How is Seedance 2.5 different from Seedance 2.0?

Seedance 2.5 doubles the single-take duration from 15 seconds to 30 seconds and raises the reference-asset ceiling to 50 total (30 images, 10 videos, 10 audio clips) per input. It also adds timestamp-level editing and a green-screen mode for localized changes, and refines image quality to reduce plastic-looking artifacts.

Can I use Seedance 2.5 today?

Seedance 2.5 is rolling out on Jimeng AI and the Pro tier of ByteDance's Doubao consumer platform. Programmatic API access is coming to Volcano Engine's BytePlus ModelArk, with a launch expected roughly a week after the public rollout. It is not open source and no weights are released.

How does Seedance 2.5 compare to Kling, Veo, and Sora?

Seedance 2.5's 30-second single-take pushes past Kuaishou's Kling 3.0 at 15 seconds and Google's Veo 3.1 at 8 seconds. OpenAI's Sora product line was discontinued in March 2026, so among mainstream video generators Seedance 2.5 currently offers the longest single-pass generation for reference-conditioned video.

Does Seedance 2.5 handle audio?

Yes. Seedance 2.5 keeps Seedance 2.0's unified audio-video architecture, so the model generates video and audio together in one pass and accepts up to 10 audio clips as reference inputs. This lets creators drive lip-sync, voice, and score from real recordings rather than editing sound onto a silent clip afterward.

ByteDance Seed · 2026-07-31 · major

Seedance 2.5 — ByteDance's video model doubles to 30 seconds per generation

ByteDance released Seedance 2.5, a video model that generates 30-second clips in one take and accepts up to 30 images, 10 videos, and 10 audio clips as references, with timestamp-level editing.

Frame from a Seedance 2.5 generated video clip

ByteDance's Seed lab doubles single-take video generation to 30 seconds with 50-asset multimodal referencing.

Quick facts

Maker	ByteDance Seed
Type	Video generation model
Max duration	30 seconds per generation
Reference inputs	Up to 30 images, 10 videos, 10 audio clips
Availability	Jimeng AI, Doubao Pro
API	Coming soon via BytePlus ModelArk
License	Proprietary

What is it?

Seedance 2.5 is ByteDance's new video generation model, released July 31, 2026. It generates 30 seconds of high-quality video in a single pass, doubling the 15-second limit of Seedance 2.0. It accepts up to 30 images, 10 video clips, and 10 audio clips as reference material per input, and supports multi-round extension for coherent longer sequences.

How does it work?

The model builds on Seedance 2.0's unified audio-video architecture and adds precise controls: timestamp-level editing lets creators target specific parts of a clip, and a green-screen mode plus reference-based editing handle localized changes without regenerating the whole video. Reference assets guide characters, scenes, camera perspective, and audio simultaneously.

Why does it matter?

For creators, Seedance 2.5 removes the two biggest limits of short-form AI video: length and multi-asset consistency. A 30-second single take with 50 reference assets covers most ad, storyboard, and short-film workflows in one call. It ships as OpenAI's Sora video product is discontinued, making Chinese video models the practical default for many teams.

Who is it for?

video creators, ad agencies, film pre-production teams

Frequently asked questions

How is Seedance 2.5 different from Seedance 2.0?: Seedance 2.5 doubles the single-take duration from 15 seconds to 30 seconds and raises the reference-asset ceiling to 50 total (30 images, 10 videos, 10 audio clips) per input. It also adds timestamp-level editing and a green-screen mode for localized changes, and refines image quality to reduce plastic-looking artifacts.
Can I use Seedance 2.5 today?: Seedance 2.5 is rolling out on Jimeng AI and the Pro tier of ByteDance's Doubao consumer platform. Programmatic API access is coming to Volcano Engine's BytePlus ModelArk, with a launch expected roughly a week after the public rollout. It is not open source and no weights are released.
How does Seedance 2.5 compare to Kling, Veo, and Sora?: Seedance 2.5's 30-second single-take pushes past Kuaishou's Kling 3.0 at 15 seconds and Google's Veo 3.1 at 8 seconds. OpenAI's Sora product line was discontinued in March 2026, so among mainstream video generators Seedance 2.5 currently offers the longest single-pass generation for reference-conditioned video.
Does Seedance 2.5 handle audio?: Yes. Seedance 2.5 keeps Seedance 2.0's unified audio-video architecture, so the model generates video and audio together in one pass and accepts up to 10 audio clips as reference inputs. This lets creators drive lip-sync, voice, and score from real recordings rather than editing sound onto a silent clip afterward.

Try it

Available on Jimeng AI and Doubao Pro; API access coming via BytePlus ModelArk.