A LoRA trained on a collection of images that were either:

• Trending on PixAI over the last few days

• Have come top in events/competitions

• Were top on detail or aesthetic focused LoRA

This model isn't trained for a specific style/composition, although it may be biased towards images that PixAI users consider good. Recommended weights 0.2 - 0.8 (see sample images for examples of different weights).

The model is trained to activate when it sees the token:

trending_on_pixai

this model is intended to be used with any baked-in quality tags in the base model.

This model works both on human and non-human targets, with a dateset comprising of:

• 553 1girl images

• 130 1boy images

• 115 group images (2girls, 3girls, 2boys, 1boy & 1girl, etc.)

• 46 still-life subjects and animals

• 78 images of detailed scenery

The first set of sample images are generated using the base Illustrious-v1.0, and this is the model that it was trained on. It works well with other finetunes. All sample images are generated using a simple negative, so that the effects of this LoRA are more obvious (the LoRA was not trained on any negative tags):

worst_quality, low_quality, bad_anatomy, bad_hands, text

The last five images in the showcase are generated using the one of the most popular finetunes of Illustrious (WAI v14) with a full negative prompt for illustrative purposes.