PixAI会員
PixAI
PixAI
Generate
Sign in
ホーム
生成
モデル
Studio
NEW
コンテスト
Mio.2 で作成
ツールボックス
NEW
ガイド

SD

Anything Evolved v1 (2D & 2D+) [Face fix recommended]

AI Tools
  • AI Anime Generator
  • AI画像生成
  • AI動画生成
  • Anime Figure Generator
  • Character Sheet Generator
  • Magazine Cover Generator
  • Plushie Generator
  • Giant Statue Generator

探索

  • 人気タグ
  • ランキング
  • モデルマーケット
  • コンテスト
  • お知らせ

PixAIについて

  • ガイド
  • PixAI ブログ
  • Tsubaki.2
  • マスコットのミオに会う
  • コンテンツ規則

料金とヘルプ

  • 会員
  • クレジットパック
  • お問い合わせ

アプリ

  • PixAI アプリを入手
  • App Store
  • Google Play
© 2026 PixAI
  • 利用規約
  • プライバシーポリシー
  • 著作権ポリシー
  • クレジット表記
  • 特定商取引法
Anything Evolved v1 (2D & 2D+) [Face fix recommended] - AI Model cover image
アップロード日時
2024/05/22 17:44
採用数
13.6k
レビュー
好評  (6)
Sampling Steps

26

Sampling Method

Euler a

CFGスケール

7

ネガティブ

(worst quality, low quality, bad quality:2), (normal quality:1.4), (poor quality:0.99), (bad-artist-anime, by bad-artist:1.3), (missing limbs, fused limbs, broken limbs, fewer digits, extra digits, mutation, deformity:1.3), (interlocked fingers, undetailed fingers:1.3), (bad eye, unclear eyes, undetailed eyes:1), (white background:1.3), (simple background:1), (monochrome:1), watermark, ng_deepnegative_v1_75t, bad-hands-5, verybadimagenegative_v1.3

権限
  • 画像生成と共有を許可する
  • ユーザーによるモデルのダウンロードを許可する
  • 生成画像の商用利用を許可する

詳細説明

Anything Evolved v1 tries to maintains Anything v3 poses while increasing the quality. Try face fix while enhancing to fix eyes. A very good model for closeups. It works better for me in Stable Diffusion. I'm still testing negatives for PixAI... Recommended resolutions: 768x512 and enhance or 1024x640. Minimum steps: 20 for closeups, 20-26 for cowboy shot, +30 for full body. Anything Evolved 2D_base used as a base for the model. Like all my models, everything is experimental so keep that in mind.

Anything Evolved 2D (still WIP) is a version that tries to fix hands and eyes from Anything V3, maintaining the 2D style. I'm gonna try all kind of things to make it a better model than models like Anything V4 or V5.

I noticed that on AnyEv v1 you can get better hands with simpler prompts. If you want the best quality I recommend HiRes into x1024 resolution.

Eyes were fixed for closeups, hands are so much better now in V1. Important: For 2D v1 you have to use enhance, highRes fix or higher resolutions like x1024. Use face fix for full body.

It is hard but I'm gonna try to make the model more appealing than release and base versions are by using other 2D models and maybe other models close to 3D, but I'll try to avoid them. I will try to make this one the best 2D model based on Anything-V3 models.

Note: It still has the prompt issues of Anything V3. Try prompts like (1girl:1.5), (1boy:1.5), (1woman:1.5), (1man:1.5), (detailed eyes)... which may help overall.

Issues with v1_r: Hands are not perfect. Issues with all v1: Hard to generate adults and gets confused with boy/girl body mostly for boy generations, which might be fixed with the upcoming versions.

I recently discovered new ways to fix models. I'm going to use each one for the final version of this model. Then maybe I'm adding even more stuff to it. Until I made an official version, it will be v1 or even v0.

コメント1