ModelRover

Gemini Video

Create Gemini Omni video tasks and retrieve platform-hosted results.

Veo 3.1 is shut down: Google discontinued the veo-3.1-*-preview series on 2026-10-22. Migrate to google/gemini-omni-1.2-flash. gemini-omni-1.1-flash automatically routes to 1.2 from 2026-12-03 at the lower 1.2 rate.

POST /v1beta/interactions

POST /v1beta/interactions
x-goog-api-key: $API_KEY
Content-Type: application/json

Gemini Omni video tasks are created asynchronously through the Interactions endpoint. The upstream accepts the task immediately; the response id is rewritten to the platform task ID and polled via GET /v1/tasks/{taskId}.

curl "https://api.modelrover.ai/v1beta/interactions" \
-H "x-goog-api-key: $API_KEY" \
-H "Content-Type: application/json" \
-d '{
  "model": "google/gemini-omni-1.2-flash",
  "input": "A glass marble rolling at high speed on a chain-reaction track, continuous smooth single shot",
  "response_format": {
    "type": "video",
    "resolution": "1080p",
    "aspect_ratio": "16:9",
    "duration": "10s"
  },
  "callback_url": "https://example.com/hooks/video?token=unguessable"
}'

Request fields

FieldTypeDescription
modelstringComplete public Omni video model ID.
inputstring, object, or arrayText prompt, or an array of text and image/video items.
response_format.typestringFixed "video"; may be omitted.
response_format.resolutionstring360p, 720p (default), 1080p, or 4k.
response_format.aspect_ratiostring16:9 (default) or 9:16.
response_format.durationstring"3s" to "15s" (default "10s"); first-last-frame mode is capped at "10s".
generation_config.video_config.taskstringOptional; inferred from input when omitted (see modes below).
callback_urlstringOptional public terminal callback URL.

Video requests reject stream: true, store: true, background: true, and continuation or agent fields: previous_interaction_id, agent, environment, webhook_config.

Video modes

The mode is inferred from the input shape. generation_config.video_config.task is a fallback for disambiguation only.

Modetask valueInput requirements
Text to videotext_to_videoText only; no images or videos. Duration 3–15s.
Image to videoimage_to_videoOne image + text. Duration 3–15s.
First and last framefirst_last_frameTwo images (first + last frame) + text. Duration capped at 10s.
Reference to videoreference_to_video1–6 reference images (with optional text). Duration 3–15s; 11–15s clips allow at most 1 reference video.
Video editeditOne video + text describing the change. Duration 3–15s.
Video extendextendOne video + text describing the continuation. Duration 3–10s.

Reference images are passed as {"type": "image", "data": "<BASE64>", "mime_type": "image/jpeg"} items in the input array. Reference, edit, or extend videos use {"type": "video", "uri": "<FILE_URI>"} or {"type": "video", "data": "<BASE64>", "mime_type": "video/mp4"}.

Reference media limits

DurationMax reference imagesMax reference videos
3–10s63 (total ≤ 15s each)
11–15s61 (≤ 3s)

Create response

{
  "id": "9dcf8ab2-c4b7-42be-a516-597223897b99",
  "object": "interaction",
  "model": "google/gemini-omni-1.2-flash",
  "status": "queued"
}

id is the platform task ID. Video results are temporarily retained for 48 hours and automatically deleted upon expiration—download and transfer them immediately after completion.

See Tasks and callbacks for polling, callback delivery, and task lifecycle, and Gemini Interactions for the text path on the shared Interactions endpoint.

Help us improve this page

Found something unclear, outdated, or incorrect?

Last updated on