Kohaku BlueLeaf
Shih-Ying Yeh 葉適穎
I am a graduate student doing my master's in the Computer Science Department at National Tsing Hua University in Taiwan.
I am interested in improving the performance of Generative AI, focusing on training better models, building efficient and robust systems, and developing useful tools.
I'm currently working on project: HomeDiffusion. HomeDiffusion seeks to enable pretrained diffusion models on consumer-level hardware. And we already have a usable model: HDM-xut-340M-anime. Which is trained with 1500 GPU hour on 5090.
"AI art should look like AI, not humans."
Recent Projects
LyCORIS
Lora beYond Conventional methods, Other Rank adaptation Implementations for Stable diffusion.
TIPO
Text to Image with Text Presampling for Optimal Prompting.
HDM (Home Diffusion Model)
A series of minimalist T2I model trained on consumer level hardware.
KohakuRAG
#1 in Wattbot2025. Domain-agnostic Retrieval-Augmented Generation (RAG) framework designed for production use.
KohakuHub
An open sourced alternative for HuggingFace, fully compatible with HuggingFace Hub, Transformers and Diffusers.
KohakuBoard
An open sourced alternative for WandB with async logging, efficient storing and fast frontend for visualization.
Recent Publications
TIPO: Text to Image with Text Presampling for Prompt Optimization
ImagenWorld: Stress-Testing Image Generation Models with Explainable Human Evaluation on Open-ended Real-World Tasks
Recent Career
Open Source Contributor
@ Kohaku LabDeveloped and maintained open-source projects related to computer vision and AI. Included personal project such as LyCORIS, TIPO, PixelOE, HakuLatent and more, and community projects such as stable-diffusion-webui, sd-scripts and more.
ML/NN Engineer (Part-time)
@ Comfy OrgCollaborated with comfy team to improve the usability and performance of Comfy UI eco system. Mainly working on lora-related system such as lora loader, weight patcher and trainer.
