.

What you can run today on Melius

Setting

What Melius supports

Inputs

Text, image, video, and audio

Clip length

4 to 30 seconds, any whole second

Resolution

480p and 720p, 720p by default

Aspect ratios

21:9, 16:9, 4:3, 1:1, 3:4, 9:16

Audio

Generated in the same pass, toggleable per node

Reference inputs

Up to 30 images, 10 videos, 10 audio clips, 50 files total

Seed

Supported, so a take you like is reproducible

The resolution row is the one to watch. Seedance 2.5 runs at 480p and 720p on Melius today because that is what is publicly available on the model's API. ByteDance has demonstrated 1080p and 4K for the Seedance family, and those tiers go into the picker as soon as they open up publicly.

Getting a good take out of it

  1. Start from a frame you already like. Image-to-video outruns text-to-video by better than two to one in our data, because a validated still removes half the uncertainty. Which means the image model you generate that still with decides a lot about the clip.
  2. Use the reference budget. If you rationed reference frames to fit older limits, stop. Fifty files is room for a character sheet, a product from several angles, a style board, and a voice sample all at once.
  3. Compare it against the field. Because Melius puts every model under one subscription and one credit pool, running the same shot through Seedance 2.5, Kling, and Veo side by side costs you nothing extra in setup.

Open Melius and pick Seedance 2.5 on any video node.

Frequently asked questions

Is Seedance 2.5 out?

Yes. ByteDance released Seedance 2.5 and it is live in the Melius video model picker. You can run it in text-to-video, image-to-video, reference-to-video, and video-to-video, with generated audio in the same pass.

How long can a Seedance 2.5 video be?

Up to 30 seconds in a single generation. Melius exposes every whole-second length from 4 to 30 seconds, so you set the shot length rather than stitching a 5-second loop into something longer.

What resolution does Seedance 2.5 support on Melius?

480p and 720p today, with 720p as the default. ByteDance has demonstrated 1080p and 4K for the Seedance family, and Melius will add those tiers to the picker as soon as they are publicly available on the model's API.

Does Seedance 2.5 generate audio?

Yes. Audio is produced together with the video rather than dubbed on afterwards, so footsteps, speech, and ambience line up with what is on screen. There is a sound toggle on the node if you want a silent clip.

How many reference inputs does Seedance 2.5 take?

Up to 30 reference images, 10 reference videos, and 10 audio clips, capped at 50 files in total across all three. That is what lets you hold a character, a product, or a house style steady across a 30-second shot.

What is the difference between Seedance 2.5 and Seedance 2.0?

Seedance 2.5 doubles the maximum clip length from 15 to 30 seconds, raises the reference ceiling to 50 files across images, video, and audio, and adds video-to-video with reference conditioning. Seedance 2.0 stays in the picker and is still the cheaper, faster option for short shots.

How much does Seedance 2.5 cost?

It runs on the same shared Melius credit pool as every other image, video, and audio model, priced per second of output, with 480p costing roughly half of 720p. There is no separate Seedance subscription to buy, so you can compare it against Kling, Veo, and Sora on the same budget.

What aspect ratios does Seedance 2.5 support?

21:9, 16:9, 4:3, 1:1, 3:4, and 9:16. The vertical ratios matter most in practice, since a large share of the video work on Melius is cut for social.

Related Articles