Used 50 images with 1024*1024 resolution and strong wlop style for 8000 steps of training, with the dataset using gemini3 fine labeling.

Under 8-step acceleration, it is recommended to set Lora strength to 1.50, sampler euler, and scheduler simple. The wlop art style is complex and difficult to imitate, so it is recommended to generate images multiple times to achieve the best results.

Compared to the flux version, the single image generation success rate, aesthetics, and style similarity have all improved. Lora does not have prompt words set, but testing has found that adding keywords like "hand-drawn style," "thick paint," or "brushstroke" in certain specific prompts yields better results.