Trained on images with the distinctive aesthetic of SD3.5L/M. It is not so flexible, but it mainly serves to enhance intricate chaotic details, backgrounds, and tattered textures.

Used categorized trigger words. While quite arbitrary, the following are the general criteria for tagging:

• 53m1r : Semi-realistic, detailed illustrations.

• 4b5tr : Abstract and surreal expressions. This trigger has incidentally absorbed many characteristics; when used, it becomes painterly, increasing detail and sharpness, but overuse can also make it noisy

• 4n1m3 : Anime-style illustrations, or those with such facial features.

• m1x : Images blending 2D and photo-like qualities, like some SD1.5 models.

• pntng : Images with strong painting or watercolor characteristics.

• s3d5 : Other realistic images. Has a flaw where it can also function as "3d".

However, effects are subtle except for "53m1r", "4b5tr".

Suggested weight: 1~1.5 : If the effect is weak, you can also try increasing the weight of the trigger words

When using highres fix with Latent (nearest-exact), the details become highly intricate. However, the structure occasionally becomes unstable.

(It seems many Stability AI models including SD3.5 and their derivatives recently became unavailable on Civitai, though I'm not sure if this LoRA is actually okay)