Agnes Video V2.0

Problem

Calling Sapiens AI's Agnes Video v2.0 is less about prompt writing and more about async task handling and parameter constraints. The API does not return a video file synchronously; it creates a task, requires polling by task_id, and exposes the final video URL in remixed_from_video_id. If your integration assumes request-then-result behavior, it can easily stall on queued, in-progress, or failed states.

How It Works

The skill wraps the agnes-video-v2.0 endpoints: POST /v1/videos creates a task, and GET /v1/videos/{task_id} checks progress. It supports four workflows:
- Text-to-video: use prompt to generate a short clip from a scene description
- Image-to-video: pass image to animate a still frame
- Multi-image transition: use extra_body.image to move between multiple images
- Keyframe animation: set extra_body.mode to keyframes for more explicit frame-level control

Key parameters include height, width, frame_rate, seed, negative_prompt, and num_frames. num_frames must satisfy 8n+1 and not exceed 441, while frame_rate determines duration and motion cadence; lowering frame_rate can extend length but changes pacing.

Scope And Notes

Best suited for scripts, CLI tools, or automation pipelines that need to submit video generation jobs. Handle task states queued, in_progress, completed, and failed; read remixed_from_video_id when completed and inspect error on failure. Auth relies on an API key, and free-tier or parameter limits should inform retry and monitoring logic.