gemini-vision

The Gemini Vision API skill provides a robust toolkit for integrating Google’s advanced multimodal AI into your agent workflows. Designed for developers and engineers, this skill enables automated visual understanding, allowing agents to interpret, classify, and manipulate image and document data programmatically. It leverages both Google AI Studio and Vertex AI endpoints, offering flexibility in deployment environments from local development to production-scale cloud infrastructure.

• Advanced Image Understanding: Perform automated captioning, image classification, and visual question answering (VQA).
• Precise Spatial Awareness: Utilize Gemini 2.0+ models for object detection with bounding boxes and Gemini 2.5+ for pixel-level image segmentation.
• High-Volume Document Processing: Ingest and analyze PDF documents up to 1,000 pages, extracting structured insights from diagrams, tables, and text.
• Multi-Image Analysis: Compare and analyze up to 3,600 images in a single request, ideal for change detection and batch visual processing.
• Flexible API Configuration: Supports multiple authentication layers and environment-based configuration for secure, secret-managed API integration.
• Scalable Model Selection: Choose between specific model variants such as Flash-Lite for speed or Pro models for maximum visual reasoning capability.

Usage notes include configuring the GEMINI_API_KEY through the provided hierarchy of .env files and utilizing the File API for large images or files exceeding 20MB. Token usage is calculated based on image tiling (768x768 units), and users are encouraged to monitor usage via Google Cloud Console. Input formats supported include PNG, JPEG, WEBP, HEIC, and PDF. Ensure that your prompts follow best practices by providing few-shot examples and specific formatting instructions for JSON or Markdown outputs to maximize accuracy in automated tasks.