Updated: Sep 6, 2026
style
The art style of the 2020 TV anime "Dragon Quest: The Adventure of Dai" (Toei Animation).
Style LoRA. All sample images were generated with PROMPT ONLY - no character LoRA.
The first sample is the same prompt and seed WITHOUT this LoRA, for comparison.
RECOMMENDED
Weight : 0.4 (0.2 - 0.6)
Base model : WAI-Illustrious SDXL v1.7
Sampler : euler_a, CFG 5
Hires fix : 0.32 denoise
Trigger : dai2020b
★ Do NOT put cel shading, watercolor, painterly, soft focus or muted in the negative.
They cancel this style completely. This is the single most common way to make it "not work".
WHAT YOU SHOULD KNOW (measured, not guessed)
0.8 and above breaks faces. At 1.0 the subject can split into several people. Start at 0.4.
- More than half of the training frames are MALE characters (1boy 178 / 1girl 97).
Applied strongly to a female character, it makes the jawline harsh.
- It was trained mostly on CLOSE-UPS (close-up 196 / portrait 200). In full-body shots the
background keeps the base model's look, and raising the weight does NOT fix it.
- Because it was trained on the show's own footage, it carries the show's faces. On a character
LoRA it will fight for the face above 0.4.
TRAINING DATA (being upfront about this)
314 frames from the 2020 TV series. Captions written with wd14, plus blur and shot-distance
written in by hand (v1 had none, so the blur itself got learned as "the style").
Trained with kohya on Illustrious.
WORKS WELL WITH
My character LoRA merle_yt (Merle), trained from the same series.
■ 説明（日本語）
『ダイの大冒険』2020年TV版の作画で焼いた画風LoRAです。本編314枚・自前学習。推奨0.4。
作例は**キャラLoRAを使わず言葉だけ**で出しています（1枚目はLoRA無しの比較）。
★ネガに cel shading watercolor, painterly, soft focus, muted があると打ち消されます。
★0.8以上で顔が崩れ、1.0で人物が分裂。★素材の半分以上が男性キャラ。★寄りの絵で学習。