πŸš€ Update News

  • 2026-03-05: Official release of KORMo-Diffusion.
  • 2026-03-02: Official release of KORMo-VL.
  • 2025-10-13: Official release of KORMo-10B-sft.

πŸ’‘ About KORMo-VL-Diffusion

KORMo-VL is a vision-language model developed from scratch by the KAIST MLP Lab (https://sites.google.com/view/aailab), built on top of KORMo-10B. The system consists of two components:

  • Vision-Language Model (VLM)
  • Image Generation Model

The KORMo-VL-Diffusion model, designed for image generation, was trained from scratch with a high proportion of images reflecting Korean daily environments and culture. Unfortunately, due to limited GPU resources during the research process, we are sharing the intermediate results of the model at this stage.


KORMo-VL은 KAIST MLP μ—°κ΅¬μ‹€μ—μ„œ from scratch둜 κ°œλ°œν•œ μ‹œκ°-μ–Έμ–΄ λͺ¨λΈλ‘œ, KORMo-10Bλ₯Ό 기반으둜 (1) μ‹œκ°-μ–Έμ–΄ λͺ¨λΈκ³Ό (2) 이미지 생성 λͺ¨λΈλ‘œ κ΅¬μ„±λ˜μ–΄ μžˆμŠ΅λ‹ˆλ‹€.

이 쀑 이미지 생성을 μœ„ν•œ KORMo-VL-Diffusion λͺ¨λΈμ€ ν•œκ΅­μ˜ μƒν™œ ν™˜κ²½κ³Ό λ¬Έν™”λ₯Ό λ°˜μ˜ν•˜κΈ° μœ„ν•΄ κ΅­λ‚΄ ν™˜κ²½ 이미지λ₯Ό κ°€λŠ₯ν•œ 높은 λΉ„μœ¨λ‘œ μ‚¬μš©ν•˜μ—¬ from scratchλΆ€ν„° ν•™μŠ΅λœ λͺ¨λΈμž…λ‹ˆλ‹€. λ‹€λ§Œ 연ꡬ μ§„ν–‰ 쀑 GPU μžμ›μ„ μΆ”κ°€λ‘œ ν™•λ³΄ν•˜μ§€ λͺ»ν•΄ ν˜„μž¬λŠ” 쀑간 결과물을 κ³΅μœ ν•˜κ²Œ λ˜μ—ˆμŠ΅λ‹ˆλ‹€.

  • LLM: KORMo-VL
  • Model Structure: Qwen-Imageλ₯Ό ꡬ쑰λ₯Ό μ°Έμ‘°ν•΄ μž¬κ°œλ°œν•¨ (20B μ •λ„μ˜ Diffusion뢀뢄을 λ³€ν˜•ν•΄ scratchλΆ€ν„° ν•™μŠ΅)
  • Languages: Korean / English
  • Training Data: Synthetic data + public datasets (e.g., AI Hub, details to be released)

ν–₯ν›„ ν•΄λ‹Ή λͺ¨λΈμ„ μΆ©λΆ„νžˆ ν•™μŠ΅ν•  수 μžˆλŠ” ν™˜κ²½μ΄ λ§ˆλ ¨λœλ‹€λ©΄ μ™„μ„±λœ λͺ¨λΈλ‘œ λ°œμ „μ‹œν‚€λŠ” 것을 λͺ©ν‘œλ‘œ ν•˜κ³  μžˆμŠ΅λ‹ˆλ‹€. 쀑간 κ²°κ³Όλ¬Ό μœ„μ—μ„œ μΆ”κ°€ νŠœλ‹μ΄λ‚˜ 연ꡬλ₯Ό μ§„ν–‰ν•˜κ³  싢은 뢄듀은 자유둭게 ν™œμš©ν•΄ λ³΄μ‹œκΈ° λ°”λžλ‹ˆλ‹€.

πŸ“ˆ T2I Performance

English Prompt

Prompt Generated Image
Prompt: Dense forest
Prompt: Black pattern mug

Korean Prompt

Prompt Generated Image
Prompt: μšΈμ°½ν•œ 숲
Prompt: 검은 무늬의 λ¨Έκ·Έμ»΅

KORMo-VL-Diffusion Demo

prompt: μ•„λ¦„λ‹€μš΄ μ •μ›μ˜ 꽃듀

πŸ“¦ Installation

uv pip install transformers==4.57.1 pillow torchvision diffusers

πŸš€ Inference Example

github repo ν™œμš© μ˜ˆμ •

Contact

  • KyungTae Lim, Professor at KAIST. ktlim@kaist.ac.kr

Contributor (https://sites.google.com/view/aailab)

  • Junghun Yuk
  • INho won
  • HANGYEOL YOO
  • Junmyeong Lee
  • KyungTae Lim

Citation

@misc{KORMo,
  author = {Minjun Kim, Hyeonseok Lim, Hangyeol Yoo, Inho Won, Seungwoo Song, Minkyung Cho, Junghun Yuk, Changsu Choi, Dongjae Shin, Huije Lee, Hoyun Song, Alice Oh, and KyungTae Lim},
  title = {KORMo: Korean Open Reasoning Model for Everyone},
  year = {2025},
  publisher = {GitHub},
  journal = {Technical Report},
  paperLink = {\url{https://arxiv.org/abs/2510.09426}},
 },
}
Downloads last month
-
Inference Providers NEW
This model isn't deployed by any Inference Provider. πŸ™‹ Ask for provider support

Paper for KORMo-VL/KORMo-VL-Diffusion