Vidu Q4 Preview on Azurade: animate one image or build a clip from 15 references

Shengshu released Vidu Q4 Preview on October 7, 2026, and both versions are on Azurade now. Vidu Q4 Preview animates one image, Vidu Q4 Preview Reference makes a clip out of up to 15 reference images and up to 3 audio clips. Clips run 3 to 16 seconds, from 540p up to 4K, with sound unless you switch audio off. We tried both on our mascot, the teal Azurade dragon, and we kept the first try of every clip.

Image to video#

You give it one picture (JPG, PNG or WebP, up to 50 MB) and the video keeps its shape, wide or tall. The prompt is optional here

We started with the teal dragon from our login page, the one breathing out photos and video frames. Our prompt was "The dragon slowly turns its head toward the camera and breathes a stronger stream of glowing photos and video frames that drift past the lens, its whiskers and mane moving in the wind, slow push-in, deep rumbling breath and soft whoosh". Around the middle of the clip it turns to face the camera and the push-in ends right on its face. One thing we didn't ask for: the photos change into new pictures while they fly (a man on a beach, a woman's portrait) which kind of fits a dragon that breathes out pictures

Vidu Q4 Preview, 720p, 5 seconds, sound on, the prompt above: 38 credits (mp4)

So what does the model do when you don't tell it anything? We gave it the ink version of the same dragon and left the prompt empty. It kept the dragon almost still and moved the ink instead: the cloud of little framed pictures in front of its mouth spreads and gets darker, and a drop of ink runs down the paper. For a still painting its a pretty good guess, and worth one run before you sit down and write a prompt. The video came out vertical because the painting is

Vidu Q4 Preview, 540p, 5 seconds, sound on, no prompt: 18 credits (mp4)

Reference to video#

Here the prompt is required. You add up to 15 images, and up to 3 MP3 clips for a voice or a sound, each 3 to 12 seconds long. Azurade checks that length before the generation starts so a clip that's too short or too long doesn't cost you credits. You pick the frame too: 16:9, 9:16, 1:1, 4:3 or 3:4. Neither version has a text-only mode, both need at least one image

Shengshu aims Q4 at short dramas, ads and social content, where one character has to look the same in every clip, and that's the job of the references. Our test had two: the Azurade logo (a dragon that's half brush strokes, half pixels) and a photo of an empty coffee shop on a rainy street that we made with Nano Banana 2.1 for 2 credits. The prompt: "A small teal dragon like the one in image 1 lands on the wooden counter of the coffee shop in image 2, sniffs the espresso machine and puffs out a little cloud of steam, rain running down the window behind it, handheld camera, quiet cafe sounds and a small snort".

The dragon flies in from the top left, lands on the counter, walks up to the expresso machine and puffs out a small cloud of steam. It's the same coffee shop as in the photo, down to the lamps and the cars outside. We liked that it kept the logo's style too, the body is made of the same square pixels as the logo, so you can tell it's our mascot straight away

Vidu Q4 Preview Reference, 16:9, 540p, 5 seconds, sound on, 2 reference images, no audio clip: 18 credits (mp4)

What it costs#

Both versions are billed per second of video, at the same rate:

The defaults are 720p and 5 seconds, which comes to 38 credits, and a 16-second 4K clip is 499.2. The three clips above cost 74 credits together, plus 2 for the coffee shop photo. A failed generation is refunded automatically and Azurade credits don't expire, so you can try an idea at 540p and only pay for 4K once you like it.

In the Azurade API and our MCP server the model ids are vidu-q4 and vidu-q4-reference.

Both versions are in the model list in Video Generation, try them on your own pictures!