Teaching Antigravity to Direct: A Stateful Video-Editing Skill Built on Gemini's Interactions API and MCP

Short summary

The author built omni-skill-agy, a FastMCP server and Antigravity CLI skill that wraps Google's Gemini Omni Flash model for stateful video generation and editing. Unlike stateless workflows, the Interactions API preserves visual context across turns, enabling incremental edits like 'make it nighttime' without re-prompting the entire scene. The tool supports five generation modes (text-to-video, image animation, keyframe interpolation, subject-consistent generation, and video restyling) plus YouTube upload, all from the terminal.

• •Built an MCP server wrapping Gemini Omni Flash for stateful video generation via Antigravity CLI
• •Interactions API preserves visual context across turns for iterative editing without re-prompting
• •Supports five multimodal generation modes plus YouTube upload from the terminal

Generated with AI, which can make mistakes.