Vidu Q3 Image-to-Video Generation

Image-to-video on Vidu's flagship Q3 model is the combination that earned the family its reputation: you hand it a still of a character and it animates them with the facial expression work that local open models can't match. This node is that, with a couple of extras the rest of the Vidu menu doesn't have.

The viduq3-pro model offers 2K as a resolution option - the only node in the Vidu family with anything above 1080p. The turbo model tops out at 1080p, which is the standard tell for how the tiers work: pro gets the delivery-grade output, turbo gets the iteration speed. The price badge makes the 2K gap visible before you queue, and it's not small. Plan your spends accordingly.

What you set

• image - the start frame. Aspect ratio must sit between 1:4 and 4:1, which is loose enough to cover everything you'll actually feed it. As ever in image-to-video, the still is the artwork; the prompt only describes the motion on top of it.
• prompt - optional, max 2000 chars. Describe the animation: "she turns to the camera and smiles slowly," not "a woman in a red dress." The dress is already in the frame.
• duration - 1 to 16 seconds, the longest range in the family.
• audio - Q3's signature toggle. On, and the clip carries dialogue and sound effects. This is where the "it's a real scene, not a GIF" feeling comes from.
• seed - defaults to 1 for a reproducible take; 0 for random.

The resolution, duration, and audio options all live inside the model dropdown - pick viduq3-pro and they expand beneath it. Newcomers consistently hunt for a separate duration field that doesn't exist; it's nested, not missing.

How it runs

ComfyUI uploads your image to Vidu's image-to-video endpoint through the Comfy proxy, polls the task, and downloads the finished video. Needs a Comfy account with credits and an internet connection - no GPU, no local model. Output is a VIDEO object; wire it into a save node.

Notes from the field

The community's use pattern for Q3 i2v is character-driven and cost-aware: build the still locally (any image generator), iterate the animation cheap on turbo at 720p, then commit to pro - and 2K only for the take that's going on the reel. The famous Q3 expression quality shows up most at longer durations, because that's where the subtle micro-movements have room to breathe; a 2-second clip barely lets the model show off. And the standing Vidu caveat applies: ComfyUI is the delivery pipeline, not a moderation loophole - the content rules are identical to Vidu's website. If your goal is unmoderated video, that's a local-model conversation, not this node.