Google

OverviewCopy Anchor Link

GoogleImageGenService generates images from text prompts with Gemini image models through the Gemini API. Gemini returns each image inline in the response, and the service decodes it to RGB pixels for the output transport.

• GoogleImageGenService: one generateContent request per image, with aspect ratio and size settings.

InstallationCopy Anchor Link

bun add @pipedog/google

import { GoogleImageGenService } from "@pipedog/google/image";
import type { ImageDecoder } from "@pipedog/google/image_decoder";

PrerequisitesCopy Anchor Link

• A Gemini API key from Google AI Studio.
• Pass the key as apiKey. The examples read it from GOOGLE_API_KEY.
• An image decoder (see below).

GoogleImageGenServiceCopy Anchor Link

ConfigurationCopy Anchor Link

SettingsCopy Anchor Link

Runtime-updatable settings, passed as settings: new GoogleImageGenService.Settings({...}).

UsageCopy Anchor Link

import { GoogleImageGenService } from "@pipedog/google/image";
import type { ImageDecoder } from "@pipedog/google/image_decoder";

// Your decoder from PNG/JPEG/WebP bytes to RGB pixels (sharp, @jsquash/*, ...).
declare const decodeImage: ImageDecoder;

const imageGen = new GoogleImageGenService({
  apiKey: process.env.GOOGLE_API_KEY ?? "",
  imageDecoder: decodeImage,
  settings: new GoogleImageGenService.Settings({
    model: "gemini-3.1-flash-image",
    aspectRatio: "16:9",
    imageSize: "2K",
  }),
});

Image decoderCopy Anchor Link

Upstream decodes the returned PNG, JPEG or WebP with Pillow. pipedog has no edge-compatible decoder yet, so the service takes an imageDecoder: a function from the encoded bytes and MIME type to { image, size, mode } (raw pixels, [width, height] and a Pillow-style mode such as "RGB"). Build one on sharp, @jsquash/* or similar.

NotesCopy Anchor Link

• The service generates an image for each TextFrame it receives and emits a URLImageRawFrame per image, with url: null since Gemini returns the image inline.
• With numberOfImages above 1, the images are generated one after another, one request each.
• A response with no image, or a failed request, yields an ErrorFrame.
• Supports TTFB metrics.
• Pass client to use your own GenAI client (anything with models.generateContent).

Runtime notesCopy Anchor Link

Generated from pipecat.runtimes and pipecat.compat in packages/google/package.json: ✅ means the compat smoke test passes on that runtime.

The client is GoogleGenAI from @google/genai, which uses fetch, so the service runs on Bun, Node, Deno and Cloudflare Workers, as long as your imageDecoder does too.

The deprecated upstream InputParams class, the params option and the negativePrompt setting are not ported.