Instructions to use gagong/korean-sumukhwa-model-ver-1 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Diffusers
How to use gagong/korean-sumukhwa-model-ver-1 with Diffusers:
pip install -U diffusers transformers accelerate
import torch from diffusers import DiffusionPipeline # switch to "mps" for apple devices pipe = DiffusionPipeline.from_pretrained("gagong/korean-sumukhwa-model-ver-1", dtype=torch.bfloat16, device_map="cuda") prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k" image = pipe(prompt).images[0] - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- Draw Things
- DiffusionBee
| license: creativeml-openrail-m | |
| language: | |
| - en | |
| - ko | |
| pipeline_tag: text-to-image | |
| tags: | |
| - art | |
| [์๋ก์ด SDXL๋ฒ์ ์ด ๋์์ต๋๋ค.](https://huggingface.co/gagong/Traditional-Korean-Painting-Model-v2.0) | |
|  | |
| ํ๊ตญ ์๋ฌตํ ๋ชจ๋ธ ์ฌ์ฉ ๊ฐ์ด๋ sd1.5๊ธฐ๋ฐ | |
| Guide for Using Korean Sumukhwa Model based on SD1.5 | |
| ์ด์ ์ ์ ์ํ๋ ํ์ดํผ๋ทํธ์ํฌ์ ์ด์ด์ ์ด๋ฒ์ ํ์ธํ๋ ํ ๋์จ ๋ชจ๋ธ์ ๊ณต์ ํฉ๋๋ค. ๊ทธ๋ฆฌ๊ณ ๋ชจ๋ธ ์ฌ์ฉ ๊ฐ์ด๋๋ก ๋จ๊น๋๋ค. ๋ฐ์ดํฐ๋ ๊ณต์ ๋ง๋น์ ์๋ ๊นํ๋ ๊ทธ๋ฆผ ์ค ์ ๋ณ ํ ์๋ฃ์ Aiํ๋ธ์ ์ฌ๋ผ์ ์๋ ํ๊ตญํ ๋ฐ์ดํฐ์ ์ ๋๋ค. | |
| Continuing from the previous Hypernetworks, I am sharing the fine-tuned model and providing a guide for its usage. The data used for training includes selected materials from Kim Hong-do's paintings available on the Gongu Sharing Market and the Korean painting dataset on AI Hub. | |
| # ๋ชจ๋ธ์ ์ฌ์ฉ๋ ์๋ฃ ์ถ์ฒ | |
| # Sources of the data used in the model: | |
| - ๊ณต์ ๋ง๋น https://gongu.copyright.or.kr/gongu/search/search/list.do?viewType=&collection=gongu_wrt&kwd=%EA%B9%80%ED%99%8D%EB%8F%84& | |
| - Aiํ๋ธ https://www.aihub.or.kr/aihubdata/data/view.do?currMenu=115&topMenu=100&aihubDataSe=realm&dataSetSn=71380 | |
| # ํ์ต๊ณผ์ | |
| # Training Process | |
| ์ด๋ฏธ์ง๋ฅผ 768ร768์ฌ์ด์ฆ๋ก ๋ฐ๊พผ๋ค, clip_interrogator๋ฅผ ํตํด ํ๋กฌํํธ๋ฅผ ๋ง๋ค์์ต๋๋ค. ์ดํ ํ๊ตญ ์๋ฌตํ ์๋ฃ์ gksrnrghk๋ผ๋ ํ๋กฌํํธ๋ฅผ ๋ถ์ด๊ณ , ๊นํ๋ ๊ทธ๋ฆผ์๋ rlaghdeh๋ผ๋ ํ๋กฌํํธ๋ฅผ ์ถ๊ฐ๋ก ๋ถ์์ต๋๋ค. | |
| ์ด ์ด๋ฏธ์ง๋ฅผ ๋ค์ 512ร512์ฌ์ด์ฆ๋ก ๋ฐ๊พผ ๋ค Stable Tuner๋ฅผ ์ด์ฉํด์ ํ์ต์ ํ์ต๋๋ค. ์ฌ์ฉํ ์ค์ ์ ์๋์ ๊ฐ์ต๋๋ค. | |
| The images were resized to 768ร768, and prompts were created using the clip_interrogator. The prompt "gksrnrghk" was added for Korean ink wash painting data, and an additional prompt "rlaghdeh" was added for Kim Hong-do's paintings. These images were then resized to 512ร512, and training was performed using the Stable Tuner. The following settings were used | |
| - pretrained model: runwayml/stable-diffusion-v1-5 | |
| - seed: 3434554 | |
| - resolution: 512 | |
| - train batch size: 24 | |
| - num train epochs: 60 | |
| - learning rate: 5e-6 | |
| ์๋ 768 ๋ชจ๋ธ์ ์๊ฐํ๊ณ ์ ์ํ์ผ๋, ์คํํ ์ ์๋ ํ๊ฒฝ์ด ์ ์ฝ์ด ํฌ๊ณ ์ปจํธ๋กค๋ท์ ์ฌ์ฉํ ์ ์์๊ธฐ์ 512๋ชจ๋ธ 1.5๋ฒ์ ์ผ๋ก ๋ค์ ์์ ์ ํ์ต๋๋ค. | |
| Originally, I intended to create a 768 model, but due to constraints on the execution environment and the unavailability of the Controlnet, I had to work with the 512 model version 1.5. | |
| ์๋ฃ๊ฐ ์ด๋ฏธ ์ค๋น ๋์ด์๊ธฐ ๋๋ฌธ์ ๊ฒฝ์ฐ์ ๋ฐ๋ผ์๋ (์์ฐ์ด๋ผ๋ ์ง) ์๋กญ๊ฒ ํ์ต ํ ์ง๋ ๋ชจ๋ฅด๊ฒ ์ต๋๋ค. | |
| Since the data is already prepared, it may not be necessary to train again in some cases (e.g., budget constraints). | |
| # ์ฌ์ฉ ๊ฐ์ด๋ | |
| # Usage Guide | |
| ํ๊ตญ ์๋ฌตํ ๋ฐ์ดํฐ๋ 6000์ฅ ์ ๋ ๊นํ๋ ๊ทธ๋ฆผ์ 1000์ฅ ์ ๋ ์ฌ์ฉ๋์์ต๋๋ค. ๊ทธ๋ฐ ์ด์ ์ธ์ง, ํ๊ตญ ์๋ฌตํ ์คํ์ผ๋ก ํ๋ ค๋ฉด CFG Scale๋ฅผ 2-7 ์ฌ์ด๋ก ๊นํ๋ ๊ทธ๋ฆผ์ ์คํ์ผ๋ก ํ๋ ค๋ฉด 4-12์ฌ์ด๋ฅผ ์ถ์ฒํฉ๋๋ค. ๋๊ฐ์ ์คํ์ผ ๋ชจ๋ ํ์ฉํ ๊ฒฝ์ฐ ์ค๊ฐ ๊ฐ์ผ๋ก ํ๋ ๊ฒ์ ์ถ์ฒํฉ๋๋ค. ์คํ ์์๋ ์ํฅ์ ๋ฐ๊ธฐ ๋๋ฌธ์ ์ ์ ํ ๊ฐ์ ์ฐพ๋ ๊ฒ์ด ์ค์ํฉ๋๋ค. | |
| For Korean ink wash painting style, it is recommended to use CFG Scale between 2-7. For Kim Hong-do's painting style, a range of 4-12 is recommended. If you want to utilize both styles, it is recommended to use an intermediate value. The step count also affects the output, so finding an appropriate value is important. | |
| ์๋์ ์ ํ์ง์์ง๋ง ๊ธฐ๋ฒ์ ๋ํ ํ๋กฌํํธ๋ ์ ์ฉ์ด ๋์ด์์ผ๋ฉฐ ์ฌ์ฉํ ๋๋ ์๋ ํ๋กฌํํธ๋ฅผ ์ฌ์ฉํ๋ฉด ๋ฉ๋๋ค. (ํ์ง๋ง ๋ฏธ๋ฌํ ์ฐจ์ด๋ง์ด ๋ฐ์ํฉ๋๋ค.) | |
| Although it does not perform well, the prompts for the techniques are applied, and when using it, you can use the following prompts. (However, only subtle differences may occur.) | |
| Translate it into English. | |
| - ๋ฐฑ๋ฌต๋ฒ/Baekmukbeob: baegmyobeob | |
| - ๋ชฐ๊ณจ๋ฒ/molgolbeob: molgolbeob | |
| - ๊ตฌ๋ฅต๋ฒ/guleugbeob: guleugbeob | |
| ๊นํ๋ ๊ทธ๋ฆผ์ ๊ฐ์กฐํ๊ณ ์ถ์ผ๋ฉด rlaghdeh style, rlaghdeh painting์ด๋ ํ๋กฌํํธ๋ฅผ ๊ฐ์ด ์ฌ์ฉํ๋ฉด ์ข๋ ๊ฐ์กฐ๊ฐ ๋ฉ๋๋ค. | |
| If you want to emphasize Kim Hong-do's painting style, using the prompts "rlaghdeh style" and "rlaghdeh painting" together will enhance the emphasis. | |
| # ์ํ ์ด๋ฏธ์ง | |
| # Sample Images | |
| txt2img์ ์ํ์ด๋ฏธ์ง ์ ๋๋ค. | |
| Here is a sample image from txt2img | |
|  | |
| gksrnrghk, sky, tree Steps: 40, Sampler: DPM++ 2M SDE Karras, CFG scale: 2.0, Seed: 1271864954, Size: 768ร512, Model hash: a710c70889, Model: gksrnrghk_15_512_60, Clip skip: 2, Script: X/Y/Z plot, X Type: Prompt S/R, X Values: โgksrnrghk,\โgksrnrghk, rlaghdeh\โ,rlaghdeh โ, Y Type: CFG Scale, Y Values: โ2,3,4,5,7,9,12,15โ, Version: v1.3.0 | |
| ํ๊ตญํ์ ๊ฒฝ์ฐ ๋์ CFG์์ ํ๋ฐฑ์ด ์๋ ์ปฌ๋ฌ๊ฐ ๋์ค๊ธฐ ์์ํฉ๋๋ค. | |
| In the case of Korean ink wash painting, colors start to appear at higher CFG values instead of black and white. |