Vidu Multi-Frame Video Generation

This is the most ambitious node in the Vidu family, and the one that most clearly shows what Vidu is for. Instead of a single start frame, you give it a start_image plus a sequence of up to nine keyframe transitions - each with its own end image, its own prompt, and its own duration - and Vidu renders the whole chain as one video, moving through your storyboard frame by frame. It's "produce a five-beat scene in a single generation" rather than "make five clips and pray."

How the keyframe chain works

The frames dropdown is the whole game. Pick a number from 2 to 9 and the node expands that many segments, each with three inputs:

• prompt{i} - what happens in this segment.
• end_image{i} - the keyframe this segment resolves to (aspect ratio between 1:4 and 4:1).
• duration{i} - 2 to 7 seconds for this segment.

The chain runs: start_image → (segment 1 → end_image1) → (segment 2 → end_image2) → … Each segment is a mini first-last-frame generation, and Vidu stitches them into one continuous video. Total length is the sum of your segment durations, so a 4-segment chain at 4 seconds each is a 16-second take. The model choice is viduq2-pro or viduq2-turbo, and resolution is 720p/1080p.

When this node is the right tool

Multi-frame shines when you have a story arc with a clear shape - "establish the room, walk to the window, look out, turn and react." You can draw that as four end frames in minutes and let Vidu animate the space between. It's also the only Vidu node where a narrative lives in the inputs, not just a scene. But it's demanding: every segment multiplies the chance of a warpy transition, and the model is only as coherent as the sequence you hand it. Each end frame should be recognizably part of the same world - same characters, same lighting, consistent composition - or the seams show.

How it runs and what it costs

ComfyUI uploads the start image and every keyframe to Vidu's multi-frame endpoint through the Comfy proxy, polls the task (longer than a single generation - the node's poll budget scales with frame count), and downloads the result. Needs a Comfy account with credits and internet; no local GPU. Output is a VIDEO object.

Pricing is the honest caveat: you're billed per segment plus per second, so this is the priciest node in the family, and the price badge is a small spreadsheet. The cost-aware path is to sketch the chain with 2–3 frames on turbo at 720p first, then commit to the full sequence on pro. And the family rule holds: same moderation as Vidu's site, with the benefit being that the finished multi-shot video lands inside your graph, ready for the next stage of the pipeline.