AIBooru Beta
Login Posts Comments Notes Artists Tags Pools Wiki Forum More »
Listing Upload Hot Changes Help

Search

  • Help
guro
scat

Artist

  • ? 1lch2 11

Model

  • ? anythingelse 833

Copyright

  • ? arknights 984

Character

  • ? angelina (arknights) 6

General

  • ? 1girl 116k
  • ? arms behind back 4.2k
  • ? bare shoulders 27k
  • ? bikini 8.8k
  • ? black collar 552
  • ? breasts 88k
  • ? breasts apart 2.7k
  • ? brown hair 17k
  • ? cameltoe 3.9k
  • ? collar 4.0k
  • ? collarbone 27k
  • ? cowboy shot 16k
  • ? groin 3.0k
  • ? hair between eyes 23k
  • ? hair ribbon 8.0k
  • ? head tilt 2.1k
  • ? infection monitor (arknights) 14
  • ? jacket 9.3k
  • ? jewelry 17k
  • ? long hair 70k
  • ? long sleeves 20k
  • ? looking at viewer 86k
  • ? medium breasts 20k
  • ? navel 36k
  • ? necklace 4.0k
  • ? off shoulder 5.2k
  • ? open clothes 7.6k
  • ? open jacket 3.0k
  • ? open mouth 26k
  • ? parted lips 10k
  • ? red bikini 713
  • ? red eyes 19k
  • ? red ribbon 3.6k
  • ? ribbon 15k
  • ? shadow 1.9k
  • ? side-tie bikini bottom 1.6k
  • ? simple background 18k
  • ? smile 39k
  • ? solo 97k
  • ? swimsuit 10k
  • ? white background 10k
  • ? white jacket 722

Meta

  • ? bad id 2.6k
  • ? ↳ bad pixiv id 1.8k
  • ? commentary request 6.7k
  • ? lora 21k

Information

  • ID: 20312
  • Uploader: Penance »
  • Date: over 2 years ago
  • Size: 805 KB .png (576x1024) »
  • Source: pixiv.net/artworks/105467319 »
  • Rating: Sensitive
  • Views: 780
  • Score: 10
  • Favorites: 10
  • Status: Active

Options

  • Resize to window
  • Find similar
  • Download

History

  • Tags
  • Pools
  • Notes
  • Moderation
  • Commentary
  • AI Metadata
angelina (arknights) generated by 1lch2 using anythingelse
Original Commentary AI Metadata

【NovelAI】德克萨斯 安洁莉娜 琴柳 巫恋

从AnythingV3换成了Anything v4.5模型,画面风格明显有了进步,差不多可以拿出来看了。

对比发现超采样放大算法对画面清晰度影响非常大,其实默认效果就已经不错了,列表中后面很多GAN的算法会导致画面模糊,巫恋那一张就是典型。

同时LoRA模型的权重过高会影响prompt效果。安洁莉娜的第一张LoRA模型权重过高,导致很多prompt完全没起作用。第二张降低到了0.8后,加了泳装无外套的prompt,还是出现了白色外套的元素,可能需要再降低一点来配合。

另外一点小发现,人物姿态的prompt对防止出现三腿,单腿等异常效果很有帮助,比如涩图中常见的张开腿等。至于负面prompt,绝大部分都是用来描述肢体,尤其是手,但实际上还是得看运气。尝试了生成后再用inpaint对手部进行修复,效果不佳,还不如指望它一开始就生成比较正常的手,或者干脆使用 arms behind back 这种挡住手的prompt。

后面准备研究下怎么自己训练LoRA模型,以及其他embedding的效果,离大魔导师还有很远的距离。

当然,AI还是图一乐,毕竟不是自己画的,这东西本质还是高级洗稿。

Prompt 1girl, (masterpiece), 4k, highres, high resolution, simple background, beautiful, anime, animal ears, no outfit, (red bikini), arms behind back, eye_contact
Negative prompt (((pubic))), ((((pubic_hair))))sketch, duplicate, ugly, huge eyes, text, logo, monochrome, worst face, (bad and mutated hands:1.3), (worst quality:2.0), (low quality:2.0), (blurry:2.0), horror, geometry, bad_prompt, (bad hands), (bad legs), (more than 2 legs), (missing fingers), multiple limbs, bad anatomy, (interlocked fingers:1.2), Ugly Fingers, (extra digit and hands and fingers and legs and arms:1.4), crown braid, ((2girl)), (deformed fingers:1.2), (long fingers:1.2), succubus wings, horn, succubus horn, succubus hairstyle, (bad-artist-anime), bad-artist, bad hand, , , bad_prompt_version2
Sampler DPM++ 2M Karras
Seed 278089968
Steps 24
Cfg Scale 7
Model Hash 6e430eb514
  • ‹ prev Search: -user:albert status:any next ›
  • Comments
  • There are no comments.

    Terms / Privacy / Contact /