모델 XL
Silene | Illustrious XL 1.0
- 업로드 날짜
- 2025. 2. 15. AM 2:38
- 이용수
- 1.26k
- 리뷰
- 훌륭함 (1)
- 샘플링 단계
35
- 샘플링 방식
Euler a
- CFG 스케일
6
- 권한
- 이미지 생성 및 공유 허용
- 다른 사용자가 내 모델을 다운로드하도록 허용합니다
- 생성된 이미지의 상업적 이용 허용
설명
Silene is a collaboration between @epiTune and @Shiiro0 <3Post your gens down below! I will ⚡ images I like! Have fun! <3Silene is a checkpoint aiming to bring forth the strengths of Illustrious 1.0 while also providing a more stable and consistent base for generating.Due to its Illustrious 1.0 base - this checkpoint handles a larger range of resolutions than earlier Illustrious checkpoints. - 832x1216 is our recommended lower-end resolution - 1024x1496 is stable and is recommended for anyone that can generate with it - it helps bring out richer and more crisp details in your generations - 1400x2048 is the highest resolution we recommend trying - it can produce crisp and delightful images but is less stable and can require trying more seedsWe believe that in future versions of Silene we can improve the high resolution stability even further, stay tuned! Silene makes the same stylistic trade-offs as Koki'o(https://civitai.com/models/1135145/kokio-illu) - aiming for a pleasing and consistent style while allowing explicit prompting to take over when needed. This expressivity is something we will aim to improve in future versions of Silene - if you run into limitations, try Koki'o!Another benefit of the Illustrious 1.0 base is better natural language prompting. While it can not be compared to models like SD3/Flux - it can definitely be very helpful for adding oft needed specificity. This is especially useful when a certain concept, pose, or detail does not have a well recognized Danbooru tag.Simple examples being index finger on tongue or both hands on neck.Your experience prompting earlier Illustrious models should transfer well - Illustrious 1.0 based models should ideally only understand tags better.Our recommendations for negative tags remain the same as for Koki'o: anatomical nonsense, artistic error, bad anatomy, bad arm, bad aspect ratio, bad ass, bad face, bad feet, bad hands, bad knees, bad leg, bad neck, bad perspective, bad proportions, bad reflection, bad vulva, different reflection, extra arms, extra digits, extra eyes, extra faces, extra legs, extra pupils, fewer digits, oversized limbs, wrong foot, wrong handTry to tailor what negative tags you use to your specific needs rather than always including them all.A trimmed down version that's suitable for general use is: anatomical nonsense, artistic error, bad anatomyThese should be used in tandem with staple negative tags such as: worst quality, bad quality, low quality, lowresOn the flip side, recommended positive tags are: incredibly absurdres, highres, masterpiece, newestYou could also consider various combinations of tags such as: detailed background, depth of field, foreshortening, dark theme, light themeBut do note that certain art styles benefit from omitting masterpiece and newest.These tags heavily lean into the refined and modern digital look, if you want to generate watercolor these might work against you.Likewise, if you're going for a retro look, avoid newest. For a traditional art style you might want to include tags such as: traditional media, painterly, good quality, average qualitySome tags that are useful for steering away from the default anime look include:POS: painterly, no lineart, blendingNEG: flat color, high contrast, anime coloring, limited palette, outline, ligne claire, lineart, oekaki, jaggy lines, dithering, pixel art, monochrome, greyscale, sketch, anime screenshot Art Mediums vs Quality Tags - Illustrious Year TagsRecommended settings: Steps: 30-40 Scale: 4-6Sampler: Euler A OR DPM++ 2M