SDXL

illust-Vector-2.5D_IL

illust-Vector-2.5D_IL - AI Model 커버 이미지
업로드 날짜
2025. 10. 27. 오전 1:20
이용수
10.3k
리뷰
훌륭함  (1)
샘플링 방식

DPM++ SDE Karras

CFG 스케일

5

제외 조건

(bad quality:1.4), (worst quality:1.4), (low quality:1.4), (normal quality:1.4), ((bad)), bad anatomy, bad hands, extra digits, too many fingers,fused fingers,(missing fingers:1.2), lowres, blurry, distorted, blur, fewer, extra, missing, error, jpeg artifacts, unfinished, displeasing, oldest, early,chromatic aberration, artistic error, scan, bad tooth,(fat:1.2),(old:1.2),(uglily:1.2),long neck,small head,bad nose,oil skin,ai-generated,embedding:lazyneg,

권한
  • 이미지 생성 및 공유 허용
  • 다른 사용자가 내 모델을 다운로드하도록 허용합니다
  • 생성된 이미지의 상업적 이용 허용

추천 LoRA

설명

Sketch, Anime, 2D, 2.5D. Danbooru tags are still available, but this is a flatter semi-realistic look, with the level of realism controlled by a prompt. she has exquisite tattoos and loves lace, stockings, armor, and cyberpunk, balancing control and imagination.

Steps: 20-40, CFG scale: 4-7, Sampler: dpmpp_sde_gpu, scheduler: beat,

The VAE is already integrated

Try it online: Tensor | Seaart | Liblib

Although other samplers can be used, for optimal results, I only recommend using dpmpp_sde_gpu. No dedicated VAEs are used, so feel free to use upscalers and ADetailers, or train Lora.

use size larger than1024x1024 for the original dimensions.

A few days ago, I wanted to create a flatter semi-realistic checkpoint. Seeing Civit @reijlita new model Rijverse gave me an idea: maybe I could create one using my semi-realistic and 2D models, so I merged them. I used illust-Semi-realistic_IL_v4 and illust-Anime-2D_IL_v1. trained a Loha for upscaling, which fixed some facial issues.

I wish you all a happy time.

댓글