输入一段文字或上传一张照片,生成带原生音效的短视频 —— 由 Wan 2.5 驱动。浏览器里直接跑,无需安装、无需显卡,注册即送起步额度。
每个新账号都有起步额度 —— 无需绑卡
没有生成视频
Wan 是阿里开源的视频生成模型系列。权重公开,任何人都能下载运行——但跑它需要 GPU 算力,这就是各家托管服务收费的原因。
The weights are public. You can run Wan yourself in ComfyUI if you own the GPU — this site just saves you the hardware and the setup.
Footsteps land when feet land. Speech matches mouth movement. Syncing audio afterwards is the slowest part of short video, and Wan removes it.
Build a scene from a sentence, or keep your own photo and add motion to it. Image-to-video gives you far more control over the result.
Wan 2.5 costs about $0.05 per second at 480p and $0.15 at 1080p in real GPU time. We tell you where the free tier ends instead of surprising you.
把成本讲清楚,把能做什么和不能做什么讲清楚。
三步出片:写提示词或传图 → 选模型 → 生成。
Upload an image to animate it, or skip straight to a prompt if the scene does not exist yet.
"Camera slowly pushes in, her hair moves in the wind" beats "a beautiful woman". The model can already see the image — tell it what changes.
5 or 10 seconds, 480p or 1080p. Iterate at 480p, then re-run the prompt you like at 1080p. Cheapest way to work.
Roughly 1–3 minutes. Audio comes back with the clip, already timed.
文生视频、图生视频,多档分辨率与多个模型可选。
Sound is produced with the video in a single pass and arrives already synced.
Draft at 480p, deliver at 1080p. Roughly a 3× cost difference between them.
The native length of short-form video. Join several clips for anything longer.
Landscape for YouTube, vertical for Shorts and Reels, square for feeds.
Your photo stays recognisable — the model adds motion rather than reinventing the scene.
Run it yourself in ComfyUI whenever you want. No lock-in, unlike closed models.
What Wan can do, what it costs, and what "free" actually means.