게임엔진 영상을 활용한 생성형 AI 영상 제작 워크플로우 연구
A Study on a Generative AI Video Production Workflow Utilizing Game Engine–Based Footage
2
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
While generative AI's Video-to-Video technology has rapidly advanced, challenges remain in video production requiring complex camera movements and precise angle control. This study proposes a workflow that validates precise camera control by inputting reference videos created with Unity game engine into Video-to-Video generative AI. We designed five camera control missions (120⁰ rotation, vertical crane shot, compound dolly zoom, spiral orbit, and multi-angle transition) and created reference videos using Unity for each mission, then generated final videos using Runway Gen-3's Video-to-Video functionality. Focus group interviews with 5 video production experts showed that Unity references accurately conveyed camera movements across all missions (average 4.6/5 points), maintaining high accuracy (4.5 points) even in complex spiral orbit missions. Experts identified frame-level precise control, ease of iterative modification, and predictable results as key advantages, with all participants expressing intention to apply this in professional practice. This study empirically demonstrates that game engine-based reference generation is a practical solution to camera control challenges in generative AI video production.
Abstract
While generative AI's Video-to-Video technology has rapidly advanced, challenges remain in video production requiring complex camera movements and precise angle control. This study proposes a workflow that validates precise camera control by inputting reference videos created with Unity game engine into Video-to-Video generative AI. We designed five camera control missions (120⁰ rotation, vertical crane shot, compound dolly zoom, spiral orbit, and multi-angle transition) and created reference videos using Unity for each mission, then generated final videos using Runway Gen-3's Video-to-Video functionality. Focus group interviews with 5 video production experts showed that Unity references accurately conveyed camera movements across all missions (average 4.6/5 points), maintaining high accuracy (4.5 points) even in complex spiral orbit missions. Experts identified frame-level precise control, ease of iterative modification, and predictable results as key advantages, with all participants expressing intention to apply this in professional practice. This study empirically demonstrates that game engine-based reference generation is a practical solution to camera control challenges in generative AI video production.