CineSim

Prototype for simulating cinematographic audiences through structured profiles and multimodal artificial intelligence

CineSim questions the relationship between cinema and automatic perception systems. The starting point is a practical question that crosses contemporary audiovisual production: how can a creative team anticipate the reception of a cut before subjecting it to a real audience? The prototype proposes a small-scale, operational response open to experimentation, built on spectator profiles defined in a JSON file and a video fragment uploaded by the user. The system analyzes the audiovisual content and generates personalized spectator reactions in real-time as playback progresses.

The technological core of the system is Gemini Flash Lite, Google’s multimodal model with a transformer architecture optimized for visual understanding and text generation tasks at low latency. The system performs an exhaustive analysis of the video in five structured dimensions: technical specifications like duration, resolution, and aspect ratio; narrative synopsis with an executive summary and detailed action; visual aesthetics with color palette, lighting, shot types, and editing; character dynamics with protagonists and relationships; and general emotional context. Based on this analysis, the model batch-generates each spectator’s thoughts at regular intervals during playback. Each spectator profile is a JSON structure with demographic variables, genre preferences, thematic sensitivities, and emotional disposition. The visualization is constructed by placing spectators in a navigable three-dimensional cinema hall with a perspective camera and orbit controls. A statistics panel records the engagement level per spectator over time and presents it as a bar graph.

The audience crisis in the film sector is not just a problem of distribution or platforms. It is also a problem of knowledge. Production teams make decisions about cuts, narrative rhythm, and emotional construction with very limited information about how different audience segments will process the material. Real-scale audience testing tools are costly, late, and inaccessible for independent projects. CineSim does not intend to replace any real audience or function as a statistical prediction system. Its value lies in allowing those who create a film to provisionally inhabit the viewpoint of specific profiles, convert those perspectives into legible text, and use that text as working material during the creative process. The question the prototype opens is cultural and epistemological: what it means to construct a spectator as data, and what that gesture reveals about how we think about cinematographic reception.

The prototype is designed to be remixed and adapted. Any team can load their own audience profiles, defined according to variables the project considers relevant, and execute the simulation on a cut in progress. Working with a few well-constructed profiles produces more legible and useful results for creative reflection than simulating mass statistics.

Application (requires uploading previous audiences in JSON):