baoyu-danger-gemini-web_skill

Overview

This skill provides text and image generation using a reverse-engineered Gemini Web API. It supports single-turn and multi-turn conversations, image generation from prompts, and vision input via reference images. Built for CLI integration, it can act as an image/text backend for other tools or direct user requests.

How this skill works

The skill runs as a TypeScript CLI that interacts with a Gemini Web client implementation. It loads cookies via a browser-based Google sign-in, accepts prompts and optional reference images, and returns generated text or image files. Sessions and preferences are stored on disk to enable multi-turn conversations and persistent settings.

When to use it

• Generate images from text prompts or using reference images for variations.
• Produce text responses or multi-turn chat where Gemini-style output is desired.
• Provide a local CLI backend for other skills that require image generation.
• Batch or script generation tasks via shell commands or npx invocations.
• Test or prototype flows that need vision-capable language model outputs.

Best practices

• Run the consent flow on first use and verify the consent file exists before automating calls.
• Use session IDs for multi-turn flows to preserve conversation state and context.
• Store large numbers of sessions or generated media in a dedicated data directory via environment variables.
• Provide reference images when you need the model to base new images on existing visuals.
• Use the --json flag for machine-readable output when integrating with other tools.

Example use cases

• Generate a product mockup image from a prompt and a reference photo for consistent styling.
• Run an automated captioning pipeline: feed images as references and collect text outputs as JSON.
• Prototype a chatbot that needs Gemini text quality and keeps state across turns via session files.
• Create prompt-driven image variations for creative exploration or A/B testing.
• Scripted batch generation of social media assets using different models and output formats.