Introduction

Vidu Q4 Preview is out. ShengShu's Vidu team added the new model to its API changelog on October 7, 2026, and the same day ComfyUI merged two partner nodes for it. That means you can try it three ways today: in the Vidu web app, through the Vidu Open Platform API as viduq4-preview, or inside ComfyUI with the new Vidu Q4 Image-to-Video and Vidu Q4 Reference-to-Video nodes.

It's a closed, API-only video model with native audio. The headline jump over Q3 is resolution: Q4 goes up to 4K, with 3 to 16 seconds per clip, up to 15 reference images, and up to 3 voice references in one generation.

What shipped

Vidu's changelog lists the Q4 Preview upgrades as more authentic performances with reference voices, smoother high-motion action (martial arts, running, fights), richer effects like explosions, fireworks, particles and energy blasts, stronger lighting and camera language, and 2K/4K output.

Two modes are live, and both share the same knobs:

  • Image-to-Video: one start frame plus an optional prompt. The output keeps your image's aspect ratio (anything between 1:5 and 5:1).
  • Reference-to-Video: 1 to 15 reference images for characters, scenes and style, plus 0 to 3 voice clips (3 to 12 seconds each, MP3 on the API). Output is 16:9, 9:16, 1:1, 3:4 or 4:3.
  • Resolution: 540p, 720p, 1080p, 2K or 4K. Default is 720p.
  • Duration: 3 to 16 seconds. Default is 5.
  • Audio: on by default, with dialogue and sound effects generated in the same pass. Vidu's docs also mention automatic camera switching for multi-shot scenes.
  • Prompt length: up to 20,000 characters on the API, 5,000 in the ComfyUI nodes.

What it costs

On the Vidu Open Platform, Q4 Preview bills per second by resolution:

  • 540p: 9 credits per second
  • 720p: 19 credits per second
  • 1080p: 24 credits per second
  • 2K: 38 credits per second
  • 4K: 78 credits per second

One credit is RMB 0.03125, so a 4K second works out to roughly RMB 2.44. For reference, 1080p Q4 costs the same 24 credits per second as Q3 Pro at 1080p, so you're paying extra only when you step up to 2K or 4K.

The ComfyUI partner nodes show their own USD price badge, which runs $0.045/s at 540p, $0.095/s at 720p, $0.12/s at 1080p, $0.19/s at 2K and $0.39/s at 4K. A default 5-second 720p test is about $0.48, and a full 16-second 4K shot is about $6.24, so draft small and upscale your keepers.

Run it in ComfyUI

The nodes landed in ComfyUI pull request #16849, merged on October 7 after the v0.39.1 tag. Until the next tagged release, you'll need a current master or nightly build to see them. Look under partner/video/Vidu for Vidu4ImageToVideoNode and Vidu4ReferenceVideoNode. Like every partner node, they run on Comfy's API credits, so sign in to your Comfy account first.

In the reference node, every image in a batch counts toward the 15-image cap, and you refer to them by order in the prompt. Voice clips only donate the voice, not the words. Write the dialogue in the prompt and assign each line to a voice by number. Reference audio also requires the audio toggle to stay on.

Prompt to try

Load a character portrait as image 1, a second character as image 2, a location photo as image 3, and a short voice clip for each character as audio 1 and audio 2. Then paste this into the Reference-to-Video node:

Rain-slick rooftop at night, the city from image 3 glowing below. Image 1 and image 2 face each other in a tense standoff, coats whipping in the wind. Image 1 lunges with a fast spinning kick; image 2 blocks and slides back through a puddle, sending up a spray of water. Sparks burst from a broken neon sign behind them. The camera cuts to a low-angle close-up of image 1, who says "You were always one step behind." in the voice from audio 1. Cut to image 2 smiling, who replies "Not tonight." in the voice from audio 2. Thunder rumbles, cinematic lighting, smooth handheld camera.

On the raw API, Vidu's reference docs use @1, @2 tags for subjects instead of "image 1", so swap those in if you're calling viduq4-preview directly.

Original Source

Conclusion

Vidu Q4 Preview is a meaty one for storytellers: 4K, a full 16 seconds, a stack of 15 references and real voice matching, all reachable from ComfyUI on day one. It's still labeled a preview, so expect the specs and prices to move. Start at 720p, lock your cast and your voices, and only pay for 4K once a shot earns it.

—Aurelia ♡