Vidu Video Extension

Video models run out of runway fast, and Vidu's answer is a dedicated extension endpoint: feed it an existing clip and it generates additional frames onto the end, so you can build a scene longer than any single generation allows. This node is that feature - viduq2-pro or viduq2-turbo, source video in, longer video out.

The mental model is worth getting right before you use it: you're not re-rendering the clip, you're continuing it. The node takes your source video (4 to 55 seconds), generates a new segment on the end, and hands back the extended result. It's the Vidu-native way to chain shots into something that holds a story beat, versus cutting between disconnected generations and hoping the eye doesn't catch it.

The inputs

• video - the source clip to extend, 4–55 seconds. This is the "start here" anchor, and the quality of the handoff depends on it.
• model - a dynamic combo: pick viduq2-pro or viduq2-turbo, and the model's own settings expand below it - duration (1–7 seconds of new footage) and resolution (720p/1080p). The extension length is a per-model choice, not a global slider.
• prompt - optional, max 2000 chars. Describe what happens in the new footage. For a clean extension you want to continue the action, not introduce a scene change the model has to invent.
• end_frame - the optional steering input. Provide an image and the extension will transition toward it. This is the difference between "keep going" and "keep going and finish looking like this" - the same idea as the last-frame input on continuation nodes, and the best tool you have for controlling how the extended clip resolves.
• seed - 1 for reproducible, 0 for random.

How it runs

ComfyUI uploads your video (and the end frame, if set) to Vidu's extend endpoint through the Comfy proxy, polls the task, and downloads the result. Needs a Comfy account with credits and internet; no local GPU, no model download. Output is a VIDEO object - save it downstream.

The honest notes

Two things to know going in. First, the handoff quality is where this node lives or dies: if your source clip ends mid-motion or on a weird frame, the extension will inherit that weirdness. Trim the source so it ends on a clean beat before you extend it - that's the whole craft of this node. Second, pricing is per-second on both the base and the extension, and the node's price badge reflects it. Building a 30-second scene by extending in 7-second chunks is five paid generations, so plan the scene in an editor before spending.

The standing Vidu caveat applies too: this is the same API as Vidu's website, same content rules. The reason to extend here instead of there is that the clip never leaves your graph - it comes back as a VIDEO object ready for the next node, whether that's an upscaler, a color pass, or the next extension.