GET

/v1/schemas/llmgateway/video/v1/videos

Accept:

kind inputactivity videohash 7ca8f5c5…extracted Nov 12, 2026version currentsource upstream spec ↗

field	recorded source
schema	Source linked above. upstream-spec Source trace `39b98bb53afb3bb9bc109695016432efb3f5c86c0cbf631948ff2ea70493a539` { "fetchedAt": 1794460248, "extractorVersion": "2", "verifiedAt": null }

Request body

17 properties · 2 required · 0 $defs

property	type	description	constraints
prompt*	string	Text prompt describing the video to generate.	length ≥ 1
seconds*	integer	Output duration in seconds. Supported values depend on the selected model and provider mapping.	n ≥ 1
model	string	The video generation model to use. Supports current Veo and Sora video models, including provider-prefixed variants like openai/sora-2 or google-vertex/veo-3.1-generate-preview.	default: "veo-3.1-generate-preview"
size	string	Output resolution in OpenAI widthxheight format. Supported values depend on the selected model and provider mapping.
callback_url	string	LLMGateway extension. When set, a signed webhook is delivered after the job reaches a terminal state.	format: uri
callback_secret	string	LLMGateway extension. Shared secret used to sign webhook deliveries with HMAC-SHA256.	length ≥ 1
input_reference	string \| object \| string \| object[]	Reference image input alias. Accepts HTTPS URLs or base64 data URLs and routes to provider-specific image-guided video generation when supported.
last_frame	string \| object	Optional ending frame for frame-to-video generation. Requires image and is routed to providers that support first/last-frame generation.
audio	boolean	Whether the generated video should include audio. Support depends on the selected model and provider mapping.	default: true
reasoning_effort	enum	FLUX 3 Max reasoning effort. Defaults to high.	low \| high
n	integer
image	string \| object	Input image for image-to-video generation. Supports HTTPS URLs and base64 data URLs. Routed to provider-specific image-to-video generation when supported by the selected model.
reference_images	string \| object[]	Reference images for provider-specific asset or material-guided video generation. ByteDance Seedance 2.x models accept up to 9 and Alibaba Wan 3.0 up to 10; other providers accept up to 3.	items ≥ 1 · items ≤ 10
reference_videos	string \| object[]	One to three reference videos (HTTPS URLs) for omni-reference video generation. Currently supported on ByteDance Seedance 2.x and Gemini Omni models and can be combined with reference_images.	items ≥ 1 · items ≤ 3
video	string \| object	LLMGateway extension. Source video (up to 10 seconds) to edit or extend, see video_task. Must be a publicly reachable HTTPS URL. Currently only supported on google-ai-studio Gemini Omni models.
video_task	enum	LLMGateway extension. What to do with `video`: `edit` (default) changes the source and keeps its length, ignoring `seconds`; `extend` appends a `seconds`-long continuation.	edit \| extend
reference_audios	string \| object[]	One to three reference audio clips (HTTPS URLs) for omni-reference video generation. Supported on ByteDance Seedance 2.x models, where they can be combined with reference_images and reference_videos, and on VEED Fabric models, which take exactly one clip as th…	items ≥ 1 · items ≤ 3

Validate a payload

before you spend tokens

shell

$ curl -X POST https://modelschemas.com/v1/validate \
    -d '{"provider":"llmgateway","endpointId":"v1/videos","payload":{…}}'
→ {"valid": false, "errors": [{"path": "#", "keyword": "required", …}]}