zhouwei713/fpv-immersive-video-prompting

fpv-immersive-video-prompting is a skill (a reusable instruction module) for AI agents that helps you write better camera-movement prompts for AI video generation tools like Seedance, Kling, Runway, and Veo. The documentation is written in Chinese with an English README also included. "FPV" here refers to first-person view camera movement -- the perspective of a camera that travels through a scene, passes characters or objects, and lands at a destination, similar to a drone or low camera sweeping through a space. The core problem this skill addresses is that AI video models often produce broken results when given vague movement descriptions. Common failures include the camera jumping suddenly, extra characters appearing or disappearing, numbered labels from a reference image leaking into the final video, the wrong camera height, or the path going through walls and furniture. The skill argues that these failures usually happen because the prompt described a visual scene rather than a planned route. The skill structures every FPV request around eight questions: who is the camera, where does it start, how many subjects are in the scene (exactly), what order does it pass them in, is each leg of the path physically possible, what interaction happens at each stop, what must stay consistent, and what must never appear in the final frame. From those answers it generates a full video prompt. It also creates a multi-image asset pack for use with GPT Image: a numbered reference frame showing character positions, one standalone reference image per character, and a clean version of the opening frame with all labels removed. Keeping the path planning and the character references in separate images reduces confusion for the video model. Installation works by copying a SKILL.md file into Hermes, Claude, Codex, Cursor, or other agent environments that support long instruction files.