INTEGRASI GENERATIVE VIDEO DALAM WORKFLOW PRODUKSI VFX FILM PENDEK “SPLIT//SECOND”

The rapid advancement of generative video technology presents novel opportunities for visual effects (VFX) asset generation, yet full reliance on artificial intelligence often creates visual inconsistencies, reduced artistic control, and integration barriers with live-action footage. This article aims to analyze the integration of generative video into the VFX post-production workflow of the short film “SPLIT//SECOND” and evaluate its effectiveness relative to conventional computer-generated imagery (CGI) practices. Employing a practice-based research methodology, the production applied a hybrid workflow where live-action footage serves as the physical reality anchor while artificial intelligence assets construct surreal virtual reality (VR) elements. Empirical results demonstrate that generative video substantially accelerates visual asset acquisition with up to 95% time efficiency, reducing manual tasks from 60–90 minutes to approximately 30 seconds to 3 minutes. Although it eliminates manual modeling labor, this method does not remove the essential requirements of digital compositing, rotoscoping, tracking, and color matching, while also introducing variable operating costs driven by platform credit consumption. ABSTRAK Perkembangan teknologi generative video membuka peluang baru dalam penyediaan aset visual effects (VFX), namun penggunaannya secara mandiri kerap memicu kendala konsistensi visual, penurunan kontrol artistik, dan hambatan integrasi dengan rekaman live-action. Artikel ini bertujuan menganalisis integrasi generative video dalam workflow produksi VFX pada film pendek “SPLIT//SECOND” serta membandingkan efektivitasnya dengan pendekatan computer-generated imagery (CGI) konvensional. Penciptaan karya menerapkan metode practice-based research melalui hybrid workflow yang menempatkan rekaman live-action sebagai jangkar realitas fisik dan elemen berbasis kecerdasan buatan (artificial intelligence / AI) sebagai pembangun dunia virtual reality (VR). Pengujian empiris membuktikan bahwa pemanfaatan generative video mampu mempercepat penyediaan aset visual secara signifikan dengan efisiensi waktu mencapai 95%, yakni dari durasi manual sekitar 60–90 menit menjadi hanya 30 detik hingga 3 menit. Kendati memangkas tahapan teknis permodelan manual, pendekatan ini tidak mengeliminasi kebutuhan digital compositing, rotoscoping, tracking, dan color matching, serta menimbulkan konsekuensi beban biaya variabel berbasis konsumsi credits.