Package com.google.genkit.plugins.googlegenai

Class OmniModel

java.lang.Object

com.google.genkit.plugins.googlegenai.OmniModel

All Implemented Interfaces:

Model, Action<ModelRequest,ModelResponse,ModelResponseChunk>, Registerable

public class OmniModel
extends Object
implements Model

Gemini Omni video generation and editing model using the Gemini Interactions API.

Gemini Omni (gemini-omni-flash-preview) natively processes text, image, audio, and video and produces video (with audio). It is served by the Interactions API (POST /v1beta/interactions) rather than generateContent, and supports conversational editing: each turn can build on a previous result.

Conversational editing. The Interactions API is stateful server-side. This model bridges it to Genkit's stateless Model interface as follows:

• On the first turn, send a prompt (and optional media). The response's custom map carries the returned interactionId.
• To iteratively edit, pass that id back as the previousInteractionId config option on the next generate call, with the edit instruction as the prompt. The model applies the change while preserving elements you did not mention.

Supported config options (via the request config map): previousInteractionId, aspectRatio (e.g. "16:9"), duration (e.g. "10s"), delivery ("inline" (default) | "uri"), task ("text_to_video" | "image_to_video"), thinkingLevel, maxOutputTokens.

Only the Gemini Developer API (API key) is supported; Vertex AI is not.

Note: the Interactions API is in preview; request/response field shapes may evolve.

Constructor Summary

Constructors

Constructor	Description
OmniModel(String modelName, GoogleGenAIPluginOptions options)	Creates a new OmniModel.

Method Summary

Modifier and Type	Method	Description
ModelInfo	getInfo()	Gets information about the model's capabilities.
String	getName()	Returns the name of the action.
ModelResponse	run(ActionContext context, ModelRequest request)	Generates a response from the given request.
ModelResponse	run(ActionContext context, ModelRequest request, Consumer<ModelResponseChunk> streamCallback)	Generates a streaming response from the given request.
boolean	supportsStreaming()	Returns whether this model supports streaming.

Methods inherited from class java.lang.Object

equals, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Methods inherited from interface com.google.genkit.ai.Model

getDesc, getInputSchema, getMetadata, getOutputSchema, getType, register, runJson, runJsonWithTelemetry

Constructor Details

OmniModel

public OmniModel(String modelName,
 GoogleGenAIPluginOptions options)

Creates a new OmniModel.

Parameters:

modelName - the model name (e.g., "gemini-omni-flash-preview")

options - the plugin options

Method Details

getName

public String getName()

Description copied from interface: Action

Returns the name of the action.

Specified by:

getName in interface Action<ModelRequest,ModelResponse,ModelResponseChunk>

Returns:

the action name

getInfo

public ModelInfo getInfo()

Description copied from interface: Model

Gets information about the model's capabilities.

Specified by:

getInfo in interface Model

Returns:

the model info

supportsStreaming

public boolean supportsStreaming()

Description copied from interface: Model

Returns whether this model supports streaming.

Specified by:

supportsStreaming in interface Model

Returns:

true if streaming is supported

run

public ModelResponse run(ActionContext context,
 ModelRequest request)

Description copied from interface: Model

Generates a response from the given request.

Specified by:

run in interface Action<ModelRequest,ModelResponse,ModelResponseChunk>

Specified by:

run in interface Model

Parameters:

context - the action context

request - the model request

Returns:

the model response

run

public ModelResponse run(ActionContext context,
 ModelRequest request,
 Consumer<ModelResponseChunk> streamCallback)

Description copied from interface: Model

Generates a streaming response from the given request.

Specified by:

run in interface Action<ModelRequest,ModelResponse,ModelResponseChunk>

Specified by:

run in interface Model

Parameters:

context - the action context

request - the model request

streamCallback - callback for streaming chunks

Returns:

the final model response