Project Yoru & Fami

Project Yoru and Project Fami are ongoing stealth research efforts exploring novel architectures and scaling infrastructures for generative vision and multimodal audio synthesis.

Research Focus Areas

🌌 Project Yoru (Vision & Image Generation)

• High-Fidelity Flow Map Formulation: Exploring continuous and decoupled flow matching formulations for accelerated, high-fidelity text-to-image synthesis.
• Architectural Exploration: Designing scale-wise transformer backbones and representation autoencoders (RAE) to eliminate traditional VAE artifacts.
• Dataset Infrastructure: Developing automated high-resolution image filtering, aesthetic scoring, and synthetic captioning pipelines.

🎙️ Project Fami (Audio & Speech Synthesis)

• Generative Audio Framework: Researching continuous flow matching and diffusion-based acoustic decoders for natural speech and audio synthesis.
• Data Engineering & Alignment: Constructing clean, multi-speaker conversational speech corpora with fine-grained prosody and acoustic feature extraction.