yoiko_style_kadenz🎼

μ„ΈλΆ€ 정보

파일 λ‹€μš΄λ‘œλ“œ

λͺ¨λΈ μ„€λͺ…

λͺ¨λΈ μΉ΄λ“œ / Model Card

λͺ¨λΈ 이름 / Model Name: yoiko_style_model
기반 λͺ¨λΈ / Parent Model: WAI-NSFW-illustrious-SDXL Ver15.0
λΌμ΄μ„ μŠ€ / License: Illustrious License
ν•™μŠ΅ 데이터 / Training Data: μ œκ°€ μƒμ„±ν•œ 이미지 (λΉ„κ³΅κ°œ 포함) / Images generated by me (including some not publicly released)


μ†Œκ°œλ¬Έ (일본어)

이 λͺ¨λΈμ€ WAI-NSFW-illustrious-SDXL Ver15.0을 기반으둜 좔가적인 **SFT(Supervised Fine-Tuning)**λ₯Ό μ μš©ν•œ 곡개용 베이슀 λͺ¨λΈμž…λ‹ˆλ‹€.
μŠ€νƒ€μΌμ€ 과거의 yoiko_lora_OD와 μœ μ‚¬ν•œ κ²½ν–₯을 보이며, μ΄λŠ” ν•™μŠ΅ 데이터와 μ‚¬μš© 이미지가 μœ μ‚¬ν•˜κΈ° λ•Œλ¬Έμž…λ‹ˆλ‹€. 이 λͺ¨λΈμ€ yoiko_lora_ODκ°€ μ§€λ‹Œ 저차원 ν‘œν˜„μ˜ κ°€μ€‘μΉ˜λ₯Ό 보닀 μ •κ΅ν•˜κ²Œ μ—…λ°μ΄νŠΈν•˜λŠ” 것을 λͺ©ν‘œλ‘œ ν–ˆμŠ΅λ‹ˆλ‹€.

이 λͺ¨λΈμ—μ„œλŠ” 또 λ‹€λ₯Έ 점을 μ˜μ‹ν–ˆμŠ΅λ‹ˆλ‹€. 그것은 κ²°ν•¨μ˜ μ΅œμ†Œν™”μž…λ‹ˆλ‹€. κ°€λŠ₯ν•œ λ²”μœ„λŠ” μ œν•œμ μ΄μ—ˆμ§€λ§Œ, μ΄λ―Έμ§€μ˜ ν’ˆμ§ˆκ³Ό 양을 늘렀 κ°œμ„ μ„ 도λͺ¨ν–ˆμŠ΅λ‹ˆλ‹€. κ·Έ κ²°κ³Ό, LoRA 단계보닀 μ•½ 3~4λ°° μ΄μƒμ˜ 이미지λ₯Ό μ‚¬μš©ν–ˆμœΌλ©°, λˆˆμ— λ„λŠ” 결함이 크게 μ€„μ–΄λ“€μ—ˆλ‹€κ³  λŠλ‚λ‹ˆλ‹€.

λͺ¨λΈ μ΄λ¦„μ˜ "kadenz"λŠ” μŒμ•… μš©μ–΄λ‘œ, '마무리' λ˜λŠ” 'μ’…κ²°'을 μ˜λ―Έν•©λ‹ˆλ‹€.
이 μ΄λ¦„μ—λŠ” "이 λͺ¨λΈμ΄ ν•˜λ‚˜μ˜ μ’…κ²°μ΄μž λ™μ‹œμ— μƒˆλ‘œμš΄ μ‹œμž‘"μ΄λΌλŠ” 의미λ₯Ό λ‹΄μ•˜μŠ΅λ‹ˆλ‹€.
완성도에 μžμ‹ μ΄ μžˆμ§€λ§Œ, λ¬Έμ œκ°€ μžˆλ‹€λ©΄ μ‚¬κ³Όλ“œλ¦½λ‹ˆλ‹€.

기반 λͺ¨λΈλ‘œ WAI-NSFW-illustrious-SDXL Ver15.0을 μ„ νƒν•œ μ΄μœ λŠ” 압도적인 완성도뿐 μ•„λ‹ˆλΌ, noob ai 계열 λͺ¨λΈλ“€κ³Ό 비ꡐ해 λΌμ΄μ„ μŠ€κ°€ μƒλŒ€μ μœΌλ‘œ μ™„ν™”λ˜μ–΄ μ‹€ν—˜μ˜ 폭이 λ„“κΈ° λ•Œλ¬Έμž…λ‹ˆλ‹€.
"kadenz"λΌλŠ” 이름을 λΆ™μ˜€μ§€λ§Œ, μƒμ„±λœ μ΄λ―Έμ§€μ˜ ν’ˆμ§ˆμ—λŠ” μ—¬μ „νžˆ λ―Έμ§€μ˜ μš”μ†Œκ°€ μžˆμŠ΅λ‹ˆλ‹€. ν”Όλ“œλ°±μ„ κΈ°λ‹€λ¦¬κ² μŠ΅λ‹ˆλ‹€.

그럼, 즐겁게 μ‚¬μš©ν•΄ μ£Όμ„Έμš”.


Description (English)

This model is a public base release built upon WAI-NSFW-illustrious-SDXL Ver15.0, with additional Supervised Fine-Tuning (SFT) applied.
Its visual style is similar to yoiko_lora_OD, as both share comparable training data and image sources. The goal was to refine and update the lower-level representational weights that define the yoiko_lora_OD aesthetic.

Another key objective was reducing visual artifacts.
While the improvements are incremental, I increased both the quality and quantity of training imagesβ€”using roughly 3 to 4 times more images than in the LoRA stage. This helped significantly reduce visible breakdowns in generated outputs.

The name β€œkadenz” comes from a musical term meaning cadence or closure.
I chose it because this model represents both an ending and a new beginning in my development cycle.
I’m confident in its completeness and balanceβ€”but if you find any issues, my apologies in advance.

I selected WAI-NSFW-illustrious-SDXL Ver15.0 as the foundation for its exceptional quality and flexible licensing, which allows for broader experimentation compared to some noob ai variants.
Even so, as the name kadenz suggests, the output quality may still hold unknownsβ€”so your feedback is very welcome.

Enjoy creating with it.

이 λͺ¨λΈλ‘œ λ§Œλ“  이미지

이미지λ₯Ό 찾을 수 μ—†μŠ΅λ‹ˆλ‹€.