Claude Drives Gemini

claude-drives-gemini

English | [简体中文](README.zh-CN.md) | [日本語](README.ja.md) | [한국어](README.ko.md)

A Claude Code skill: let your AI agent drive the Chrome where **your Google AI Pro account is already signed in**, and generate **images (Nano Banana, 2816×1536) / video (Veo) / music (Lyria)** through the Gemini web app. Output lands on disk losslessly, and media bytes never enter the agent's context.

**Make the Google Gemini subscription you already pay for do extra work.**

Features

• One command, end to end: auto-launch the dedicated Chrome → ensure Pro + Extended Thinking → submit the prompt → stage-aware wait → native download → optional watermark removal, returning a single line of JSON.
• Never burns quota twice: only an explicit refusal triggers a safe resubmit; a timeout only extends the wait; a failed download only retries the download.
• Full resolution, lossless: the native 2816×1536 original is downloaded directly; --dewatermark removes the visible corner logo with exact math (SynthID excluded — see the risk section).

Prerequisites

• Node ≥ 22 + Chrome + a Google AI Pro (or higher) subscription — the consumer web subscription, not an API key.
• Verified on Windows; the macOS / Linux scaffolding is in place but untested (feedback welcome).
• A local proxy if your network cannot reach Google directly.
• UI text matching is verified against the Chinese Gemini interface; locales/en.json is an unverified best guess — calibration reports are very welcome.

Quick start

**Recommended: let Claude Code install it for you.** Paste this to Claude Code:

Install claude-drives-gemini for me: clone https://github.com/MagicYangG/claude-drives-gemini into `~/.claude/skills/claude-drives-gemini`, run `node setup/init.mjs` to generate the local config, walk me through the one-time seeding login, then run the start-chrome launcher and verify with `node scripts/smoke.mjs`.

Manual install is three steps:

git clone https://g