Skip to main content
ExploreExplore

Research

Reports (4)

Reproduction study

CoSER: An Independent Reproduction Report

Reproduces CoSER’s literary role-play evaluation. Scores varied substantially with the judge, conversation length, and evaluation run. The leaderboard was withdrawn. This report documents the findings, reproduction differences, and limitations.

Reproduction study

CharacterEval: An Independent Reproduction Report

Evaluates 77 characters from Chinese fiction across 12 criteria using CharacterRM. This report traces the implementation to the authors’ code and data, documents deployment checks, and examines numerical differences from the published results.

Reproduction study

CharacterBench: An Independent Reproduction Report

Documents the CharacterBench setup, from the prompts in Table 11 to context construction and CharacterJudge scoring. Each setting is traced to the paper or official code.

RESEARCH DIRECTIONS

What we’re exploring

We study model architecture, data, and systems through real tasks.

Social intelligence

We study how identity, emotion, beliefs, and relationships shape dialogue. Our work connects social state and event context with character consistency across long conversations.

Explore OmniSoul

Real-time interaction

We study motion control, visual continuity, and audio coordination in live interactions, measuring identity consistency, transitions, and latency.

Explore OmniFlow

Multimodal generation

We explore generation from text, images, video, and audio, including causal chunking and production-grade 4K. We study character performance, scene structure, and complex motion together.

Explore OmniX-1

Efficient inference

We measure quality, latency, and compute cost together, and study distillation, caching, and scheduling for real-time generation.

Explore OmniFlow

Data and self-improving training

We study how dialogue and audiovisual data shape training, using verifiable evaluations to guide large-scale RSI self-improvement.

Explore OmniX-1

Evaluation and reproducibility

We trace data, prompts, judges, and statistical settings to their sources. Reproduction studies document implementation choices, deviations, and limitations so others can review the results.

Explore SociaBench

RESEARCH COLLABORATION

Research collaboration

Work with us