Research

---

in generative media, that world-model is built from explicit modalities and constraints; different modality bundles and coordination produce different coherence signatures, so outputs can look coherent while failing to preserve continuity, identity, scale, or geometric stability the way ecological perception would. the key variable is not the tool; it’s whether structure is being specified or resolved at that step.

my core research question is how each modality supplies different spatial cues and constraint conditions, how those constraint sets interact, and how model architecture (alignment vs separation) determines what spatial structure can stay consistent—especially when human steering functions as an added modality during generation.

method: input modality → constraints → constraint interaction → alignment/separation → architecture-shaped structure → representation → human interpretation.

humans rely on spatial structure to reduce interpretive effort and support action.
teaching people how to reason about inferred worlds reframes imperfections as intelligible outcomes rather than failures, enabling communication, adaptation, and intentional intervention in hybrid human–AI systems.

deliverable: a repeatable casework framework for inferential normalization and world-model shaping that links a system’s modality bundle and coordination to the coherent-enough spatial hypothesis it constructs—and to the interpretations humans reliably take from it in hybrid environments. it includes a documentation workflow (inputs → human interventions → breakpoints → constraint dynamics mapped) that makes hybrid continuity observable, comparable, and archivable over time.

track A: specified-heavy workflows (explicit 3D authoring + ai modifiers)

track B: resolved-heavy workflows (generative worldbuilding)