Seedance (ByteDance)

Seedance is ByteDance's family of video generation models, made by the company behind TikTok and CapCut. Its current model, Seedance 2.5, released on July 31, 2026, generates up to 30 seconds of video with native audio from as many as 50 reference files, and as of October 20, 2026 it ranks third on Artificial Analysis's text-to-video-with-audio leaderboard. Seedance is also the most legally contested model in generative video: its predecessor, Seedance 2.0, drew cease-and-desist letters from three major studios within days of its release.

Seedance 2.0 and the Copyright Fight

ByteDance released Seedance 2.0 on February 12, 2026. Rights holders objected almost at once: Disney sent ByteDance a cease-and-desist letter on February 13, Paramount Skydance on February 15 and Netflix on February 17, and the Motion Picture Association and SAG-AFTRA condemned the model. US senators wrote to ByteDance's chief executive on March 16. ByteDance said it respects intellectual property and strengthened its safeguards, and when Seedance 2.0 rolled out globally in CapCut on March 26 it blocked real faces, added watermarks and left out the United States. No lawsuit or settlement had been reported as of October 2026.

What Seedance 2.5 Does

• Length. 4 to 30 seconds per pass, extendable twice, with a beta long-video mode of up to three minutes in ByteDance's Dreamina app.
• References. Up to 50 per generation (30 images, 10 video clips and 10 audio clips), up from 12 in Seedance 2.0, which helps keep characters, products, locations and soundtracks consistent.
• Audio. Sound and lip-synced speech generated with the picture.
• Resolution. 480p and 720p through the major API hosts; some ByteDance surfaces advertise higher resolutions.
• Weights. Closed; no open-weight Seedance model has been released.

Where It Runs

ByteDance distributes Seedance mainly through its own consumer apps. Seedance 2.5 launched on Dreamina, and on Jimeng and Doubao in China, and reached CapCut's desktop app on August 7, 2026 for Pro and Ultra members, billed per second in credits; CapCut's pricing page quotes it "from $0.12/sec". Developers can use BytePlus ModelArk, ByteDance's enterprise platform run by a Singapore entity, or third-party hosts, where per-second prices vary widely. Creative platforms such as Runway, Higgsfield and Kapwing offer Seedance 2.5, and HeyGen bundles Seedance 2.0 in its Pro plan. Whether Seedance 2.5 is available to CapCut users in the US has not been confirmed.

How It Compares

Seedance sits in a tight cluster at the top of the leaderboards. On Artificial Analysis's text-to-video-with-audio board as fetched on October 20, 2026, it scored an Elo of 1159, behind Alibaba's Wan 3.0 (1173) and Utopai X (1166) and level with Black Forest Labs' FLUX 3 Max. Kuaishou's Kling 4.0, released in mid-October, also reaches 30 seconds per pass, offers keyframe control and an MCP server, but had not yet been rated. Against Wan 3.0, Seedance is stronger on reference depth and lip-sync, while Wan has published per-second API prices and an open-weight predecessor. Against Google's Gemini Omni, it offers longer clips and more references but no conversational editing.

Two considerations apply beyond quality. First, the rights scrutiny that followed Seedance 2.0 means brands should put Seedance output through legal review and take care with references and likenesses (see deepfakes and AI content labeling). Second, ByteDance is headquartered in Beijing, so teams handling sensitive material should check where requests and uploads are processed (see data sovereignty), or use the model through a Western platform whose terms they have reviewed. See Seedance vs Kling, Wan vs Seedance and text-to-video.