GoogleGeminiConnector

Connects to Google Gemini models for content generation, image creation, image editing, speech, and embeddings.

Typical Uses

Generating text responses, summaries, or structured JSON from a prompt using a Gemini model
Creating or editing images from a text prompt using a Gemini image model
Synthesizing speech audio from text, including multi-speaker dialogue
Generating vector embeddings for text, image, audio, video, or PDF input

How does it work?

The GoogleGeminiConnector uses your credentials to connect to Google Gemini and perform various tasks:

Action	Task
Generate Content	Send a text prompt (with optional supporting media, chat history, and tool use) to a Gemini model and return the generated text response. This also can be used to analyze or describe audio or video input.
Generate Image	Generate an image from a text prompt using a Gemini image model.
Edit Image	Edit an existing input image according to a text prompt, using a Gemini image model.
Generate Speech	Synthesize speech audio from text, with single- or two-speaker voice configuration.
Generate Embeddings	Generate a vector embedding for text, image, audio, video, or PDF input.

Optional Input Port

This transformer has two modes, depending on whether a connector is attached to the Input port or not:

Input-driven: When input features are connected, the transformer runs once for each feature it receives in the Input port.
Run Once: When no input features are connected, the transformer runs one time.

When the Input port is in use, the Initiator output port is also enabled.

Usage Notes

All actions except Generate Embeddings call the same underlying Gemini Interactions API endpoint; Generate Embeddings uses a separate endpoint (embedContent)
For Generate Image and Edit Image the generated image is always attached to the output feature as raster geometry, regardless of Save Response To. Only PNG and JPEG images support this; other formats log a warning and are skipped, without affecting the non-geometry response data.
If multiple supporting inputs need to be provided for an action (such as multiple images for Edit Image or multiple JSON files for Generate Content ), set Supporting Input to File. This is the only input method which supports multiple supporting inputs.
The transformer will only include parts of the response explicitly marked as model_output when outputting its response. To see the full response, include _googleai_json_response in Attributes To Add. To include model reasoning output (thinking), include _googleai_thinking_summaries{} in Attributes To Add. Note that including thinking summaries will increase token usage.
To continue a Generate Content workflow across multiple requests, enable Chat History and provide a Previous Interaction ID set to the _googleai_interaction_id attribute value of the previous feature.
Users are responsible for verifying output. Text content which is blocked ( due to content filters, for example) may not always result in a <Rejected> feature.
Google releases some Gemini models and functionality on its beta API before the stable one. If a model or functionality returns an error, try setting the web connection's Base URL to the v1beta option.
Users can edit the AI Model parameter to specify a custom Google Gemini model. For a list of all models, see https://ai.google.dev/gemini-api/docs/models. Note that not all models are supported by the underlying API and that only models suited for the selected Action should be used.
Data sent to Google AI is subject to Google's data handling and privacy policies.

Configuration

Input Ports

Output Ports

Output

Features with added attributes, as specified in parameters and according to Action.

Generated content, speech, and embeddings are output according to the Save Response To parameter. Images are output as raster geometry.

Action	Output - Input-Driven	Output - Run Once
Generate Content	Input feature with the generated response.	New feature with the generated response.
Generate Image	Input feature with the generated image.	New feature with the generated image.
Edit Image	Input feature with the revised image.	New feature with the revised image.
Generate Speech	Input feature with the generated speech.	New feature with the generated speech.
Generate Embeddings	Input feature with the generated embedding vector.	New feature with the generated embedding vector.

Parameters

Request

Action	Select an operation to perform. Choices include: Generate Content (default) Generate Image Edit Image Generate Speech Generate Embeddings
AI Model	Specify the Gemini model to use.

Action > Generate Content

Generate Content Options

User Prompt

Provide a text prompt to send to the model.

Supporting Input

Images, audio clips, video, or documents can be submitted along with the User Prompt so the model can answer questions about them.

Select a method of adding additional materials:

None (default)
File
Attribute

Various image, audio, video, and Document (e.g. PDF, CSV, JSON, or other text documents) file types are supported.

Media up to 100 MB (50 MB for a PDF) is sent inline with the request. Anything larger is uploaded to Google first and referenced by URL, which happens automatically and needs no configuration — expect the feature to take longer, since the whole file has to be transferred before the prompt can run. The upload accepts up to 2 GB; beyond that the feature is rejected.

Uploaded media is held by Google for 48 hours and is not reused between features. Each feature uploads its own copy.

Files

When Supporting Input is File, specify the files to include.

Attribute

When Supporting Input is Attribute, specify the attribute containing the supporting media.

MIME Type

When Supporting Input is Attribute, specify the media type of the material held in the specified attribute.

Thinking Level

Select an option specifying how much reasoning the model does before answering.

Default (default). Use the model’s own default value.
Minimal
Low
Medium
High

More reasoning means better answers on hard questions; less means a faster, cheaper response.

Note that not every reasoning level will be supported by every model. Increasing the thinking level will increase token usage, so the value of Max Output Tokens (if set) should be adjusted accordingly.

Max Output Tokens

(Optional) Specify a limit to the length of the response.

This parameter caps tokens spent as part of the model's internal reasoning as well as its reply, so a low cap can result in no response being returned.

If setting the cap, it is recommended to set it to at least a few hundred tokens above the length of the expected response.

Stop Sequences

(Optional) Provide any stop sequence phrases as comma-delimited text.

If any of these phrases are generated, the response will end right before the stop sequence.

Stop sequences are case-sensitive.

Seed

(Optional) Provide a fixed seed for more repeatable output.

Note that this is best-effort, and the same seed is not guaranteed to give the same response.

Chat History

When enabled, a previous conversation is continued. Only the interaction ID will be included in the request. The conversation history is stored by Google.

Previous Interaction ID

Provide the id of a previous response to continue from, typically set to the _googleai_interaction_id attribute value from an earlier Generate Content action.

Note that IDs are tied to the account that created them, so an Interaction ID cannot be used with two different sets of credentials. Google controls when Interaction IDs are deactivated.

Context Tools

Additional tools are available to ground results.

Use Google Search	When Yes, ground responses using Google Search results.
Use Code Execution	When Yes, allow the model to write and execute code as part of generating its response.
Use URL Context	When Yes, direct the model to retrieve content from URLs referenced in the prompt.
Use Google Maps	When Yes, direct the model to ground its response in a specific geographic location.
Latitude	When Use Google Maps is Yes, provide a latitude coordinate. The coordinate system is assumed to be EPSG:4326.
Longitude	When Use Google Maps is Yes, provide a longitude coordinate. The coordinate system is assumed to be EPSG:4326.
Use File Search	When Yes, direct the model to ground the response in a searchable document store.
Store Names	When Use File Search is Yes, specify the stores to search.

Output Attributes

Attributes to Add

Select any available attributes to be included on the output features:

_googleai_annotations{}	A list attribute holding the citations behind the response, one JSON record per element. Available in the Generate Content action when the File Search or Google Search tools are enabled.
_googleai_interaction_id	The unique ID of the API response. Not offered for Generate Embeddings, which never returns one.
_googleai_json_response	The raw steps array from the Interactions API response, as a JSON string. Not offered for Generate Embeddings, which never returns one.
_googleai_model_status	The finish/status reason reported by the model (for example, STOP). Not offered for Generate Embeddings, which never returns one.
_googleai_model_version	The specific model version that served the request.
_googleai_thinking_summaries{}	A list attribute containing a summary of the model's reasoning per reasoning pass. This should be treated as commentary, and not as a record of how an answer was reached. Thinking summaries are only requested from the model when this attribute is selected. Note that selecting this attribute increases token usage. If no reasoning took place (e.g. Thinking Level was set to Minimal), no summaries will be generated. Not offered for Generate Embeddings.
_googleai_usage	Token/usage metadata returned by the API for the request.

Action > Generate Image

Generate Image Options

Prompt

Provide a text prompt describing the image to generate.

Aspect Ratio

Select an aspect ratio for the generated image.

If the selected aspect ratio is not supported by a model, it will fall back to the nearest available ratio.

Set to Default to use the model default.

Output Attributes

Attributes to Add

Select any available attributes to be included on the output features:

_googleai_annotations{}	A list attribute holding the citations behind the response, one JSON record per element. Available in the Generate Content action when the File Search or Google Search tools are enabled.
_googleai_interaction_id	The unique ID of the API response. Not offered for Generate Embeddings, which never returns one.
_googleai_json_response	The raw steps array from the Interactions API response, as a JSON string. Not offered for Generate Embeddings, which never returns one.
_googleai_model_status	The finish/status reason reported by the model (for example, STOP). Not offered for Generate Embeddings, which never returns one.
_googleai_model_version	The specific model version that served the request.
_googleai_thinking_summaries{}	A list attribute containing a summary of the model's reasoning per reasoning pass. This should be treated as commentary, and not as a record of how an answer was reached. Thinking summaries are only requested from the model when this attribute is selected. Note that selecting this attribute increases token usage. If no reasoning took place (e.g. Thinking Level was set to Minimal), no summaries will be generated. Not offered for Generate Embeddings.
_googleai_usage	Token/usage metadata returned by the API for the request.

Action > Edit Image

Edit Image Options

Prompt

Provide a text prompt describing how to edit the input images.

Input Images

Provide the file paths containing the image(s) to edit. Note that even if multiple reference images are provided, the model will only output a single image.

Aspect Ratio

Select an aspect ratio for the edited image.

If the selected aspect ratio is not supported by a model, it will fall back to the nearest available ratio.

Set to Default to use the model default.

Output Attributes

Attributes to Add

Select any available attributes to be included on the output features:

_googleai_annotations{}	A list attribute holding the citations behind the response, one JSON record per element. Available in the Generate Content action when the File Search or Google Search tools are enabled.
_googleai_interaction_id	The unique ID of the API response. Not offered for Generate Embeddings, which never returns one.
_googleai_json_response	The raw steps array from the Interactions API response, as a JSON string. Not offered for Generate Embeddings, which never returns one.
_googleai_model_status	The finish/status reason reported by the model (for example, STOP). Not offered for Generate Embeddings, which never returns one.
_googleai_model_version	The specific model version that served the request.
_googleai_thinking_summaries{}	A list attribute containing a summary of the model's reasoning per reasoning pass. This should be treated as commentary, and not as a record of how an answer was reached. Thinking summaries are only requested from the model when this attribute is selected. Note that selecting this attribute increases token usage. If no reasoning took place (e.g. Thinking Level was set to Minimal), no summaries will be generated. Not offered for Generate Embeddings.
_googleai_usage	Token/usage metadata returned by the API for the request.

Action > Generate Speech

Generate Speech Options

Number of Speakers	Select the number of speakers in the Text: One (default): One narrator voice Two: Two-speaker dialogue
Text	Provide the text to synthesize into speech. Optional style directions may be included, as in Speaker should sound pleasant and cheerful. When Number of Speakers is Two, format the text as one line per turn, prefixing each line with the same names provided in the Speaker One Name and Speaker Two Name parameters, followed by a colon, as in: Alice: Hi, my name is Alice. What's your name? Bob: My name is Bob. Alice: Nice to meet you, Bob.
Voice	When Number of Speakers is One, select a voice to use.
Speaker One Name	When Number of Speakers is Two, provide a name for the first speaker. This should be used as an identifier in the dialogue script provided in the Text .
Speaker One Voice	When Number of Speakers is Two, select a voice for the first speaker.
Speaker Two Name	When Number of Speakers is Two, provide a name for the second speaker. This should be used as an identifier in the dialogue script provided in the Text .
Speaker Two Voice	When Number of Speakers is Two, select a voice for the second speaker.

Output Attributes

Attributes to Add

Select any available attributes to be included on the output features:

_googleai_annotations{}	A list attribute holding the citations behind the response, one JSON record per element. Available in the Generate Content action when the File Search or Google Search tools are enabled.
_googleai_interaction_id	The unique ID of the API response. Not offered for Generate Embeddings, which never returns one.
_googleai_json_response	The raw steps array from the Interactions API response, as a JSON string. Not offered for Generate Embeddings, which never returns one.
_googleai_model_status	The finish/status reason reported by the model (for example, STOP). Not offered for Generate Embeddings, which never returns one.
_googleai_model_version	The specific model version that served the request.
_googleai_thinking_summaries{}	A list attribute containing a summary of the model's reasoning per reasoning pass. This should be treated as commentary, and not as a record of how an answer was reached. Thinking summaries are only requested from the model when this attribute is selected. Note that selecting this attribute increases token usage. If no reasoning took place (e.g. Thinking Level was set to Minimal), no summaries will be generated. Not offered for Generate Embeddings.
_googleai_usage	Token/usage metadata returned by the API for the request.

Action > Generate Embeddings

Generate Embeddings Options

Input Type	When AI Model is gemini-embedding-2, select the type of content to embed: Text (default) Image/Audio/Video/PDF
Task Mode	When Input Type is Text, specify the embedding task mode: Retrieval (Document): The embedding is the query side of a retrieval pair. Retrieval (Query): The embedding is the document side of a retrieval pair. Symmetric (default): This is a symmetric single-input task.
Task Type	When AI Model is gemini-embedding-001, select the intended downstream use of the embedding, which optimizes its quality for that task: Classification Clustering Code Retrieval Query Fact Verification Question Answering Retrieval Document Retrieval Query Semantic Similarity (default) When AI Model is gemini-embedding-2 and Input Type is Text, select the task type that the embedding should optimize for. Options vary depending on the Task Mode selected.
Document Title	(Optional) When Task Type is Retrieval Document (gemini-embedding-001) or Task Mode is Retrieval (Document) (gemini-embedding-2), provide the title of the document to embed.
Input Text	When AI Model is gemini-embedding-001 or Input Type is Text, provide the text content to embed.
Supporting Input	When Input Type is Image/Audio/Video/PDF, select the source of the embedded input: File (default) Attribute
Files	When Supporting Input is File, select any files to embed. The MIME type of the input will be derived from the file extensions. If multiple files are selected, these will be sent together in a single request, producing a single aggregate embedding for all files, rather than separate embeddings for each file. To produce separate embeddings for each file, use multiple input features to the transformer.
Attribute	When Supporting Input is Attribute, specify the attribute holding the input to embed.
MIME Type	When Supporting Input is Attribute, select the MIME type of the attribute's content.
Description	(Optional) When Input Type is Image/Audio/Video/PDF, provide a text description of the non-text input being embedded. This description will be embedded along with the input.
Output Dimensionality	Specify number of dimensions in the output embedding vector.

Output Attributes

Attributes to Add

Select any available attributes to be included on the output features:

_googleai_model_version	The specific model version that served the request.
_googleai_usage	Token/usage metadata returned by the API for the request.

AI Model Behavior

System Instructions

(Optional) When Action is Generate Content, Generate Image, or Edit Image, provide instructions that steer the model's overall behavior, separate from the user prompt.

Structured Output

When enabled and Action is Generate Content, constrain the response to a specific JSON schema.

JSON Schema

Provide the JSON schema the response must conform to. Keys should be specified as properties under a top level object, for example:

Copy

{
  "type": "object",
  "properties": {
    "city":        { "type": "string" },
    "province":    { "type": "string" },
    "postal_code": { "type": "string" }
  },
  "required": ["city", "province", "postal_code"]
}

See https://ai.google.dev/gemini-api/docs/structured-output for more details.

Advanced

Transport and Resilience

Request Timeout (seconds)

(Optional) Specify the number of seconds to wait for a response before timing out.

Retry Count

(Optional) Specify the number of times to retry a failed request. Range 0–10, default 3.

Custom HTTP Headers

When enabled, additional headers may be defined.

Each custom Header Name/Value pair will be included with API requests.

Header Name	Provide a header name.
Value	Provide a value.

Destination

Save Response To	Select a destination type for the response: Attribute (default) File Folder - when writing a response to a folder, a filename will be auto-generated.
Download to Attribute	When Save Response To is Attribute, specify the attribute to write the response to.
Filename	When Save Response To is File, specify the name of the output file. This name should not include the file extension (as in output_file instead of output_file.txt). The file extension will be determined automatically or can be set using the File Extension parameter, if it is available. If the default value of Auto-generate is selected, the filename will be <yymmddhhmmss>_<interaction_id>.
Download to Folder	When Save Response To is Folder or File, specify the full path of the folder to write to.
File Extension	When Save Response To is Folder or File and Action is Generate Content, set the extension to use for text-based responses.
If File Exists	When Save Response To is Folder or File, specify behavior when the output file already exists: Overwrite (default) Reject

Editing Transformer Parameters

Transformer parameters can be set by directly entering values, using expressions, or referencing other elements in the workspace such as attribute values or user parameters. Various editors and context menus are available to assist. To see what is available, click beside the applicable parameter.

How to Set Parameter Values

Defining Values

There are several ways to define a value for use in a Transformer. The simplest is to simply type in a value or string, which can include functions of various types such as attribute references, math and string functions, and workspace parameters.

Using the Text Editor

The Text Editor provides a convenient way to construct text strings (including regular expressions) from various data sources, such as attributes, parameters, and constants, where the result is used directly inside a parameter.

Text Editor

Using the Arithmetic Editor

The Arithmetic Editor provides a convenient way to construct math expressions from various data sources, such as attributes, parameters, and feature functions, where the result is used directly inside a parameter.

Arithmetic Editor

Conditional Values

Set values depending on one or more test conditions that either pass or fail.

Parameter Condition Definition Dialog

Content

Expressions and strings can include a number of functions, characters, parameters, and more.

When setting values - whether entered directly in a parameter or constructed using one of the editors - strings and expressions containing String, Math, Date/Time or FME Feature Functions will have those functions evaluated. Therefore, the names of these functions (in the form @<function_name>) should not be used as literal string values.

Content Types

String Functions	These functions manipulate and format strings.
Special Characters	A set of control characters is available in the Text Editor.
Math Functions	Math functions are available in both editors.
Date/Time Functions	Date and time functions are available in the Text Editor.
Math Operators	These operators are available in the Arithmetic Editor.
FME Feature Functions	These return primarily feature-specific values.
FME Parameters	FME and workspace-specific parameters may be used.
Creating and Modifying User Parameters	Create your own editable parameters.

Dialog Options - Tables

Table Tools

Transformers with table-style parameters have additional tools for populating and manipulating values.

Row Reordering

Enabled once you have clicked on a row item. Choices include:

Add a row
Remove a row
Move current row up one
Move current row down one
Move current row to top
Move current row to bottom

Cut, Copy, and Paste

Enabled once you have clicked on a row item. Choices include:

Cut a row - delete and copy to clipboard
Copy a row to the clipboard
Paste a row from the clipboard

Cut, copy, and paste may be used within a transformer, or between transformers.

Filter

Start typing a string, and the matrix will only display rows matching those characters. Searches all columns. This only affects the display of attributes within the transformer - it does not alter which attributes are output.

Import

Import populates the table with a set of new attributes read from a dataset. Specific application varies between transformers.

Reset/Refresh

Generally resets the table to its initial state, and may provide additional options to remove invalid entries. Behavior varies between transformers.

Note: Not all tools are available in all transformers.

For more information, see Transformer Parameter Menu Options.

Reference

Processing Behavior	Feature-Based
Feature Holding	No
Dependencies
Aliases	GeminiAIConnector, GeminiConnector, GoogleAIConnector
History

FME Online Resources

The FME Community and Support Center Knowledge Base have a wealth of information, including active forums with 35,000+ members and thousands of articles.

Search for all results about the GoogleGeminiConnector on the FME Community.

Examples may contain information licensed under the Open Government Licence – Vancouver, Open Government Licence - British Columbia, and/or Open Government Licence – Canada.