ABOUT

EXHIBITION

Horizons of Synesthesia

CONFERENCE

Palimpsest of Reality

FILM

Echoes of Memories

ARCHIVE

Development of a Virtual Casting Service (Pre-Cast AI) Utilizing Face-Swap and Voice-Conversion Algorithms in the Previsualization Stage

Speaker

GSAIM AI Team 3
| Gilwoo Jeong, Heejae Jo, Junhyeok An | Korea [Team 3]
Gilwoo Jeong is currently enrolled in the Graduate School of Advanced Imaging Science, Multimedia & Film at Chung-Ang University, majoring in Film and Media Production. In graduate school, he directed a film titled Molted Shell, and his AI-CFT research project, titled Pre-Cast AI, aims to develop a virtual casting service for the previsualization stage of film production.

Heejae Jo is a master’s researcher at the VILab, Graduate School of Advanced Imaging Science, Multimedia and Film, Chung-Ang University,She studies Deepfake detection and Trustworthy AI. In the AI-CFT project,she is developing virtual casting service technologies utilizing AI-based Face Swap and Voice Conversion methods.

JunHyeok An is Currently enrolled in the Film Directing & Production major at the Graduate School of Advanced Imaging Science, Multimedia & Film (GSAIM), Chung-Ang University. He directed the graduate workshop film and is currently working on the AI-CFT project, aiming to develop 'Pre-Cast-AI,' a virtual casting service for the pre-visualization stage.

Abstract

This study aims to develop a virtual casting service that utilizes Face Swapand Voice Conversiontechnologies during the previsualization (previs)stage of film, drama, and commercial production. For this purpose, Roopis primarily used for facial synthesis, while Coqui.aiand other TTS (Text-to-Speech)algorithms are applied for voice synthesis.
The proposed service generates realistic previsualization videos by compositing the target actor’s face, expressions, performance, and voice onto a stand-in actor or animatic footage based on real facial and vocal data. This allows production teams to efficiently evaluate casting decisions, reduce production costs and time, and explore diverse actor combinations across global production environments.
All synthesis processes are conducted only with prior consent from the performers, and the system explicitly indicates that the generated content is synthetic. Furthermore, data security and ethical complianceare prioritized to minimize potential legal and moral risksassociated with the use of synthetic media.

List