Other video models

Parameters, sizes, aspect ratios, and input/output reference for Hailuo 3, Grok Imagine, Kling, Veo, Gemini Omni, and Motion.

Every video model other than the Seedance family is served by the le-video provider and shares one asynchronous task API: submit and receive a task ID, poll for status, download the MP4 when it completes.

Each model accepts only its own fixed enumeration of sizes and durations, and no two are alike. Any value outside the table returns 400 and is never forwarded upstream. The tables below are the complete enumerations — there is no "derivable aspect ratio" rule here.

The Seedance family (dreamina-seedance-*) is a separate group — see Dreamina video generation.

Model overview

These models are billed per second of video (meter video_seconds, unit: second).

Model ID	Name	Input	Audio	Duration (s)
`hailuo-3`	MiniMax Hailuo 3	Text-to-video / image-to-video	Always on, cannot be disabled	5–15 (default 5)
`grok-imagine-1.5`	Grok Imagine 1.5	Image-to-video only	Toggleable (default on)	3–15 (default 6)
`kling-3.0-turbo`	Kling 3.0 Turbo	Text-to-video / image-to-video	Toggleable (default on)	3–15 (default 5)
`veo-3.1-generate-001`	Veo 3.1	Text-to-video / image-to-video	Toggleable (default on)	4 / 6 / 8 (default 8)
`gemini-omni-flash`	Gemini Omni Flash	Text-to-video / image-to-video	Not supported	3–10 (default 8)
`motion_2.0-fast`	Motion 2.0 Fast	Text-to-video / image-to-video	Not supported	Fixed 5

hailuo-03 is an equivalent spelling of hailuo-3; both IDs work and are identical in capability and billing. Treat the unit price returned by the console or the model discovery endpoint as authoritative. Models your organization has not enabled return 403.

Sizes and aspect ratios

size takes a "WIDTHxHEIGHT" string. The tables are the complete set of legal values.

`hailuo-3` / `hailuo-03`

2K only; 720p and below are rejected.

size	Aspect ratio
`3360x1440`	21:9
`2560x1440`	16:9
`1920x1440`	4:3
`1440x1440`	1:1 (default)
`1440x1920`	3:4
`1440x2560`	9:16

`grok-imagine-1.5`

size	Aspect ratio
`1280x720`	16:9 (default)
`720x1280`	9:16
`960x960`	1:1
`544x544`	1:1 (small)
`736x400`	16:9 (small)
`400x736`	9:16 (small)

`kling-3.0-turbo`

size	Aspect ratio	Tier
`1280x720`	16:9	720p (default)
`960x960`	1:1	720p
`720x1280`	9:16	720p
`1920x1080`	16:9	1080p
`1440x1440`	1:1	1080p
`1080x1920`	9:16	1080p

`veo-3.1-generate-001`

size	Aspect ratio	Tier
`1280x720`	16:9	720p (default)
`720x1280`	9:16	720p
`1920x1080`	16:9	1080p
`1080x1920`	9:16	1080p
`3840x2160`	16:9	4K
`2160x3840`	9:16	4K

`gemini-omni-flash`

size	Aspect ratio
`1280x720`	16:9 (default)
`720x1280`	9:16

`motion_2.0-fast`

size	Aspect ratio
`1280x720`	16:9 (only legal value)

Input and output

Model	Text-to-video	Image-to-video	Reference image limit	Output
`hailuo-3`	Yes	Yes	4	MP4 with audio
`grok-imagine-1.5`	No	Required	1 (first frame)	MP4
`kling-3.0-turbo`	Yes	Yes	1 (first frame)	MP4
`veo-3.1-generate-001`	Yes	Yes	3	MP4
`gemini-omni-flash`	Yes	Yes	4	MP4 (no audio)
`motion_2.0-fast`	Yes	Yes	1 (first frame)	MP4 (no audio)

grok-imagine-1.5 has no text-to-video path upstream. Submitting without input_reference returns 400 invalid_reference and is never forwarded upstream.

Endpoints

Operation	Endpoint
Create task	`POST /v1/videos`
Get task	`GET /v1/videos/{task_id}`
Download result	`GET /v1/videos/{task_id}/content`

The key must carry the ai:video scope:

Authorization: Bearer $TOS_API_KEY

Request parameters

Field	Type	Required	Description
`model`	string	Yes	A model ID from the table above
`prompt`	string	Yes	An empty prompt returns 400
`size`	string	No	`"WIDTHxHEIGHT"`. A value outside the model's enumeration snaps to the legal size with the closest aspect ratio; omitting it uses the model's default
`duration`	number	No	Seconds. A value outside the enumeration snaps to the nearest legal value, ties rounding up; omitting it uses the default. Aliases: `duration_seconds`, `seconds`
`audio`	boolean	No	Only meaningful for models that support audio. `hailuo-3` always produces audio regardless
`input_reference`	string or string[]	Required for `grok-imagine-1.5`	Publicly reachable `http(s)` image URL. Exceeding the model's reference limit returns 400

Billing follows the duration actually rendered, i.e. the value after snapping. Asking a model that only accepts 4/6/8/10 seconds for 5 seconds renders 6 seconds and is billed as 6.

Text-to-video

curl --noproxy '*' "https://ai.tos.run/v1/videos" \
  -H "Authorization: Bearer $TOS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "hailuo-3",
    "prompt": "A red apple rolling slowly across a white table, cinematic",
    "size": "2560x1440",
    "duration": 5
  }'

{
  "id": "78a2ab60-9a53-4d2c-99fb-b5f8ca8d4018",
  "task_id": "78a2ab60-9a53-4d2c-99fb-b5f8ca8d4018",
  "object": "video",
  "model": "hailuo-3",
  "status": "queued",
  "progress": 0,
  "created_at": 1786370088,
  "expires_at": 1786974888,
  "metadata": { "provider": "le-video" }
}

Image-to-video

grok-imagine-1.5 must use this path:

curl --noproxy '*' "https://ai.tos.run/v1/videos" \
  -H "Authorization: Bearer $TOS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "grok-imagine-1.5",
    "prompt": "Slow camera push-in with natural light",
    "size": "1280x720",
    "duration": 6,
    "input_reference": "https://example.com/first-frame.jpg"
  }'

The reference image must be publicly downloadable and stay reachable for the duration of the task — upstream fetches it while generating.

Polling and download

Field	Description
`id` / `task_id`	Gateway task ID, used for both polling and download
`status`	`queued` → `in_progress` → `completed` / `failed`
`progress`	0–100, display only; never use it to decide terminal state
`metadata.url`	Result URL, present once `status` is `completed`
`expires_at`	When the stored result expires

status is the only terminal-state signal.

curl --noproxy '*' "https://ai.tos.run/v1/videos/$TASK_ID" \
  -H "Authorization: Bearer $TOS_API_KEY"

curl --noproxy '*' "https://ai.tos.run/v1/videos/$TASK_ID/content" \
  -H "Authorization: Bearer $TOS_API_KEY" -o output.mp4

2K generation takes considerably longer than 720p: a 5-second 2K clip from hailuo-3 typically needs more than 7 minutes. Poll no more often than every 15 seconds.

Errors

HTTP	code	Meaning
400	`invalid_size`	Size is outside the model's enumeration. The error lists every legal size for that model
400	`invalid_duration`	Duration is outside the model's enumeration
400	`invalid_reference`	Missing a required reference image, or more references than the model allows
400	`model_not_available_for_routing`	The model ID is not in the video catalog
403	`model_not_authorized_for_org`	Your organization has not enabled this model
502 / 504	`upstream_error`	Upstream returned an error or timed out

On this page