IAMBARRIA • Reflex

Reflex / The Poetic Life of a Bolex H Brochure —
Stochastic Music Video

Subtext Mask / Synthetic Casting Framework —

A visual research project exploring fictional actor generation, identity continuity, and controlled resistance to generative homogenization.

The idea behind was a simple question: can a synthetic face be directed away from internet-glamour conventions?

Asking a model for a “natural realistic face” is not enough. Terms like natural, realistic, cinematic, or beautiful are vague. Generative image models tend to resolve them through familiar visual conventions. The question became less about prompting and more about defining the right constraints. How can a fictional face hold a close-up without becoming overdesigned, generic, or emotionally obvious?

The answer came through film history, personal references, performer studies, anatomy, gesture, semiotics, and casting logic. The goal was not beauty standards or nostalgia. The goal was dramatic restraint: a face that can hold the camera, where the close-up becomes a place of withholding rather than explanation.

The process involved text-to-image visual exploration, mood descriptions, era mixtures, casting descriptors, and image-remix experiments. No intentional likeness references of living or deceased individuals were used at any stage. Selected outputs became anchor portraits, then were translated into functional casting headshots through visual identity analysis, headshot refinement, expression testing, wardrobe control, and continuity correction.

The first refinements often drifted toward generic AI-image styling. To correct that, I developed an iterative feedback process focused on specificity: facial structure, eye behavior, skin texture, hair identity, expression discipline, photographic credibility, and resistance to cosmetic homogenization.

That process eventually became a set of instructions, rules, and source documents designed to govern the work. The framework was built from anatomical, photographic, and casting-based criteria rather than style labels alone. The goal was not simply to ask for a “look,” but to use precise visual descriptions to condition, evaluate, and correct the output.It was never intended as an automated system for producing casting sheets. The point was the opposite: to test how much human judgment is still required when working with generative images.

There is a lot of discussion about what tools can do, but far less about the criteria used to direct them. This project is about that missing layer: how images are evaluated, constrained, refined, and made visually coherent.

The result is less a fixed formula or tool than a framework based on judgment and curation. The value is not the tool. The value is not the tool. The work has to remain tool-agnostic. What matters is the mindset that precedes the machine: the ability to evaluate, constrain, refine, and decide what an image is doing.

Midjourney Explorations