03Projects
Projects
Research projects and systems the lab built itself. Academic and teaching activities are on the Activities page.
R&D
Research projects · systems
-
GENERATIVE AIOngoing
Improving Generative AI Models for Cultural Content Production
Virtual Convergence Technology Research Institute R&D · 2025 –
Raising the quality and controllability of generative models used in cultural content production pipelines: conditional control and editing methods for image and video generation.
-
GENERATIVE AIOngoing
Generative AI for Content Production Workflows
Industry–academia joint R&D · Dongseo University with Themantastory · 2026 –
A three-year R&D programme, based on an industry needs survey, applying generative AI to content production. Aims at on-site use of image and video generation models and follow-on government funding.
-
LLM · AGENTSOngoing
Local LLM Multi-Agent System
Lab project · 2026 –
Language-model agents that do not depend on the cloud: four agents — orchestrator, research, development, and assistant — run on local models and connect through a messenger. Open-weight model selection and serving optimisation are the core.
-
GENERATIVE AIOngoing
ComfyUI Content Generation Pipeline
Lab project · 2026 –
Text-to-image, video, and speech generation plus lip-sync video production in one workflow. Runs headless on a remote GPU server, with long-audio segmentation and Korean speech synthesis.
-
INFRASTRUCTURECompleted
Lab AI Server
Lab project · 2026
An open-weight LLM and generative-model serving environment on a GPU server. Coding agents connect to local models through vLLM and Ollama; lab members access it remotely.
RESULTS
Research results
Results backed by published papers.
-
GENERATIVE MODELS
Pixel-level control and dual attention on the MMDiT architecture edit images without additional training.
Related paper · The Journal of Korean Institute of Communications and Information Sciences -
GENERATIVE MODELS
A GAN pipeline that generates and controls Asian facial textures using skin condition maps.
Related paper · Journal of the Institute of Convergence Signal Processing -
LLM & NLP
Korean text generation that summarises news trends with TF-IDF topic guiding and a fine-tuned KoBART.
Related paper · Journal of the Institute of Convergence Signal Processing -
LLM & NLP
Tokenising the deltas of time-series data so a language model can predict human motion.
Related paper · Journal of Artificial Intelligence Convergence Technology -
TOOLS & AGENTS
A reinforcement-learning agent that optimises disinfectant dosing while guaranteeing water quality safety.
Related paper · The Journal of the Korea Institute of Electronic Communication Sciences -
TOOLS & AGENTS
Reinforcement-learning control that optimises full-charge timing to user patterns and prevents overcharging.
Related paper · Journal of the Institute of Convergence Signal Processing -
GENERATIVE MODELS
Combining CBAM attention with CycleGAN to improve day-to-night image translation quality.
Related paper · Journal of the Korea Institute of Information and Communication Engineering -
TOOLS & AGENTS
A reinforcement-learning agent that adjusts game difficulty dynamically to player behaviour.
Related paper · Journal of the Institute of Convergence Signal Processing -
MULTIMODAL & VISION
A framework and authoring tools that estimate motion from MYO EMG sensors with Bayesian probability.
Related paper · Multimedia Tools and Applications