Each line becomes one clip: your reference image drives the visuals, the assembled prompt drives the delivery. Generation runs on fal.ai using Gemini Omni Flash image→video. Reference images are required. Every clip spends fal usage on your account key.