Vidu2 Start/End Frame-to-Video Generation

Vidu's first image-to-video node locked you to exactly 5 seconds, and it felt like it. This is Vidu 2's answer to that complaint: a start/end-frame node with a real duration slider - 2 to 8 seconds - plus a model dropdown that actually has choices. It's a partner/video API node built into ComfyUI core: no install, nothing runs on your GPU, and every render bills against your Comfy account credits.

The shape is the standard bookend pattern. first_frame and end_frame are the anchors, prompt describes what happens between them (capped at 2000 characters), and the node uploads both frames, generates, and returns a single VIDEO. model is where Vidu 2 shows its range: viduq2-pro-fast, viduq2-pro, and viduq2-turbo - three cost/speed tiers, and the pricing badge on the node reflects them differently. Pro-fast is the cheap everyday tier; pro is the quality push; turbo sits in its own pricing lane. Pick per job, not once.

duration runs 2–8 seconds with a default of 5, which is the flexibility the original Vidu I2V node denied you. resolution (720p or 1080p, tucked into advanced) and movement_amplitude (auto, small, medium, large) complete the set. The movement knob deserves the same respect it gets in the rest of the Vidu family - static-feeling results are usually an amplitude problem before they're a prompt problem. And seed defaults to 1 here rather than 0, a minor cosmetic difference from the other Vidu nodes that changes nothing about how you use it.

Two validation quirks from the source are worth knowing before they bite. Each frame input accepts exactly one image - a batch errors out rather than picking a frame. And the two frames' aspect ratios have to be close to each other (within roughly 0.8–1.25 relative), because Vidu 2 won't animate between wildly different canvas shapes; mismatch them and the node rejects the pair. Keep your start and end frames the same size and shape and you'll never see it.

The practical pitch is the same as every bookend node: pin the ending you actually want. Two storyboard stills become a shot with a guaranteed conclusion; a character's pose at the start and their pose at the end become a motion arc with identity held at both ends. Vidu 2's contribution is doing it with real duration control and tiered pricing, so you can decide how much quality you're paying for per render.

Standard partner-node rules: your frames leave your machine for Vidu's servers, vendor-side moderation can reject a generation, and the price badge - which varies by model, duration, and resolution - is your guide to drafting cheap before committing. The Vidu 2 nodes landed in core in January 2026. If you liked the Vidu look but hated the fixed 5 seconds, this is the upgrade you were waiting for.