Gemini-API

A reverse-engineered asynchronous python wrapper for Google Gemini web app (formerly Bard).

Features

• Persistent Cookies - Automatically refreshes cookies in background. Optimized for always-on services.
• ImageFx Support - Supports retrieving images generated by ImageFx, Google's latest AI image generator.
• Extension Support - Supports generating contents with Gemini extensions on, like YouTube and Gmail.
• Classified Outputs - Automatically categorizes texts, web images and AI generated images in the response.
• Official Flavor - Provides a simple and elegant interface inspired by Google Generative AI's official API.
• Asynchronous - Utilizes asyncio to run generating tasks and return outputs efficiently.

Table of Contents

• Features
• Table of Contents
• Installation
• Authentication
• Usage
  • Initialization
  • Generate contents from text
  • Generate contents from image
  • Conversations across multiple turns
  • Continue previous conversations
  • Retrieve images in response
  • Generate images with ImageFx
  • Save images to local files
  • Generate contents with Gemini extensions
  • Check and switch to other reply candidates
  • Control log level
• References
• Stargazers

Installation

Note

This package requires Python 3.10 or higher.

Install/update the package with pip.

pip install -U gemini_webapi

Optionally, package offers a way to automatically import cookies from your local browser. To enable this feature, install browser-cookie3 as well. Supported platforms and browsers can be found here.

pip install -U browser-cookie3

Authentication

Tip

If browser-cookie3 is installed, you can skip this step and go directly to usage section. Just make sure you have logged in to https://gemini.google.com in your browser.

• Go to https://gemini.google.com and login with your Google account
• Press F12 for web inspector, go to Network tab and refresh the page
• Click any request and copy cookie values of __Secure-1PSID and __Secure-1PSIDTS

Note

If your application is deployed in a containerized environment (e.g. Docker), you may want to persist the cookies with a volume to avoid re-authentication every time the container rebuilds.

Here's part of a sample docker-compose.yml file:

services:
  main:
    volumes:
      - ./gemini_cookies:/usr/local/lib/python3.12/site-packages/gemini_webapi/utils/temp

Note

API's auto cookie refreshing feature doesn't require browser-cookie3, and by default is enabled. It allows you to keep the API service running without worrying about cookie expiration.

This feature may cause that you need to re-login to your Google account in the browser. This is an expected behavior and won't affect the API's functionality.

To avoid such result, it's recommended to get cookies from a separate browser session and close it as asap for best utilization (e.g. a fresh login in browser's private mode). More details can be found here.

Usage

Initialization

Import required packages and initialize a client with your cookies obtained from the previous step. After a successful initialization, the API will automatically refresh __Secure-1PSIDTS in background as long as the process is alive.

import asyncio
from gemini_webapi import GeminiClient

# Replace "COOKIE VALUE HERE" with your actual cookie values.
# Leave Secure_1PSIDTS empty if it's not available for your account.
Secure_1PSID = "COOKIE VALUE HERE"
Secure_1PSIDTS = "COOKIE VALUE HERE"

async def main():
    # If browser-cookie3 is installed, simply use `client = GeminiClient()`
    client = GeminiClient(Secure_1PSID, Secure_1PSIDTS, proxies=None)
    await client.init(timeout=30, auto_close=False, close_delay=300, auto_refresh=True)

asyncio.run(main())

Tip

auto_close and close_delay are optional arguments for automatically closing the client after a certain period of inactivity. This feature is disabled by default. In an always-on service like chatbot, it's recommended to set auto_close to True combined with reasonable seconds of close_delay for better resource management.

Generate contents from text

Ask a one-turn quick question by calling GeminiClient.generate_content.

async def main():
    response = await client.generate_content("Hello World!")
    print(response.text)

asyncio.run(main())

Tip

Simply use print(response) to get the same output if you just want to see the response text

Generate contents from image

Gemini supports image recognition and generating contents from images. Optionally, you can pass images in a list of file data in bytes or their paths in str or pathlib.Path to GeminiClient.generate_content together with text prompt.

async def main():
    response = await client.generate_content(
            "Describe each of these images",
            images=["assets/banner.png", "assets/favicon.png"],
        )
    print(response.text)

asyncio.run(main())

Conversations across multiple turns

If you want to keep conversation continuous, please use GeminiClient.start_chat to create a ChatSession object and send messages through it. The conversation history will be automatically handled and get updated after each turn.

async def main():
    chat = client.start_chat()
    response1 = await chat.send_message("Briefly introduce Europe")
    response2 = await chat.send_message("What's the population there?")
    print(response1.text, response2.text, sep="\n\n----------------------------------\n\n")

asyncio.run(main())

Tip

Same as GeminiClient.generate_content, ChatSession.send_message also accepts image as an optional argument.

Continue previous conversations

To manually retrieve previous conversations, you can pass previous ChatSession's metadata to GeminiClient.start_chat when creating a new ChatSession. Alternatively, you can persist previous metadata to a file or db if you need to access them after the current Python process has exited.

async def main():
    # Start a new chat session
    chat = client.start_chat()
    response = await chat.send_message("Fine weather today")

    # Save chat's metadata
    previous_session = chat.metadata

    # Load the previous conversation
    previous_chat = client.start_chat(metadata=previous_session)
    response = await previous_chat.send_message("What was my previous message?")
    print(response)

asyncio.run(main())

Retrieve images in response

Images in the API's output are stored as a list of Image objects. You can access the image title, URL, and description by calling image.title, image.url and image.alt respectively.

async def main():
    response = await client.generate_content("Send me some pictures of cats")
    for image in response.images:
        print(image, "\n\n----------------------------------\n")

asyncio.run(main())

Generate images with ImageFx

In February 2022, Google introduced a new AI image generator called ImageFx and integrated it into Gemini. You can ask Gemini to generate images with ImageFx simply by natural language.

Important

Google has some limitations on the image generation feature in Gemini, so its availability could be different per region/account. Here's a summary copied from official documentation (as of February 15th, 2024):

Image generation in Gemini Apps is available in most countries, except in the European Economic Area (EEA), Switzerland, and the UK. It’s only available for English prompts.

This feature’s availability in any specific Gemini app is also limited to the supported languages and countries of that app.

For now, this feature isn’t available to users under 18.

async def main():
    response = await client.generate_content("Generate some pictures of cats")
    for image in response.images:
        print(image, "\n\n----------------------------------\n")

asyncio.run(main())

Note

by default, when asked to send images (like the previous example), Gemini will send images fetched from web instead of generating images with AI model, unless you specifically require to "generate" images in your prompt. In this package, web images and generated images are treated differently as WebImage and GeneratedImage, and will be automatically categorized in the output.

Save images to local files

You can save images returned from Gemini to local files under /temp by calling Image.save(). Optionally, you can specify the file path and file name by passing path and filename arguments to the function and skip images with invalid file names by passing skip_invalid_filename=True. Works for both WebImage and GeneratedImage.

async def main():
    response = await client.generate_content("Generate some pictures of cats")
    for i, image in enumerate(response.images):
        await image.save(path="temp/", filename=f"cat_{i}.png", verbose=True)

asyncio.run(main())

Generate contents with Gemini extensions

Important

To access Gemini extensions in API, you must activate them on the Gemini website first. Same as image generation, Google also has limitations on the availability of Gemini extensions. Here's a summary copied from official documentation (as of February 18th, 2024):

To use extensions in Gemini Apps:

Sign in with your personal Google Account that you manage on your own. Extensions, including the Google Workspace extension, are currently not available to Google Workspace accounts for school, business, or other organizations.

Have Gemini Apps Activity on. Extensions are only available when Gemini Apps Activity is turned on.

Important: For now, extensions are available in English, Japanese, and Korean only.

After activating extensions for your account, you can access them in your prompts either by natural language or by starting your prompt with "@" followed by the extension keyword.

async def main():
    response1 = await client.generate_content("@Gmail What's the latest message in my mailbox?")
    print(response1, "\n\n----------------------------------\n")

    response2 = await client.generate_content("@Youtube What's the latest activity of Taylor Swift?")
    print(response2, "\n\n----------------------------------\n")

asyncio.run(main())

Note

For the available regions limitation, it actually only requires your Google account's preferred language to be set to one of the three supported languages listed above. You can change your language settings here.

Check and switch to other reply candidates

A response from Gemini usually contains multiple reply candidates with different generated contents. You can check all candidates and choose one to continue the conversation. By default, the first candidate will be chosen automatically.

async def main():
    # Start a conversation and list all reply candidates
    chat = client.start_chat()
    response = await chat.send_message("Recommend a science fiction book for me.")
    for candidate in response.candidates:
        print(candidate, "\n\n----------------------------------\n")

    if len(response.candidates) > 1:
        # Control the ongoing conversation flow by choosing candidate manually
        new_candidate = chat.choose_candidate(index=1)  # Choose the second candidate here
        followup_response = await chat.send_message("Tell me more about it.")  # Will generate contents based on the chosen candidate
        print(new_candidate, followup_response, sep="\n\n----------------------------------\n\n")
    else:
        print("Only one candidate available.")

asyncio.run(main())

Control log level

You can set the log level of the package to one of the following values: DEBUG, INFO, WARNING, ERROR and CRITICAL. The default value is INFO.

from gemini_webapi import set_log_level

set_log_level("DEBUG")

References

Google AI Studio

acheong08/Bard

Stargazers