Agnes AI Image and Video Generation

Problem It Solves

In multimodal agent workflows, image and video generation often require assembling POST /v1/images/generations, POST /v1/videos, polling task status, and handling Agnes AI authentication, parameter limits, and response field quirks. This skill wraps API key configuration, base URL, and model calls into a usable workflow, reducing boilerplate for HTTP requests and status handling.

Core Capabilities and Workflow

• Text-to-image: uses agnes-image-2.1-flash with /v1/images/generations for prompt-based image generation.
• Image editing: uses agnes-image-2.0-flash, passing one or more reference images via extra_body.image, requires tags: ["img2img"], and supports a fixed seed for reproducible results.
• Video generation: uses agnes-video-v2.0 as an asynchronous task; poll /v1/videos/{task_id} after creation. num_frames must be 8n+1 and no greater than 441, commonly 81, 121, 241, or 441.
• Status handling: lifecycle moves from queued to in_progress to completed / failed; when completed, read the video download URL from remixed_from_video_id.

Boundaries

This skill relies on the free Agnes AI API, so it may hit RPM limits, video queuing, or 500/502/503 errors during peak periods. Store the API key in environment variables or configuration, and avoid committing it to Git.