What Is Seedance? ByteDance’s AI Video Model, Version by Version

What is Seedance: ByteDance’s AI video model family, from Seedance 1.0 in June 2025 to Seedance 2.5, with release dates, features and limits.

By Melies AI · Updated · 6 min read

Two young men in denim jackets walking down a 1990s New York street, a frame from a Seedance image-to-video clip

Seedance is ByteDance’s family of AI video generation models. It makes short video clips, with sound since late 2025, from a text prompt, from images, and in newer versions from video and audio references too. ByteDance’s Seed research team builds it, and the newest version is Seedance 2.5, released on July 31, 2026. This article covers who makes Seedance, every version and its release date, what the model can and cannot do, and how it differs from ByteDance’s other models.

Who makes Seedance, and is it Chinese?

Yes, Seedance is Chinese. ByteDance, the Chinese technology company behind TikTok, Douyin and CapCut, owns Seedance, and its Seed research team develops it. Seedance is a model, not a company or a standalone app. ByteDance puts it inside its own products: Jimeng AI and Doubao in China, Dreamina, CapCut and the Dola assistant internationally, and its BytePlus cloud for developers outside the United States.

Every Seedance version and its release date

ByteDance released four main Seedance versions between June 2025 and July 2026: 1.0, 1.5 pro, 2.0 and 2.5, each with a launch post on the ByteDance Seed website. Seedance 2.0 also comes in lighter Fast and Mini editions. Seedance 2.5 is the newest, and Seedance 3.0 has not been announced.

Seedance versions and release dates
VersionReleasedSoundWhat it added
Seedance 1.0 (Pro and Lite)June 11, 2025No1080p text to video and image to video, with several shots in one clip
Seedance 1.5 proDecember 16, 2025YesThe first Seedance with sound: voices, effects and music made with the picture, lip sync in six languages
Seedance 2.0February 12, 2026Yes, two-channelText, image, video and audio inputs (up to 9 images, 3 videos and 3 audio clips), 15-second multi-shot clips, editing and extension
Seedance 2.0 Fast and Mini2026YesQuicker and lower-cost versions of Seedance 2.0 for drafts and short clips
Seedance 2.5July 31, 2026Yes30 seconds in one generation, repeated extensions, up to 30 images, 10 videos and 10 audio clips, timestamp-level editing
Seedance 3.0Not announced—No official details as of November 18, 2026

Seedance 1.0 came out on June 11, 2025. ByteDance built it for steady motion and multi-shot storytelling: its demo is a 10-second clip with two to three shot changes. ByteDance reports that it rendered a 5-second 1080p clip in 41.4 seconds on an NVIDIA L20 and ranked first on Artificial Analysis’s text-to-video and image-to-video leaderboards in June 2025. BytePlus sold it in a Pro version and a cheaper Lite version.

Seedance 1.5 pro, released on December 16, 2025, added sound generated together with the video. ByteDance lists speech in Chinese, English, Japanese, Korean, Spanish and Indonesian, plus regional Chinese dialects such as Sichuanese and Cantonese.

Seedance 2.0, released on February 12, 2026, is the version that made Seedance widely known. It takes four kinds of input at once and makes multi-shot clips of up to 15 seconds with two-channel audio. ByteDance also sells two lighter versions, Seedance 2.0 Fast and Seedance 2.0 Mini, for quicker and cheaper drafts. To start making clips, see where and how to use Seedance 2.0.

Seedance 2.5, released on July 31, 2026, doubles the length of one generation to 30 seconds and can extend a clip several times for multi-minute videos. It rolled out first on Jimeng AI and Doubao Pro in China, then on Dreamina, CapCut and BytePlus ModelArk. BytePlus describes it as taking up to 50 multimodal references in one clip. You can also make 30-second Seedance 2.5 clips online.

What can Seedance do?

Seedance makes video in three main ways, and the newer versions also edit video. Each mode suits a different starting point:

  • Text to video: a written prompt becomes a clip, including camera moves, several shots and spoken dialogue.
  • Image to video: a still image becomes the first frame, and an optional second image becomes the last frame.
  • Reference to video: images, and in Seedance 2.0 and 2.5 also video and audio clips, set the characters, products, places, motion or voice the clip must keep.
  • Editing and extension: from Seedance 2.0 on, a prompt can change part of an existing clip or continue it. Seedance 2.5 adds edits tied to exact timestamps, green-screen background swaps and camera changes.

ByteDance also names the weak spots. For Seedance 2.5, it says complex motion is not always physically plausible and scenes where several people interact can be unstable. For Seedance 2.0, it lists detail stability and occasional audio distortion as areas still being improved.

Is Seedance 3.0 out?

No. As of November 18, 2026, ByteDance has not announced Seedance 3.0, and the ByteDance Seed website lists Seedance 2.5 as its newest video model. Websites that offer a “Seedance 3.0 generator” are not run by ByteDance, and posts about 3.0’s length or release date are rumours. A new version would be announced on the Seed blog first, as every earlier version was.

Why did Seedance 2.0 go viral and draw criticism?

Seedance 2.0 went viral in February 2026 because people used it to make realistic clips of famous actors and characters, such as a fight between Brad Pitt and Tom Cruise and Will Smith battling a spaghetti monster. The Motion Picture Association criticised it over copyright, and Disney sent ByteDance a cease-and-desist letter on February 13, 2026. On February 16, ByteDance said it would strengthen its safeguards. Its platforms now limit real faces: a real person’s portrait needs identity verification or legal authorization before it can be used as a reference.

Is Seedance open source?

No. ByteDance has published technical reports for Seedance 1.0, 1.5 pro and 2.0, but it has not released the model weights of any Seedance version. You cannot download Seedance or run it on your own GPU. You use it through ByteDance’s apps, its BytePlus API, or platforms that pay for that API.

Is Seedance free, and where can you use it?

Seedance is free to try in small amounts, but not unlimited. ByteDance’s Dreamina gives free accounts a daily credit allowance with watermarked downloads, and some independent sites give sign-up credits; Melies AI gives 4 credits on Google sign-up, enough for one short Seedance 2.0 Mini clip. Longer or higher-quality clips need a paid plan or credits on every platform. Our guide compares the platforms that run Seedance, and a free Seedance 2.0 trial runs in the browser.

How is Seedance different from Seedream?

Seedance makes video and Seedream makes still images. Both come from ByteDance’s Seed team, and both appear on the Seed website’s model list. A common workflow uses them together: make a picture with Seedream, then animate it with Seedance as the first frame. On this site, ByteDance’s Seedream image model has its own generator.

Sources

More questions about Seedance

Something else? Email support@meliesai.com.

Seedance is not a company. It is an AI video model made by ByteDance, the Chinese technology company that also owns TikTok, Douyin and CapCut. ByteDance’s Seed research team develops it and ByteDance offers it in its own apps, such as Dreamina and CapCut, and through its BytePlus cloud.

Seedance 2.0 Mini is the lowest-cost version of ByteDance’s Seedance 2.0. It makes the same 4- to 15-second clips with sound from text, images or references, at a lower price than full Seedance 2.0. It suits drafts, tests and short social clips.

Seedance 2.0 Fast is the quicker-rendering version of Seedance 2.0. It keeps the same inputs and 4- to 15-second clip length with sound, and costs less than full Seedance 2.0. Use it to try several prompt ideas before rendering the final version on the full model.

Yes, from Seedance 1.5 pro onward. Seedance 1.5 pro, 2.0 and 2.5 generate dialogue, sound effects and music together with the picture, with lip sync. Seedance 2.0 and 2.5 produce two-channel audio and can take audio clips as references. Seedance 1.0 made silent video.

One Seedance 2.0 generation makes 4 to 15 seconds. One Seedance 2.5 generation makes up to 30 seconds, and ByteDance says it can extend a clip several times for videos lasting several minutes. Not every platform offers extension, so the longest single clip depends on where you use it.

Seedance is among the leading AI video models for motion, multi-shot scenes and sound, by ByteDance’s own tests and Artificial Analysis’s leaderboards in June 2025. It is weaker at physically complex motion and scenes with several interacting people, by ByteDance’s own account. Results depend on the prompt, so test a few clips first.

The official Seedance pages are on ByteDance Seed’s website, seed.bytedance.com, with one page per version and its launch post. ByteDance’s consumer apps for using Seedance are Dreamina and CapCut internationally, and Jimeng AI and Doubao in China. Other Seedance-named websites are independent.

No. ByteDance’s platforms block nudity, sexual content, hate and the unauthorized use of a real person’s likeness, through prompt blocking and content moderation. Independent platforms that run Seedance through ByteDance’s API usually add their own content rules on top. Refused prompts usually return an error instead of a video.

Make your first Seedance clip

Write a prompt or upload an image and get a video with sound in a few minutes.

Start creating