Cowpokes riding through the ghost town of an abandoned Western. Isolated cabin, solitary frames, staring out onto the flickering plains of a mythology fading to red. A crossroads. Two mediums, revolvers drawn, caught in a deadly standoff.
Cf. Dead Crossing installation (YouTube). Camera: Faust Perillaud. Edit: Douglas Edric Stanley
Playable Cinema is a research project exploring how artificial intelligence can ride the invisible frontier between cinema history and interactive gameplay. By using machine learning to analyze classical western cinema, the project constructs a generative database where fragments of film history and the streams of live gameplay bleed into one another; where the joystick becomes an editing tool for an infinite fever dream looping through the ghosts of cinematic history.
The project began by building a dataset that connects patterns between dystopian cinema and the award-winning indie game Inside (Playdead, 2016). This installation has been exibited in various locations and contexts (San Francisco, Lausanne, Marseille, Genève, …). This testbed served as a prototype for a larger-scale dataset comparing Western films and the iconic western video game, Red Dead Redemption 2.
Dead Crossing is the latest physical manifestation of the Playable Cinema research project. Inside of an unfinished wooden cabin, visitors explore the video game world of Red Dead Redemption 2. As they move through its three-dimensional virtual terrain, AI models synchronize their movements with shots echoing from the history of Western cinema.
In this current form, the project explores how hybrid strategies can emerge new curatorial methodologies as generative tools collaborate with humans in assembling the haunted archive of our shared hallucinations of the West.
An exhibition document explaining the installation Dead Crossing can be found here: dead-crossing-exhibition-document (PDF)
- Douglas Edric Stanley, Project Lead, concept, design, development, production, documentation
- Faust Perillaud, Research Assistant, training & labelling, production, documentation, photography
- Guillaume Stagnaro, Cowpoke Controller development
- Colin Castellano, Wood construction consulting & production
A training and playback tool is currently in development.
You can watch a short excerpt of the software synchronizing in real-time images from gameplay and shots from historical western cinema:
Sync Demo - Train Robbery (https://youtu.be/-g9P9GaXHlI)
Sync Demo - Carriage Entering Stables (https://youtu.be/cOG3Zf-KX_0)
A cabin has been constructed that integrates all the parts of the project.
Photo by Faust Perillaud.
Blueprints to build this cabin can be found in the repository hardware/cowpoke-cabin
A physical game controller is currently in developement. This allows visitors to interact with the installation.
CAD model files and KiCAD circuit diagram can be found in hardware/cowpoke-controller
There is a list of the western films we are using to train the deep learning models of this project.
- Test Phase (Q2 2025): Annotating sequences from Inside using Roboflow and SAM tools.
- Training Prototype Model (Q3 2025): Building initial inference pipeline and testing labeling workflow.
- Scaling Up (Q3 2025): Transitioning to a larger dataset involving Western films and Red Dead Redemption 2.
- Deployment (Q4 2025): Developing interactive installation, annotation tool and real-time inference API.
- Anthony Masure, Dean of Research, IRAD, HEAD – Genève, HES-SO
- Christelle Granite-Noble, Administrative Coordination, IRAD, HEAD – Genève, HES-SO
- Valentin Dubois, HEAD – Genève, Head of Materials + Prototyping Pool
- Alexandre Simian, Technician Wood Workshop, HEAD – Genève
- Sébastien Pitteloud, Technician Wood Workshop, HEAD – Genève
- Charles Cuccu, Régisseur HEAD – Genève
This project was financed with a research grant from the Network of Expertise in Design and Visual Arts / Réseau de compétences Design et Arts visuels.
