# 字节发布视频模型 Dreamina Seedance 2.5

- 来源：Rohan Paul (@rohanpaul_ai)
- 发布时间：2026-08-03 12:54
- AIHOT 分数：59
- AIHOT 链接：https://aihot.virxact.com/items/cmscrhqp70kcvroeul9lvpsaf
- 原文链接：https://x.com/rohanpaul_ai/status/2084140807263498513

## AI 摘要

字节跳动推出视频模型 Dreamina Seedance 2.5，单片段最多支持 50 个参考文件，可一次生成 30 秒连续视频，前 7 天仅在 Dreamina 平台运行。该模型支持同时输入 30 张图像、10 段视频和 10 段音频，并允许在提示词中使用时间戳（如 [0s-5s]）作为分镜脚本。

## 正文

ByteDance has launched its new video model Dreamina Seedance 2.5 and its genuinely impressive.

Use up to 50 reference files for a single clip and produce 30 seconds of continuous video in one take.

For the first 7 days it will only run on Dreamina, making this a world-first global launch.

AI video has been mostly a guessing game until now: somebody types a long paragraph and hopes the model imagines the right face, the right room, the right camera move.

In Seedance 2.5 that is changed to 30 images, 10 videos and 10 audio clips all fed in at the same time.

This way the creator can use characters, sets, shot styles and music from files they already have.

Now with the ability to use timestamps in prompts, a creator can write [0s-5s] for one action and [5s-10s] for the next and the model will use that as a shot list.

1/