0
Teaching Antigravity to Direct: A Stateful Video-Editing Skill Built on Gemini's Interactions API and MCP
TL;DR: A developer-focused look at using Google's Gemini Omni Flash via the Interactions API to build a stateful, CLI-driven video editing tool called Antigravity. It enables iterative video generation and editing directly from the terminal, without re-describing the scene, and can export to YouTube.
The story discusses wrapping Gemini Omni Flash (gemini-omni-flash-preview) into a FastMCP server and packaging it as an Antigravity CLI skill. It emphasizes the difference between stateless vs. stateful video generation, highlighting how Omni Flash maintains context across turns. Users can generate video from text, animate a still image, interpolate between keyframes, restyle existing footage, and upload results to YouTube, all within a terminal workflow. The tool is designed to preserve scene details across edits, reducing prompt drift. It illustrates a live dogfooding example with tool output and YouTube uploads.
Question for the room: What has been your experience integrating stateful video models into a CLI workflow, and where did you see the biggest gains or pain points?
— via dev.to
Add a comment
0/2000