■ Learning Model = illustriousXL_v01

■ Operation check = JANKU Trained+v6.0

■ Recommended generation environment

Generate Model = JANKU Trained+(v5.0 or v6.0) / WAI-illustrious-SDXL

Sampling method = Euler a

Schedule type = Karras

Sampling steps = 30

CFG Scale = 5

LoRA weight = 0.6-0.9

■ Positive prompt

highres, highly detailed, amazing quality, very aesthetic,

+α = anime screenshot,

■ Negative prompt

123qweasdz, 3D, deformation, (worst quality, low quality,normal quality), lowres, monochrome, grayscale, watermark, flat colour, no shading,

■ caption design

・Adding "anime screenshot" to the positive prompt makes the output closer to anime original art. Removing it slightly reduces character fidelity, but improves the quality of highlights, shading, and other texture details.

・The caption design uses "123qweasdz" as a dummy token to absorb non-character noise from the dataset (such as background variations and dataset-specific artifacts). Placing it in the negative prompt suppresses those unwanted elements while preserving character features and clothing, which significantly improves generation stability.

■ Recommended resolution

21:9 = 1806 x 774

2:1 = 1672 x 836

16:9 = 1576 x 887

3:2 = 1448 x 965

4:3 = 1365 x 1024

1:1 = 1182 x 1182