ai-multimodal

Transcription with timestamps processes multimedia content through Google Gemini's multimodal API to extract spoken content and temporal markers. Use the gemini_batch_process.py script with the transcribe task parameter to handle audio and video inputs, producing JSON or Markdown output compatible with downstream RAG systems.

Document extraction applies Gemini's multimodal analysis to PDFs and Word files via the document_converter.py script. The Skill processes documents through OCR and structured extraction, outputting JSON or Markdown for data retrieval and integration into retrieval-augmented generation workflows.

Image analysis uses Gemini's multimodal API for captioning and object detection across batch-processed image files. The Skill supports inline data and File API input methods, producing structured insights in JSON format suitable for downstream processing and storage.

Video analysis, including YouTube inputs, leverages Gemini to perform scene and temporal analysis. The Skill chunks long media content for efficient processing and outputs structured scene-level insights in JSON or text format for retrieval systems.

Cost-aware processing manages billing through the Skill's ecosystem, which includes free-tier fallbacks and long media chunking strategies. Configure input methods (inline data vs File API) and select appropriate models (Gemini 2.5/3, Imagen 4, Veo 3) to optimize token usage and expense.

Image and video generation integrates Imagen 4 and Veo 3 models within the multimodal workflow for content creation and editing. The Skill orchestrates generation alongside analysis and extraction tasks, producing assets ready for downstream retrieval and RAG integration.