1 / 11
Identidad Fluida is an interactive installation that transforms the viewer's face in real time using generative AI. A webcam captures the participant's face; MediaPipe isolates facial landmarks and feeds them into a StreamDiffusion pipeline that continuously re-imagines the face. The viewer witnesses their own face flow, dissolve, and mutate in front of them.
The artwork explores the relationship between open source ai, physical identity and its digital representation. Each viewer is confronted with the same question the piece asks: who are we, the mirror or the own algorithm? By making the transformation visible and participatory, Identidad Fluida proposes identity not as a fixed image but as a living, computational process. The audience is asked to stand in front of the system, see their face drift through styles, color fields, and deformations, and reflect on what remains of the "self" when every frame is rebuilt by a local model.
Inputs: live webcam feed (HD, indoor lighting), optional on-screen prompt selection.
Components: - MediaPipe Face Mesh: 478 facial landmarks extracted per frame, isolating eyes, mouth, jaw, and contour. - StreamDiffusion pipeline (SD-Turbo): a real-time diffusion model accelerated on a local NVIDIA GPU. Runs locally, no cloud, no data exfiltration. - Prompt conditioning: a curated set of prompts (identity drift, color fields, glitch, painterly styles) cycled or triggered by user input. - Output compositor: blends the transformed face with the live capture to preserve presence.
Process: each frame, the webcam image is preprocessed, fed to MediaPipe for landmark extraction, and routed to the StreamDiffusion pipeline. The pipeline generates a stylized re-imagining of the face at near-real-time latency. The transformed image is composited back to the display, projected at 1:1 scale so the viewer sees themselves "becoming" in real time.
The system is designed for data sovereignty: all inference runs on the local machine, the camera feed is never stored, and prompts are seeded locally. The art script is single-file, deterministic from its seed, and reproducible.
Petra (Luis Betancourt Nuñez)
Kinetic AI, Microversos Ludografías Transmedia, 8ADN Cuerpos en Acción
Petra es artista de nuevos medios cuya obra se sitúa en el cruce entre danza, arte digital y tecnología. Investiga, experimenta y crea sistemas coreográficos experimentales donde el movimiento humano dialoga con procesos algorítmicos y entornos interactivos. Formadx en danza/coreografía y con una práctica sostenida en la programación creativa, su trabajo explora nociones de coreografía expandida, danza generativa y meta-improvisación. A través de performance, animación y video-danza-experimental, Petra construye paisajes sensibles donde el cuerpo funciona como interfaz y archivo vivo. Su obra ha sido presentada en contextos internacionales de arte contemporáneo y nuevos medios, incluyendo el Museo de Arte Moderno MX, Salón de la Plástica Mexicana, CVPR Art Gallery, Mad Arts, ASVOFF, Miami Art Week, Superchief Gallery, Free The Kitsch Festival, London Fashion Week, NFT NYC, NFT Festa, Plexus Projects, así como en plataformas de investigación artística y cultura digital. www.petrasynthetic.com
