Image generation
Generate images from text descriptions using Google Gemini, and store the result for later workflow steps
Overview
The image-generation integration turns a text description into an image using Google Gemini (Nano Banana). The generated image is stored, and the tool returns its storage metadata so later activities can reference the file rather than passing pixels around the workflow.
It is a single-purpose integration: one tool, one job.
Capabilities
The image-generation MCP server exposes one tool.
generate_image
Generate an image from a text description using Google Gemini (Nano Banana). Returns the canonical storage metadata for the generated file, so a later step can attach, render, or hand it to a user.
Because the tool returns a storage reference rather than raw image data, the image never has to travel through the agent's context. Pass the returned key to a Respond node, an email tool, or file-generation to embed it in a document.
Configuration
Image generation needs a Google Gemini API key, supplied when the integration is connected. It is stored in HashiCorp Vault like every other credential, and is typically configured once by an administrator and locked, so project members can use the tool without handling the key.
Connect the integration
On the MCP Tools page, connect Image Generation.
Supply a Gemini API key
Provide the Google Gemini API key. It is written to Vault and injected at runtime, never exposed to agents or logs.
Enable it on an agent
Give the tool to the agents that should be able to generate images. Agents without it cannot call it.
Use cases
Illustrated reports
An agent writes a report section, generates a supporting illustration for it, then passes both to file-generation to render a branded PDF.
Marketing drafts
A workflow takes a product description as input, generates several image options, and routes them to a Human Task node for a person to pick one before anything is published.
Presentation visuals
Generate slide imagery, then use render_document and revise_document to assemble and iterate on a deck without regenerating it from scratch.