Vidu Request Parameters and Media

Understand Q2, Q3, and Q4 Preview durations, resolution tiers, image inputs, audio, and native extension parameters.

This page covers common parameters for the enabled models in the overview. Use the exact lowercase model name in model. Parameter support varies by model and operation.

Duration, resolution, and sound

Model	Integer `duration` in seconds	`resolution` used in this series
`viduq2`, `viduq2-pro`, `viduq2-turbo`	1–10; default 5	`540p`, `720p`, `1080p`; default `720p`
`viduq2-pro-fast`	1–10; default 5	`720p`, `1080p`; default `720p`
`viduq3-pro`, `viduq3-turbo`	1–16; default 5	`540p`, `720p`, `1080p`; default `720p`
`viduq4-preview` (availability unconfirmed)	3–16; default 5	`540p`, `720p`, `1080p`, `2K`, `4K`; default `720p`

resolution selects a quality tier. Output dimensions also depend on aspect ratio and model alignment; 540p does not guarantee an exact 540-pixel height for every ratio.

Q3 and Q4 default audio to true; false produces silent video. Q2 text generation enables audio by default, while Q2 image and reference generation disable it by default. Set the value explicitly when needed. Sound parameters for first/last-frame generation depend on the model; do not assume image-to-video audio behavior applies. Audio and prompt expansion can affect pricing.

Select an image input mode

Operation	Path	Image meaning
Image to video	`/ent/v2/img2video`	One first-frame image in `images`
First/last frames	`/ent/v2/start-end2video`	Two images, ordered first frame then last frame
Reference to video	`/ent/v2/reference2video`	Reference images; limits and subject support vary by model

Q2 image references accept 1–7 images and require a nonempty prompt. This series covers image references only; video-reference and video-editing operations are outside its scope.

Use publicly downloadable image URLs rather than local paths or pages requiring login. Supported formats include PNG, JPEG/JPG, and WEBP. Complete image data URIs such as data:image/png;base64,... are also supported. Image dimensions, ratios, and body limits depend on the selected model and operation.

Q3 image to video

Replace the example image URL with a real asset and send this JSON to POST /ent/v2/img2video:

{
  "model": "viduq3-pro",
  "images": ["https://example.com/first-frame.png"],
  "prompt": "The camera slowly pulls back as the person in the reference image waves.",
  "duration": 5,
  "resolution": "720p",
  "audio": false,
  "is_rec": false,
  "seed": 0,
  "watermark": false
}

Q3 first/last frames

Send to POST /ent/v2/start-end2video. Image order determines the opening and final frames:

{
  "model": "viduq3-turbo",
  "images": ["https://example.com/first.png", "https://example.com/last.png"],
  "prompt": "A steady camera move transitions from a flower bud to a fully opened blossom.",
  "duration": 5,
  "resolution": "720p",
  "is_rec": false,
  "watermark": false
}

Q2 reference to video

Send viduq2 references to POST /ent/v2/reference2video. A single reference image does not make this an image-to-video request:

{
  "model": "viduq2",
  "images": ["https://example.com/teapot.png"],
  "prompt": "Keep the reference teapot's appearance and place it on a sunlit wooden table. The camera slowly orbits it.",
  "duration": 5,
  "resolution": "720p",
  "aspect_ratio": "16:9",
  "audio": false,
  "watermark": false
}

Q4 Preview image parameters

The following describes the official Q4 Preview image-to-video protocol. Production availability is unconfirmed; enabled Q2 / Q3 models do not establish Q4 availability.

images is required and accepts exactly one first-frame image, at most 50 MB. The HTTP body must be smaller than 20 MB. Prefer URLs because Base64 increases body size.
prompt is optional, with a 20,000-character limit. Official is_rec defaults to true; set false explicitly to disable expansion.
duration is an integer from 3 to 16. See the table above for resolution and audio.
An omitted or zero seed requests randomness; a nonzero value specifies the seed.
Set watermark explicitly rather than relying on a default. wm_position accepts 1 (top left), 2 (top right), 3 (bottom right, default), or 4 (bottom left).

Only after channel access is available, send this body to POST /ent/v2/img2video:

{
  "model": "viduq4-preview",
  "images": ["https://example.com/first-frame.png"],
  "prompt": "The person waves to the camera, then the camera slowly rises.",
  "duration": 5,
  "resolution": "720p",
  "audio": true,
  "is_rec": false,
  "seed": 0,
  "watermark": false
}

Q4 image-to-video does not accept a first/last-frame pair or provide text-only generation. An official reference-to-video protocol also exists; access depends on the model and channel.

Native extension fields

Field	Type and purpose
`aspect_ratio`	String; output ratio for supported text/reference operations, for example `16:9`
`is_rec`	Boolean; prompt expansion. For current Q3 access, explicitly use `false`
`off_peak`	Boolean; current channels allow omission or `false` only
`payload`	Custom pass-through string, at most 1,048,576 characters
`wm_url`	Custom watermark image URL, subject to upstream requirements
`meta_data`	JSON-encoded string for metadata labeling, not an object
`callback_url`	URL receiving upstream POST notifications; callbacks may contain upstream IDs
`sounds`, `subjects`, `auto_subjects`	Model-specific audio/subject fields; use only with supported operations

Native requests can include supported extension fields such as Q2 bgm; field availability is model-specific. Generic requests map only the listed fields. Use native requests for complete provider parameters.

Example meta_data:

{"meta_data": "{\"Label\":\"your-label\",\"ContentProducer\":\"your-company\"}"}

Upstream callbacks are auxiliary notifications. After a notification, confirm the result by querying with the saved public task ID and ModelSell API Key; use polling as the main workflow. This series does not provide a ModelSell callback receiver endpoint.