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

ユーザーLoRA

XL
キャラ

Rydia (Realistic) - Final Fantasy 4 | IllustriousXL

#レオタード#リアル絵#リディア#ラテックス#リディア#リディア#リディア#リディア

創作

  • AI画像生成
  • AI動画生成
  • ツールボックス
  • テーマジェネレーター
  • LoRA学習
  • Mio.2 エージェント
  • Studio

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
Rydia (Realistic) - Final Fantasy 4 | IllustriousXL - AI Model cover image
アップロード日時
2026/08/12 10:10
採用数
0
レビュー
レビューがありません
トリガーワード

Rydia_(Final_Fantasy_IV), green hair, long wavy hair, green eyes, curvy figure, wide hips, thick thighs, green glossy bodysuit, gold embroidery, long green gloves, green thigh-high boots, white lace trim, gold hair ornament, pink gemstone, brown beaded necklace, bridal fingerless gauntlet gloves, fingerless opera gloves, elbow-length gloves, glossy green latex, patent leather material, delicate gold filigree, leaf acanthus gold embroidery, pointed cuff design,

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

Rydia — Final Fantasy IV | Realistic Character LoRAA realistic character LoRA for Rydia from Final Fantasy IV, trained for realistic image generation.The LoRA was trained specifically to capture Rydia's recognizable character appearance while allowing her to be used with different poses, compositions, outfits, and environments.

Trigger Word

Rydia_(Final_Fantasy_IV)

Recommended LoRA Strength

0.8–1.0

Prompt

Character:

Rydia_(Final_Fantasy_IV), green hair, long wavy hair, green eyes, curvy figure, wide hips, thick thighs,

Outfit:

green glossy bodysuit, gold embroidery, long green gloves, green thigh-high boots, white lace trim, gold hair ornament, pink gemstone, brown beaded necklace, bridal fingerless gauntlet gloves, fingerless opera gloves, elbow-length gloves, glossy green latex, patent leather material, delicate gold filigree, leaf acanthus gold embroidery, pointed cuff design,

The idea is to use Rydia_(Final_Fantasy_IV) to activate the character, then describe the pose, clothing, camera angle, lighting, environment, etc. separately.

Recommended Negative Prompt

A general realistic-generation negative prompt can be used:

low quality, blurry, distorted face, deformed hands, extra fingers, extra limbs, bad anatomy, duplicate, poorly drawn face

The negative prompt is not required to activate the LoRA and can be adjusted according to the checkpoint/workflow being used.

Training InformationBase Model:novaRealityXL_ilV90Architecture:SDXLTraining method:DreamBooth LoRATraining images:124Repeats:5Effective training images:620 per epochEpochs:3Total training steps:1860Resolution:1024 × 1024Bucket resolution:512–1024Bucket step:64Batch size:1Gradient accumulation:1LoRA Rank / Network Dimension:32LoRA Alpha:16Optimizer:8-bit AdamWPrecision:BF16Text Encoders:CLIP Text Encoder 1 CLIP Text Encoder 2U-Net LoRA modules:722Text Encoder LoRA modules:264Regularization images:NoneLatent caching:Enabledxformers:EnabledHardware:NVIDIA GeForce RTX 4090 24GB 128GB system RAMTraining time:~24 minutesCharacter AppearanceThe LoRA is intended to reproduce Rydia's recognizable appearance, including her characteristic: Green hair Long/wavy hairstyle Green eyes Distinctive facial appearance Curvy body proportions The original character design has been adapted toward a realistic interpretation rather than reproducing the original game's exact visual style.TipsThe model was trained on NovaRealityv9.0It can produce images using other models like CyberIllustrious or UnNamedIXLBut the best results were on NovaRealityv9.0 as it was trained with.NotesThis LoRA was trained on 124 images with 5 repeats for 3 epochs, resulting in 1,860 optimization steps.The training dataset contained varied image compositions and aspect ratios, with SDXL bucket training enabled.The model was trained with no NSFW aspects of any image. So while the checkpoint might produce results, it may not be the best or might look uncanny lacking training data.This is a character LoRA, so the trigger is intended to identify the character rather than lock the generation to a particular pose, camera angle, or environment.

コメント