n8n Google Gemini nodeConfigure Google Gemini in n8n.

Sends one or more audio files to a Gemini model with a question, and returns the model's answer as text. Good for describing what a recording contains without reading a full transcript.

• Model: the Gemini model that handles the request, required, chosen from the list or set by expression.
• Input Type: url for URL(s), or binary for Input Data Field Name(s), with several values separated by commas.
• Length of Description (Max Tokens): fewer tokens give a shorter, less detailed audio description.

Turns audio into text. Unlike the analyze operation there is no question to ask: the output is the spoken content itself, ready for search, storage or a second model call.

• Model: required, the model that performs the transcription.
• URL(s): one or several audio addresses separated by commas, or Input Data Field Name(s) when the file arrives as binary data.
• Start Time and End Time: bounds in MM:SS or HH:MM:SS, so a two hour webinar can be reduced to the segment you care about.

Feeds a document to the model and answers a question about it, so a PDF stops being an opaque attachment and becomes fields the workflow can use.

• Text Input: the question, for example what the document contains or which totals it shows.
• URL(s): the document addresses, often {{ $json.fileUrl }} from the node before.
• Simplify Output: on, the node returns a trimmed response instead of the raw payload.

Creates an empty File Search store, the container Gemini uses for Retrieval Augmented Generation: the model answers from documents you uploaded rather than from its training alone.

• Display Name: required, a human-readable name for the store, which is how you recognize it later in the list.

Removes a store and, optionally, everything inside it. The operation is the cleanup counterpart of the create step, useful when stores are built per project and thrown away at the end.

• File Search Store Name: required, the full name in the format fileSearchStores/..., not the display name.
• Force Delete: on, related documents and objects go too; off, the deletion fails when the store still holds documents.

Returns the File Search stores owned by the account, which is how a workflow finds the full store name it needs before uploading or deleting.

• Page Size: how many stores come back per page, up to a maximum of 20.
• Page Token: the token from a previous page, used to fetch the next one; without it the workflow only ever sees the first page.

Adds a file to an existing store so the model can cite it later. This is the operation that actually makes a RAG setup useful, once the store exists.

• File Search Store Name: required, the full fileSearchStores/... name of the target store.
• File Display Name: required, and worth care because it shows up in citations.
• Input Type: url to point at URL, or binary to read the file from Input Data Field Name.

Asks a question about one or more images and returns the answer as text, which is how a workflow reads a screenshot, a photo or a scanned page.

• Text Input: what to ask, for instance what the picture shows or whether a label is present.
• URL(s): image addresses, several separated by commas, or Input Data Field Name(s) for binary fields.
• Length of Description (Max Tokens): caps the answer, fewer tokens meaning a shorter image description.

Creates images from a written description and returns them as binary data the workflow can store or send. No input file is involved here, only the prompt.

• Prompt: the text description of the image you want, as concrete as possible.
• Number of Images: how many variants come back in one run.
• Put Output in Field: the name of the output field that carries the binary file data, which the next node then reads.

Takes images you already have and reworks them according to an instruction, so the starting point is your own asset rather than a blank prompt.

• Prompt: the instruction describing the edit, such as combining the first image with the second.
• Images: the collection where you add one or more binary fields to send along with the prompt.
• Binary Field Name: the field holding the image data, when the default name does not match what the previous node produced.

Pushes a file to the Google Gemini API and keeps it there for later use, which saves resending the same large asset on every call.

• Input Type: url to fetch from URL, or binary to take the file from the workflow itself.
• Input Data Field Name: the binary property carrying the file, typically data when it comes from a download node.

The plain text operation: a completion from a Gemini model, with the widest set of options on the node.

• Messages: the conversation you send, each entry carrying a Prompt and a Role, user to ask, model to set a tone or personality.
• System Message: standing instructions that apply to the whole call.
• Output Randomness (Temperature): lower it for repeatable answers, raise it when variety is the point.
• Thinking Budget: 0 disables automatic thinking, -1 leaves it dynamic.

Answers questions about video content, which turns a file nobody has time to watch into a few usable sentences.

• Text Input: the question put to the model about the footage.
• URL(s): video addresses, comma separated for several at once, or Input Data Field Name(s) for binary input.
• Simplify Output: on, the response comes back trimmed rather than raw.

Builds a video from a written prompt. Generation takes longer than an image, so the operation lets you choose how the result comes back.

• Prompt: the description of the shot you want.
• Return As: video for a binary file in the workflow, url for an address you can download later.
• Aspect Ratio: 16:9 for screens, 9:16 for short-form vertical.
• Person Generation: dont_allow, allow_adult or allow_all, depending on what the brand accepts.

Fetches a generated video from the Google Gemini API using the address the API gave you, and brings the file into the workflow as binary data.

• URL: the Gemini API address to download from, usually {{ $json.url }} carried over from the generation step.
• Put Output in Field: the output field that receives the binary file data.