Vidu Image To Video Generation

Vidu's hosted image-to-video API, exposed as a first-party node in ComfyUI core, is about as stripped-down as this family gets: an image, an optional prompt, and a fixed 5-second clip out. It's a partner/video API node - no install, nothing runs on your GPU, and every render bills against your Comfy account credits through the Comfy proxy.

Let's be direct about the layout, because the input list looks padded but isn't. image is your start frame and the only required creative input besides the model. prompt is optional - feed it motion directions ("slow push in, leaves drifting past") or leave it blank and let Vidu improvise. model is currently viduq1 (Vidu Q1), and duration is where the joke lives: min 5, max 5, default 5. That's not a slider, that's a label. This node does exactly one duration, and the Vidu 2 nodes exist precisely because of that limitation.

The controls that actually matter are buried in the advanced section. movement_amplitude - auto, small, medium, large - is the knob that decides whether your clip is a gentle sway or an energetic scene. Start with auto, then go large when the video comes back too static; most Vidu disappointment is a motion-energy problem, not a prompt problem. resolution offers 1080p (the only current option in this build), and seed follows the family convention: 0 for random, otherwise a re-run control on a non-deterministic model.

Two practical constraints from the source, both the kind that eat an afternoon if you discover them the hard way. The node accepts exactly one input image - feed it a batch and it errors rather than picking a frame for you. And the image must have an aspect ratio between 1:4 and 4:1; anything more extreme is rejected. Normal photos and most renders are fine; a 1:10 banner is not.

Output is a single VIDEO, ready for the Save/Preview or downstream editing. The rhythm is the standard partner-node one: draft cheap, confirm the motion reads, then commit. The price badge sits on the node and updates with your settings - and since duration is fixed, the resolution is the only real cost lever.

Where it fits: quick "animate this still" jobs where Vidu's quality tier is what you want and you don't need bookending, camera control, or a longer cut. The moment you want a guaranteed ending frame, step up to the Vidu2 start/end node; the moment you need more than 5 seconds, the text-to-video or reference nodes in the same family. Vidu's video nodes have been in core since August 2025, and this one has stayed deliberately simple ever since. Your image leaves your machine, the vendor's filter runs on the other end, and the clip comes back as a VIDEO you can wire anywhere.