Esther Olofsson

The strongest evidence comes from the people who make her. In its December 2022 press release, RAUWcc called Esther Olofsson 3.0 the first Dutch AI-generated virtual human. The agency said it trained Stable Diffusion on a full 3D model of her and inpainted her into royalty-free photos of real places. Movio animates the stills and handles the lip-sync, Replica Studios generates her voice and ChatGPT writes her captions. RAUWcc says all it still does is direct the story, write the prompts and post. Emerce reported the same shift to a fully AI-driven model. Director Maarten Reijgersberg, who calls himself her spiritual father, presented it at Immersive Tech Week.

She was not always generated end to end. RAUWcc's June 2022 release says that until then she was built by applying CGI and deepfake technology to photos or footage of an existing model. So between 2020 and 2022 a real person's body was probably under the face. That model has never been named, and from version 2.0 onward she was built entirely in 3D, with no model on location. No page names a human who performs or voices her today.

VirtualHumans.org describes her as a fictional character with an invented backstory and lists her as inactive or discontinued. Still unknown: who that early model was, and whether anyone at RAUWcc still runs the newer @estherolofssonexplores account day to day.