NON MI TROVI - SHIKOBA | AI MUSIC VIDEO

PROCESS

I developed *NON MI TROVI* as a reference-driven AI production rather than generating isolated clips and assembling them afterwards. I first created the original song with Suno, then broke it into timed sections to plan the performance, coverage and editing rhythm. I designed a consistent visual library for the singer, wardrobe, glasses, piano, cigarette, wolf and location. Every asset was numbered and assigned a specific role so it could be referenced consistently across generations. The main technical challenge was creating a believable performance that was never physically filmed. The singer had to articulate fast Italian lyrics while maintaining natural facial expressions, head movement, breathing, body weight and continuous interaction with the piano. I generated different camera angles using Dreamina Seedance 2.5, often extracting approved frames from previous clips to establish the correct identity, lighting, background and camera position for the next shot. Magnific was used to develop and refine character, prop and environment references. Topaz Labs supported selected enhancement and upscaling. I manually reviewed every generation for performance, anatomy, perspective, geography, lighting and continuity, then selected, timed and restructured the material in CapCut. The final result came from repeated directing, testing and editorial decisions—not from a single prompt. My role covered concept development, original music, visual direction, reference design, shot planning, prompt direction, performance supervision, generation selection, editing, colour treatment and final delivery.