> ## Documentation Index
> Fetch the complete documentation index at: https://docs.qingbo.dev/llms.txt
> Use this file to discover all available pages before exploring further.
# Omni-Flash-Ext
> The fixed-duration Omni video model: text-to-video, first-frame or three-image references, and reference-video rework
`omni-flash-ext` is the Omni model whose length you control: `duration` is `4`, `6`, `8` or `10` seconds, output runs from 360p to 4K, and the input can be a prompt alone, one first-frame image, three reference images, or a reference clip. When the model should pick the length itself, or you want to keep editing the same clip over several rounds, use [Gemini Omni Flash](/en/api-reference/video/omni/gemini-omni-flash).
## Quick start
```bash Text to video theme={"system"}
curl -X POST https://www.qingbo.dev/v1/tasks \
-H "Authorization: Bearer $WAVE_API_KEY" \
-H "Content-Type: application/json" \
-H "Idempotency-Key: omni-ext-001" \
-d '{
"model": "omni-flash-ext",
"action": "generate",
"prompt": "sunset over the sea, gulls skimming the water, cinematic light",
"duration": 6,
"resolution": "720p",
"aspect_ratio": "16:9"
}'
```
```bash First-frame image theme={"system"}
curl -X POST https://www.qingbo.dev/v1/tasks \
-H "Authorization: Bearer $WAVE_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "omni-flash-ext",
"action": "generate",
"prompt": "the character smiles and slowly turns around, gentle push-in",
"generation_type": "frame",
"image_urls": ["https://cdn.example.com/character.jpg"],
"duration": 6,
"resolution": "1080p",
"aspect_ratio": "9:16"
}'
```
```bash Three reference images theme={"system"}
curl -X POST https://www.qingbo.dev/v1/tasks \
-H "Authorization: Bearer $WAVE_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "omni-flash-ext",
"action": "generate",
"prompt": "combine the scene, the character and the product into one continuous shot",
"generation_type": "reference",
"image_urls": [
"https://cdn.example.com/scene.jpg",
"https://cdn.example.com/character.jpg",
"https://cdn.example.com/product.jpg"
],
"duration": 10,
"resolution": "1080p",
"aspect_ratio": "9:16"
}'
```
```bash Reference video theme={"system"}
curl -X POST https://www.qingbo.dev/v1/tasks \
-H "Authorization: Bearer $WAVE_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "omni-flash-ext",
"action": "generate",
"prompt": "the same scene, shot at night under neon lights",
"resolution": "720p",
"aspect_ratio": "16:9",
"video_urls": ["https://cdn.example.com/reference.mp4"]
}'
```
A successful submission returns a `task_id`. Retrieve the result with [`GET /v1/tasks/{task_id}`](/en/api-reference/task/status), wait inline with [`Prefer: wait`](/en/api-reference/task/submit#in-request-waiting-prefer-wait), or configure a webhook.
## Request parameters
Must be `omni-flash-ext`.
Value `generate`; it may be omitted. The mode is not selected with `action` but by what you send: `prompt` alone is text-to-video, `image_urls` makes it image-to-video, and `video_urls` reworks an existing clip.
The video description. The more specific the scene, subject, action, setting, camera movement and visual style, the steadier the result.
Clip length in seconds: `4`, `6`, `8` or `10`.
Do not send this field together with `video_urls` — the reference video determines the length.
Output resolution: `360p`, `720p`, `1080p` or `4k`, case-insensitive.
Frame ratio: `16:9` landscape or `9:16` portrait. With `video_urls`, the output usually follows the reference video's ratio.
Reference image URLs, publicly accessible. Send none, one, or three; how the count maps to `generation_type` is under "Limits".
It can be combined with `video_urls`: the images act as identity or composition references while the video acts as the motion reference.
Reference video URLs, publicly downloadable direct links. At most one clip, up to 10 seconds. Do not send `duration` alongside it.
How images are used: `frame` (exactly 1) or `reference` (1 or 3).
See [Submit Task](/en/api-reference/task/submit) for `callback_url`, `callback_events`, `Prefer: wait`, `Idempotency-Key`, and the maximum-cost header.
This model does not support `seed`, `first_frame_image`, `last_frame_image` or `ref_task_id`; a first frame is expressed as `generation_type: "frame"` with one image in `image_urls`. Any other parameter returns `400` and is not billed.
## Limits
* One video per task; 0, 1 or 3 images (2 images are rejected upstream).
* `generation_type=frame` takes exactly one image as the first frame.
* `generation_type=reference` takes 1 or 3 reference images.
* With a reference video duration cannot be sent (omit it); at most one video of ≤ 10 s, billed per second of the reference video.
| Item | Limit |
| ---------------- | -------------------------------- |
| Reference images | Up to 3 |
| Reference video | At most 1, up to 10 seconds |
| Duration | `4` / `6` / `8` / `10` seconds |
| Resolution | `360p` / `720p` / `1080p` / `4k` |
| Aspect ratio | `16:9` / `9:16` |
## Pricing
**Without a reference video**, billing is a **whole-clip price by resolution × duration**: `resolution` and `duration` are the only two billing dimensions, every combination of the four resolutions and four durations has its own price, and the cost of a call is known before you submit it. Reference images cost nothing extra.
**With a reference video** (`video_urls`) the shape changes: billing runs **per second of the reference video** and the whole-clip price is not charged. Each resolution tier has its own per-second rate, and a reference video may be at most 10 seconds. How reference-video seconds are rounded is covered in [Video Generation Overview](/en/api-reference/video/overview#input-video-duration-and-billing).
Submitting a task places a hold; the task then settles on the actual generated result and the difference is released.
You are only billed for a successfully generated video. Failed and cancelled tasks, and tasks that return no usable video, are refunded in full. What a call actually cost is the `cost` field on the task response — an integer quota at **500,000 quota = 1 USD**, not dollars.
Per-model rates are in `price_config` on `GET /v1/models` and in the console's Model Market.
This section describes the billing dimensions only; it carries no amounts.
## Response
```json Completed task theme={"system"}
{
"task_id": "taski-f4df1815445069da8995fa77f7cfc947-task-wave1788784802b950097545",
"model": "omni-flash-ext",
"action": "generate",
"status": "completed",
"progress": "100%",
"created_at": 1788784802,
"completed_at": 1788784868,
"result": {
"videos": [
{
"expires_at": 1789089508,
"url": ["https://cdn.example.com/result.mp4"]
}
]
},
"billing_status": "settled",
"cost": 67500,
"urls": {
"get": "https://www.qingbo.dev/v1/tasks/{task_id}",
"cancel": "https://www.qingbo.dev/v1/tasks/{task_id}/cancel"
}
}
```
The generated video is returned in `result.videos`, where `url` is an array and `expires_at` is the link's expiry as a Unix timestamp in seconds. See [Query Task Status](/en/api-reference/task/status) for every field.
## Related
* [Omni Video Series](/en/api-reference/video/omni/overview)
* [Gemini Omni Flash](/en/api-reference/video/omni/gemini-omni-flash)
* [Video Generation Overview](/en/api-reference/video/overview)
* [Submit Task](/en/api-reference/task/submit)
* [Query Task Status](/en/api-reference/task/status)
* [Task System](/en/docs/task-system)