- client
- last_frame_image
- url
- task_id
Every video model in this pack has a hard ceiling on clip length - Wan is stuck at 5 seconds, even Vidu caps out at 16. So when your scene needs to run longer, you have two options: regenerate and pray, or extend what you already have. Vidu Extend is the second option. You feed it the URL of an existing video (4 seconds to 1 minute), tell it how many more seconds to add, and it continues the clip from where it stopped.
That's it - that's the whole job, and it's genuinely useful. It's how you turn a 5-second Wan shot into a 12-second scene, or how you keep a character talking past the model's built-in limit. It's also the node you reach for when a clip is almost right and just needs more breathing room at the end.
The inputs that matter
client- standard wiring from the Modelverse Client or Secret Client.model-viduq2-turbo(default, faster) orviduq2-pro(better quality). Only the q2 line supports extend, per the tooltips.video_url- required - the clip to extend, 4 seconds to 1 minute. This is a URL string, so the source video has to be reachable; a URL from another Modelverse video node works perfectly.duration- how much to add, 1–7 seconds. Not total length - added length.resolution-540p,720p(default),1080p.- Optional
prompt- text steering what the extension contains. Handy if you want the added seconds to do something specific rather than just continue. - Optional
last_frame_image/last_frame_url- a target end frame for the extension (one or the other, never both). This is the "extend until it reaches this image" mode - useful for looping or matching to a specific endpoint.
Outputs are url and task_id; the new url is the lengthened video, and you'd wire it into Modelverse Preview Video like any other video result.
How it works
The node submits the source URL with vidu_type: extend, polls the task every 5 seconds, and returns the extended clip. One caveat from the source: because the input is a URL, there's no image-tensor handling here - you can't wire an IMAGE in as the source. So if your clip lives on a web host or as a Modelverse result URL, use that string directly rather than trying to pass a tensor.
Install
Pack install: ComfyUI Manager → search "ComfyUl-UCloud" → Install → restart, or git clone https://github.com/ucloud/comfyui-plugin into custom_nodes. Light deps, no model downloads, and an API key from the Modelverse console.
The honest take: Extend is a utility, not a showpiece - you won't reach for it first, but it's the node that saves a whole regeneration cycle when a clip just needs more seconds. And since each extend run is metered, adding 7 seconds to a video costs roughly one generation's worth of credits. Plan your extensions, don't chain ten of them and eat your balance.
Inputs (8)
| Name | Type | Default | Description |
|---|---|---|---|
| client | MODELVERSE_API_CLIENT | — | |
| model | COMBO | viduq2-turbo | viduq2-pro: 效果好, viduq2-turbo: 生成快 |
| video_url | STRING | 需要延长的视频URL (4秒-1分钟) | |
| duration | INT | 51–7 | 延长时长(秒), 1-7 |
| resolution | COMBO | 720p | 分辨率 |
| last_frame_imageopt | IMAGE | 延长到尾帧的参考图像 | |
| last_frame_urlopt | STRING | 延长到尾帧的参考图像URL | |
| promptopt | STRING | 延长提示词,控制延长内容 |
Outputs (2)
| Name | Type | Description |
|---|---|---|
| url | STRING | — |
| task_id | STRING | — |