What is Seedance 2.5?

What Seedance 2.5 does, according to ByteDance

ByteDance Seed describes Seedance 2.5 as an audio-video joint generation model built on the architecture of Seedance 2.0. Its launch post of 31 July 2026 lists three main upgrades:

• Longer clips: up to 30 seconds in a single pass, with several shots that follow a story, and extensions that continue a clip with the same characters and setting.
• More references: up to 30 images, 10 video clips and 10 audio clips in one request, including clay-render, motion and style references.
• Editing by timestamp: change the camera, an action or a character in a chosen part of a clip, plus green screen and camera-angle edits.

Seedance 2.5 vs Seedance 2.0

Pick Seedance 2.5 for longer takes and story-like clips; pick Seedance 2.0 when a short clip is enough, or when you need 4K.

TikTok

TikTok

Where you can use Seedance 2.5

ByteDance rolled Seedance 2.5 out first on its own apps, Jimeng AI and Doubao Pro, and announced API access through BytePlus ModelArk. Other video sites run it through API providers, each with its own limits and prices.

Capafy AI is one of them: Seedance 2.5 is the default model of the studio, in both text to video and image to video. What we do not offer: video or audio references, extensions and timestamp edits. You can add up to 9 reference photos and choose 16:9, 9:16, 1:1, 4:3, 3:4, 21:9 as the ratio.

What Seedance 2.5 costs in Capafy AI

Seedance 2.5 bills by the second, so a 10 s clip costs twice a 5 s clip at the same resolution. The sound switch does not change the price.

How to prompt Seedance 2.5

• For a long clip, write it as timed beats: what happens in the first 5 seconds, the next 5, and so on. ByteDance's own examples are written this way.
• Say "one continuous take" when you want no cuts, or describe each shot when you want several.
• With several reference photos, say who comes from which photo: "the woman from photo 1", "the dog from photo 2".
• Describe the sound you want, from dialogue to ambience, since the model makes it with the picture.
• More on structure in how to write AI video prompts.

Seedance 2.5 limits

ByteDance says there is still room to improve the physical plausibility of complex motion and the stability of scenes where several subjects interact. In practice: keep fast, tangled action short, and test a scene with many people at 480p before you render it at 1080p.