Independent AI research · North Carolina

Character research.
Measurable control.

We study how AI characters can be understood, controlled and evaluated. Our work began with facial-feature analysis in anime-style imagery and continues into broader character systems.

Explore the research Open the demo
RESEARCH STUDY / 001 PARAMETRIC RETRIEVAL
Selected character frame at head X minus 0.60
head_x −0.60
Selected character frame at head X zero
head_x 0.00
Selected character frame at head X plus 0.60
head_x +0.60
One parameter, three requested positions.Actual frames returned by the research demo. Labels show requested coordinates; retrieval is bounded by the available index and extraction accuracy.
153,451 frames in the
public research index
2 character categories
in the demonstration
Jan 2026 technical report
and validation study

Research areas

From visual features
to useful control.

We investigate the connection between a character image, its measurable properties and the systems that act on those properties.

01 / ANALYSIS

Reading visual features

Studying how head pose, eye gaze and expression can be represented as measurable parameters in stylized character imagery.

02 / CONTROL

Making selection controllable

Exploring coordinate-based retrieval and the relationship between a requested visual state and the frames a system returns.

03 / VERIFICATION

Evaluating what a system produces

Investigating visual consistency, extraction limits and methods for comparing generated outputs against intended parameters.

Published work

A starting point you can inspect.

Our public work includes the underlying approach, an interactive example, and an account of what the measurements do and do not establish.

Research figure showing the distribution of indexed head and eye coordinates

Technical report · January 2026

Selection over Generation

A proof of concept for retrieving character frames through head-pose, eye-gaze and expression coordinates. The report describes the extraction approach, internal benchmarks and known limitations.

Read the report View validation

The public study is a research demonstration. Historical benchmark figures describe their stated test conditions, not a guarantee of current service performance.

The lab today

Building on the research.

We continue to develop controllable character technology and the methods used to evaluate it. Our first product is in private development.

We share selected research here while keeping ongoing product work private.

About Hollywood Reborn

Research infrastructure

Compute for exploration.

We have secured cloud credits from Amazon Web Services, Microsoft Azure and Google Cloud, and use them to support research and development.

Our research uses NVIDIA GPUs for compute-intensive experiments. This support helps us investigate ideas, evaluate results and develop our own technology.

Amazon Web Services Microsoft Azure Google Cloud

Research is a conversation.

We welcome thoughtful discussions about the work.

Contact the lab