Research

A lab for cinematic intelligence.

Palindromix.Studio is a research-driven company. We work on the systems that let creators direct generative models — with the precision of cinema and the iteration speed of software.

Cinematic previs scene with camera blocking diagrams

Position

Generation without direction is novelty. Direction without generation is friction. Palindromix unifies them.

The next decade of visual storytelling won't be won by larger video models alone. It will be won by the interfaces, representations, and control systems that sit between the creator and the model — what we call cinematic infrastructure.

Our research targets the missing middle: spatial scene understanding, camera-aware conditioning, and persistent world representations that make AI output directable, editable, and reusable.

Active threads

T-01ONGOING

Immersive scene intelligence

Decomposing narrative text into structured scene graphs — entities, geometry, intent, and continuity constraints — robust enough to drive downstream generation.

T-02ONGOING

Spatial narrative systems

Mapping story structure onto navigable 3D space, so beats, blocking, and camera coverage become first-class authoring primitives.

T-03ONGOING

Camera-directed AI generation

Conditioning generative video on lens, blocking, lighting, and approved keyframes — moving the field from prompts to direction.

T-04ONGOING

3D Gaussian splat workflows

Photoreal capture-to-edit pipelines for environments and props, treated as editable cinematic assets, not static scans.

T-05ONGOING

Browser-native visualization

Three.js-based real-time previs that runs anywhere, collapsing the gap between concept, blocking, and delivery.

T-06ONGOING

Reusable cinematic environments

Worlds as long-lived assets — versioned, composable, and portable across projects, models, and pipelines.

Publishing

Research log, technical notes, and case studies — coming soon.

We share working notes, evaluation methods, and field studies with our research partners and waitlisted creators first.