Gemini Omni Flash Video API

Gemini Omni Flash is a high-performance multimodal model for fast video generation, editing, and cinematic control. It can use text, images, audio, and video together, supports conversational editing through the Interactions API, and can apply Gemini world knowledge to physical scenes and storytelling.

Calling Method

Endpoint and Authentication

Use these Interactions endpoints to create requests or retrieve results:

text

POST https://api.exchangetoken.ai/v1beta/interactions
GET  https://api.exchangetoken.ai/v1beta/interactions/{id}

Use your Exchange Token API key in either of the following authentication headers:

http

x-goog-api-key: YOUR_API_KEY

http

Authorization: Bearer YOUR_API_KEY

Content-Type Header:

http

Content-Type: application/json

Request Body

The request body follows the Vertex AI Interactions API create schema. The examples below use these common fields:

Field	Type	Required	Description
`model`	string	Yes	Model name. Text and video editing examples use `gemini-omni-flash-preview`; set `$MODEL` in image examples to the model you are calling.
`previous_interaction_id`	string	No	The previous interaction's `id`, used to continue editing a generated video.
`input`	string, content object, or content array	Yes	Main interaction input. Use a string for text-to-video, or an array for multimodal image/video editing input.
`input[].type`	string	Conditional	Content item type, such as `text`, `image`, `video`, or `user_input`.
`input[].text`	string	Conditional	Text prompt when `type` is `text`.
`input[].data`	string	Conditional	Base64 image content. Videos use `uri` and do not accept Base64 `data`.
`input[].uri`	string	Conditional	Media file address.
`input[].mime_type`	string	Conditional	MIME type for media inputs, such as `image/jpeg`, `image/png`, or `video/mp4`.
`input[].content`	array	Conditional	Nested content array when using a `user_input` wrapper.
`background`	boolean	No	Run in the background; defaults to `false`.
`store`	boolean	No	Defaults to `true`.
`stream`	boolean	No	Return SSE events; defaults to `false`.
`response_format`	object or array	No	Output format options. Examples use an object; for an array, configure each output item.
`response_format.type`	string	No	Set to `video` for video output.
`response_format.delivery`	string	No	Delivery mode. Examples use `uri` to receive a video link.
`response_format.aspect_ratio`	string	No	Video aspect ratio: `16:9` or `9:16`.
`response_format.resolution`	string	No	Output resolution, such as `720p`.
`response_format.duration`	string	No	Video duration, such as `5s`; include the number and unit in one string.
`response_format.gcs_uri`	string	Omit	The platform supplies storage for video URI output.
`generation_config`	object	No	Generation options.
`generation_config.video_config.task`	string	No	Optional video task hint. Values include `text_to_video`, `image_to_video`, `reference_to_video`, `edit`, and `extend`; follow the selected model's parameter constraints.

TIP

To receive a video link, set both response_format.type: video and response_format.delivery: uri, and omit gcs_uri. The platform allocates output storage and converts storage URIs in model output to signed download links.

Response Body

Common fields:

Field	Type	Description
`id`	string	Unique identifier of the interaction
`object`	string	Resource type identifier, typically `interaction`
`status`	string	Current interaction status, such as `in_progress`, `completed`, or `failed`
`role`	string	Role associated with the response payload, typically `model`
`created`	string	Interaction creation time in ISO 8601 format
`updated`	string	Last interaction update time in ISO 8601 format
`model`	string	Name of the model used for the interaction
`steps`	array	Ordered step objects returned by the interaction
`usage`	object	Token usage summary for the interaction when provided by upstream

Common steps fields:

Field	Type	Description
`type`	string	Step type, for example `model_output`
`content`	array	Content items carried by the step
`content[].uri`	string	Video download URL; read it from a `type: model_output` step

Example:

json

{
  "id": "v1_...",
  "status": "completed",
  "role": "model",
  "created": "2026-05-27T06:15:03Z",
  "updated": "2026-05-27T06:15:03Z",
  "steps": [
    {
      "content": [
        {
          "text": "",
          "type": "text"
        }
      ],
      "type": "model_output"
    },
    {
      "content": [
        {
          "mime_type": "video/mp4",
          "uri": "https://storage.googleapis.com/example-bucket/output.mp4?X-Goog-Signature=EXAMPLE",
          "type": "video"
        }
      ],
      "type": "model_output"
    }
  ],
  "object": "interaction",
  "model": "gemini-omni-flash-preview"
}

Get an Input Video URI

Video input must be provided through uri and only supports Google Cloud Storage addresses (gs://...). Base64 video input is not supported. If you have a publicly accessible gs:// address, use it directly. Otherwise, use the platform's Google Cloud Storage upload feature: follow the steps below to upload the video and use the returned gcs_uri as the input URI.

Get an Upload URL

Call the upload URL endpoint with your Exchange Token API key. The request body accepts only an optional filename to specify the file name.

bash

curl --fail-with-body -sS -X POST "https://api.exchangetoken.ai/v1/storage/upload-urls" \
  -H "x-goog-api-key: $API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"filename":"source.mp4"}' \
  -o upload.json

Example response:

json

{
  "upload_url": "https://storage.googleapis.com/BUCKET/uploads/.../source.mp4?X-Goog-...",
  "gcs_uri": "gs://BUCKET/uploads/.../source.mp4",
  "expires_at": "2026-10-14T00:00:00Z",
  "method": "PUT",
  "headers": {}
}

Use upload_url to upload the file and set the video input's uri field in the Interactions API to gcs_uri. The upload link is valid for 24 hours; check expires_at for its exact expiry. This request only allocates an upload URL; the video file still needs to be uploaded.

Upload a Local Video

Read the addresses from upload.json and upload the local file with PUT. The example uses Bash, curl, and jq.

bash

VIDEO_FILE="./source.mp4"
UPLOAD_URL=$(jq -er '.upload_url' upload.json)
VIDEO_URI=$(jq -er '.gcs_uri' upload.json)

curl --fail-with-body -sS -X PUT "$UPLOAD_URL" \
  --upload-file "$VIDEO_FILE"

After upload, $VIDEO_URI can be used as the uri of a video content item in an Interactions API request.

Text to Video

Generate a video with audio from a text prompt. For best results, describe the scene, camera movement, lighting, mood, and key actions.

bash

curl -X POST "https://api.exchangetoken.ai/v1beta/interactions" \
  -H "x-goog-api-key: $API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gemini-omni-flash-preview",
    "input": "A marble rolling fast on a chain reaction style track, continuous smooth shot.",
    "response_format": { "type": "video", "delivery": "uri" }
  }'

Control Aspect Ratio

Set response_format.aspect_ratio to 9:16 for portrait video or 16:9 for landscape video.

bash

curl -X POST "https://api.exchangetoken.ai/v1beta/interactions" \
  -H "x-goog-api-key: $API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gemini-omni-flash-preview",
    "input": "A futuristic city with neon lights and flying cars, cyberpunk style",
    "response_format": {
      "type": "video",
      "delivery": "uri",
      "aspect_ratio": "9:16"
    }
  }'

Image to Video

Put images and the text prompt in the same input array. Set $MODEL to the model you are calling and $BASE64_IMAGE to the image content encoded as Base64.

bash

curl -X POST "https://api.exchangetoken.ai/v1beta/interactions" \
  -H "x-goog-api-key: $API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "'"$MODEL"'",
    "input": [
      { "type": "image", "data": "'"$BASE64_IMAGE"'", "mime_type": "image/jpeg" },
      { "type": "text", "text": "turn this into realistic footage, using the drawing only as a guide for movement, do not show the drawing in the final video" }
    ],
    "response_format": { "type": "video", "delivery": "uri" }
  }'

TIP

Images can also reference uploaded files through uri. For data, supply plain Base64 content and a mime_type matching the file.

Subject Reference

Put multiple reference images in the input array and describe the role of each image in the prompt.

bash

curl -X POST "https://api.exchangetoken.ai/v1beta/interactions" \
  -H "x-goog-api-key: $API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "'"$MODEL"'",
    "input": [
      { "type": "image", "data": "'"$CAT_B64"'", "mime_type": "image/png" },
      { "type": "image", "data": "'"$YARN_B64"'", "mime_type": "image/png" },
      { "type": "text", "text": "A cat playfully batting at a ball of yarn." }
    ],
    "response_format": { "type": "video", "delivery": "uri" }
  }'

Task Parameter

Use generation_config.video_config.task to specify a task mode. The gateway forwards this field; when omitted, the model determines the task from the prompt and input media. See the Vertex AI VideoConfig definition.

Parameter values (follow the selected model's constraints):

text_to_video
image_to_video
reference_to_video
edit
extend

bash

curl -X POST "https://api.exchangetoken.ai/v1beta/interactions" \
  -H "x-goog-api-key: $API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "'"$MODEL"'",
    "input": [
      { "type": "image", "data": "'"$BASE64_IMAGE"'", "mime_type": "image/jpeg" },
      { "type": "text", "text": "turn this into realistic footage, using the drawing only as a guide for movement, do not show the drawing in the final video" }
    ],
    "generation_config": {
      "video_config": {
        "task": "image_to_video"
      }
    },
    "response_format": { "type": "video", "delivery": "uri" }
  }'

Edit Your Own Videos

Submit the edit with $VIDEO_URI ready. See Get an Input Video URI for how to obtain the address.

bash

curl -sS -w "\n[HTTP %{http_code}]\n" "https://api.exchangetoken.ai/v1beta/interactions" \
  -H "x-goog-api-key: $API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gemini-omni-flash-preview",
    "input": [
      {
        "type": "user_input",
        "content": [
          {
            "type": "video",
            "mime_type": "video/mp4",
            "uri": "'"$VIDEO_URI"'"
          },
          {
            "type": "text",
            "text": "When the person touches the mirror, make the mirror ripple beautifully like liquid, and the person arm turns into reflective mirror material"
          }
        ]
      }
    ],
    "generation_config": { "video_config": { "task": "edit" } },
    "response_format": { "type": "video", "delivery": "uri" }
  }'

Multi-turn Editing

After the previous turn completes, previous_interaction_id references that interaction and input provides the new editing instructions. Subsequent requests must use the same API key and model.

When using previous_interaction_id, omit generation_config.video_config.task and describe the edits in input to avoid an upstream parameter conflict error.

bash

PREVIOUS_INTERACTION_ID="YOUR_INTERACTION_ID"

curl -X POST "https://api.exchangetoken.ai/v1beta/interactions" \
  -H "x-goog-api-key: $API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gemini-omni-flash-preview",
    "previous_interaction_id": "'"$PREVIOUS_INTERACTION_ID"'",
    "input": "Keep the subject and camera motion unchanged, and change the background to a snowy mountain.",
    "response_format": { "type": "video", "delivery": "uri" }
  }'

Background Generation and Retrieval

background: true enables background execution. Creation and retrieval requests must use the same API key for authentication.

bash

curl -X POST "https://api.exchangetoken.ai/v1beta/interactions" \
  -H "x-goog-api-key: $API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gemini-omni-flash-preview",
    "input": "A paper boat drifts across a quiet pond in soft morning light.",
    "background": true,
    "response_format": { "type": "video", "delivery": "uri" }
  }'

GET /v1beta/interactions/{id} retrieves the task status and result. The {id} path parameter is the interaction identifier returned by the creation request.

bash

INTERACTION_ID="YOUR_INTERACTION_ID"

curl --fail-with-body -sS "https://api.exchangetoken.ai/v1beta/interactions/$INTERACTION_ID" \
  -H "x-goog-api-key: $API_KEY" \
  -o result.json

The response field status indicates the task state:

Status	Description
`in_progress`	The task is running and can be polled periodically.
`completed`	The task has completed; generated results are available in `steps`.
`failed`	The task has failed; error details are included in the response.
`incomplete`	The task did not complete fully; details are included in the response.

Gemini Omni Flash Video API ​

Calling Method ​

Endpoint and Authentication ​

Request Body ​

Response Body ​

Get an Input Video URI ​

Get an Upload URL ​

Upload a Local Video ​

Text to Video ​

Control Aspect Ratio ​

Image to Video ​

Subject Reference ​

Task Parameter ​

Edit Your Own Videos ​

Multi-turn Editing ​

Background Generation and Retrieval ​

Gemini Omni Flash Video API

Calling Method

Endpoint and Authentication

Request Body

Response Body

Get an Input Video URI

Get an Upload URL

Upload a Local Video

Text to Video

Control Aspect Ratio

Image to Video

Subject Reference

Task Parameter

Edit Your Own Videos

Multi-turn Editing

Background Generation and Retrieval