Vidu Q4 API: 4K AI Video with Native Audio

What is Vidu Q4?

Vidu Q4 is ShengShu Technology's video model, released as Vidu Q4 Preview on October 7, 2026. It renders 540p to 4K video at 24 fps with an audio track generated in the same pass. Two input modes are available: image-to-video animates a single image for 3 to 16 seconds, and reference-to-video builds a new scene from up to 15 reference images plus up to 3 audio references, such as a voice the character should speak with. There is no text-to-video mode; every request starts from at least one image.

Key features of Vidu Q4

Up to 16 seconds of multi-shot video

ShengShu demo video, 15 seconds at 1920x1080: one generation cuts from a character on a cliff to airships drifting through clouds and a wide view of a floating city, with sound throughout. Shot changes and camera moves happen inside a single request, so a short scene comes back finished instead of as separate pieces to edit together.

Fast action and effects

ShengShu demo video, 15 seconds at 1920x1080: a soldier sprints through a ruined city while debris, fire and smoke fill the frame, with a close-up of her face between the running shots. Her body and face keep their shape through the fast running shots.

Multi-image reference with dialogue

House test at 1920x1080, 6 seconds, 16:9, two reference images: a woman in a denim jacket as [Image1] and a red apple as [Image2]. Prompt: [Image1] stands in a bright kitchen, holds up the apple from [Image2], smiles and says a line. She keeps her face, hair and jacket from the first photo, the apple keeps its exact striped skin from the second, and the line comes back in the audio track.

Spoken lines from a single image

House test at 1680x1234, 5 seconds, image-to-video from a portrait of a woman in a hat. Prompt: she looks into the camera, smiles and says a short line while wind moves her hair. Her face, hat and makeup stay exactly as in the photo, the output keeps the photo shape, and the spoken line and ambient sound arrive in the same file. Motion stays small and close to the source pose.

Camera moves through a referenced scene

House test at 1920x1080, 8 seconds, 16:9, one reference image of an alpine ridge. Prompt: a cinematic drone flight over the ridge from [Image1], gliding forward and rising above the peaks. The camera move is smooth and the ridge keeps the shape of the reference photo. The golden-hour light asked for in the prompt did not appear; the scene stayed in daylight.

Voice references

Pass up to 3 audio files of 1 to 16 seconds each and point to them as [Audio1] to [Audio3]. Use them to give a character a consistent voice across videos, or to set the sound a scene should carry.

540p to 4K in both modes

Every resolution from 540p to 4K works in image-to-video and in reference-to-video. Draft at 540p while the prompt settles, then render the final version at 1080p, 2K or 4K with the same request.

What Vidu Q4 is best for

Product videos with a presenter
Send the presenter and the product as two reference images and write the spoken line into the prompt.
Character-consistent series
Reuse the same reference images and voice reference to keep one character across a whole campaign.
Animating stills
Image-to-video turns one photo or illustration into a 3 to 16 second video that keeps its framing and shape.
4K deliverables
Render the final version at 2K or 4K without a separate upscaling step.
Fast drafts
540p renders the same request for a fraction of the 4K cost while you tune the prompt.

Vidu Q4 use cases

Build a talking-presenter ad: send a photo of the host and a photo of the product, then write the line the host should say. Keep a mascot consistent across a month of social posts by reusing the same character sheet and voice sample as references. Turn product photos or concept art into short animated videos with image-to-video, which keeps each image's framing. Render vertical 9:16 videos for social feeds and 16:9 versions for the web from the same references by changing aspect_ratio.

Vidu Q4 limitations

Vidu Q4 has no text-to-video mode and no start-and-end frame control: every request needs at least one image. Image-to-video needs at least 3 seconds, so 1 and 2 second videos come from reference-to-video only. Videos cap at 16 seconds, and reference inputs are images and audio only, with no video references or video editing. The model is a preview release, so its behavior may change before the final Q4 version.

Vidu Q4 API examples

Call Vidu Q4 from any language by POSTing to /v1/tasks. Full parameter docs live at docs.unifically.com/models/video/vidu/q4.

curl -X POST https://api.unifically.com/v1/tasks \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -d '{
    "model": "vidu/q4",
    "input": {
      "prompt": "[Image1] stands in a bright kitchen, picks up the apple from [Image2] and says: \"Fresh from the market.\"",
      "reference_image_urls": [
        "https://example.com/presenter.jpg",
        "https://example.com/apple.jpg"
      ],
      "resolution": "1080p",
      "duration": 6,
      "aspect_ratio": "16:9"
    }
  }'

Successful submission returns a task_id. Poll GET /v1/tasks/<task_id> or set a callback_url on the request to receive the finished result.

FAQs

Vidu Q4 API