ByteDance Seedance 2.5 adds 30-second AI video with audio
ByteDance says Seedance 2.5 can generate 30-second clips with audio, giving creators a longer one-pass video tool for ads and shorts.
By Wei-Lin Zhao · AI Correspondent
· 3 min read
ByteDance Seedance 2.5 is the company’s newest AI video model, and ByteDance says it can generate video and audio together in clips of up to 30 seconds. The release matters for creative tools and ad production because longer native clips with synchronized sound reduce the amount of stitching teams need to do across separate generation steps.
The company says Seedance 2.5 can create a clip in one pass and then extend it multiple times. That is a meaningful product detail in a category where many systems still produce short fragments that have to be assembled, matched and revised outside the model. By comparison, The Decoder reported that Google’s Gemini Omni Flash generates clips of about 10 seconds.
ByteDance did not disclose pricing, model size, training data details or performance benchmarks in the announcement. It also did not give a specific date for API availability beyond saying access through BytePlus ModelArk will come later.
What is Seedance 2.5?
Seedance 2.5 is ByteDance’s AI model for generating video, now with audio produced as part of the same generation process. The model is aimed at creating longer scenes, including projects that require multiple characters, reference material and camera variation.
ByteDance says users can provide up to 30 images, 10 video clips and 10 audio files as references. That input set is meant to help the model construct scenes with different characters and camera angles while keeping creative direction closer to the supplied material.
The company also claims improvements in textures, lighting and skin detail. Those are standard areas of competition in AI video, where visual consistency and human rendering remain practical limits for commercial use. ByteDance did not publish third-party test results for those specific improvements.
Where is Seedance 2.5 available?
Seedance 2.5 is available now through Jimeng AI and Doubao Pro, according to ByteDance. API access is planned through BytePlus ModelArk, the company’s model platform, but ByteDance has not said when developers will be able to use it there.
ByteDance is positioning the release for advertising and visual-content workflows. The company showed a short film, “The Missing Pair,” which it says was produced entirely with Seedance 2.5. The claim is useful as a demo of workflow ambition, though it does not say how much human editing, prompting or post-production was involved.
The new model follows Seedance 2.0, which Artificial Analysis ranks at the top of its image-to-video leaderboard for models with audio. That earlier version also drew attention outside product demos: “District 9” director Neill Blomkamp used Seedance 2.0 to make “Nightborne,” a 13-minute short film generated entirely with AI.
For ByteDance, the release keeps its video generation stack in the same competitive conversation as Google and other AI labs pushing toward longer, more controllable clips. For creative teams, the practical question is whether Seedance 2.5’s longer duration and reference inputs reduce production time enough to outweigh the usual review burden around consistency, rights and brand control.
This story draws on original reporting from The Decoder.
