ByteDance has introduced Seedance 2.5, a new version of its AI video model built to generate video and audio together. The release gives creators a way to produce clips up to 30 seconds long, while also using reference images, video, and audio to guide the result.
The update matters because AI video tools are moving from short visual experiments toward more complete production workflows. Instead of creating separate pieces and assembling them afterward, Seedance 2.5 is designed to produce more of the scene in one pipeline.
What Seedance 2.5 can generate
Seedance 2.5 creates video and audio in one pass. That means the model is not limited to silent footage or a workflow where sound must be added separately after the visual clip is generated.
The model can produce clips up to 30 seconds long. Those clips can also be extended multiple times, giving users a way to build beyond a single generated segment while keeping the workflow inside the same system.
That length is a key part of the release. In the same source comparison, Google's Gemini Omni Flash tops out at about 10 seconds. Seedance 2.5 therefore gives users a longer clip window for AI video with built-in audio.
References give creators more control
ByteDance is also emphasizing the amount of source material users can bring into a generation. Seedance 2.5 accepts up to 30 images, 10 video clips, and 10 audio files as references.
Those references can be used to create scenes with multiple characters and camera angles. For teams working on ads or visual content, that could make the system more useful than a simple prompt-to-video generator, because the output can be guided by more than text alone.
The source article also notes improvements in textures, lighting, and skin detail. Those are important areas for AI video because visual believability often depends on small surface details, consistent light, and how people appear on camera.
- Seedance 2.5 generates video and audio together.
- Clips can be up to 30 seconds long.
- Users can extend clips multiple times.
- The model can use images, videos, and audio files as references.
- ByteDance says textures, lighting, and skin detail have improved.
Why the workflow shift matters
The practical value of Seedance 2.5 is not only the longer clip length. It is the combination of longer generation, audio, and reference-driven creation inside one workflow.
For ads and visual content, the source frames this as a way for teams to build full productions in one pipeline instead of stitching together individual clips. That distinction is important. A tool that only creates isolated shots may still require substantial manual assembly, while a more integrated pipeline can reduce the number of separate production steps.
Seedance 2.5 was also showcased through a short film called “The Missing Pair,” which ByteDance produced entirely with the model. The example is meant to demonstrate how the system performs when used for a complete piece rather than a single test clip.
The earlier Seedance 2.0 release had already been used in a larger AI film context. According to the source, “District 9” director Neill Blomkamp used Seedance 2.0 to create “Nightborne,” a 13-minute short film generated entirely with AI.
Where Seedance 2.5 is available
Seedance 2.5 is live on Jimeng AI and Doubao Pro. API access through BytePlus ModelArk is planned to come later.
The previous version, Seedance 2.0, leads the image-to-video leaderboard for models with audio, according to Artificial Analysis. That ranking gives useful context for ByteDance’s position in AI video, though the new release is presented as a step forward from that earlier model.
For creators and production teams, the main question is how reliably the model can turn references into usable scenes across multiple characters, camera angles, and extended clips. The source does not provide performance benchmarks for Seedance 2.5, but it does make clear that ByteDance is positioning the model around longer AI video clips, built-in audio, and more consolidated production.
In plain terms, Seedance 2.5 points to where AI video generation is heading: fewer disconnected assets, longer generated sequences, and more control through reference material. The release does not remove the need for creative direction, but it gives users a broader starting point for building scenes with audio already included.