This LoRA aims at improving the realism of checkpoints. It doesn't require trigger keyword, can be used as a negative (don't see the point but you can) or with a weight between 0 and 2.

Please note that the sample pictures are made without Adetailer, without specific prompting. Also, more realistic doesn't mean prettier.

While it is intended to be used with semi-realistic checkpoints, it can be with all and can improve even realistic ones.

Due the tagging method it should also be efficient with non-Pony SDXL checkpoints.

For all the following examples, the positive prompt includes score_9, score_8_up, score_7_up, and the negative only score_5, score_4, score_3, score_2, score_1.

Checkpoint: Real dream, Prompt: a blonde woman, long hair, green eyes, seductive smile, looking at viewer, shy, blush, hand on chin, realistic

Checkpoint: Pony, Prompt: a bird on a branch, looking at viewer

Checkpoint: AutismMix, Prompt: outdoors, sky, day, tree, no humans, leaf, sunlight, nature, scenery, road, autumn leaves, autumn, realistic

Checkpoint: Real dream, Prompt: 1girl, redhead, portrait,realistic,freckles

Technical point of view

Training has been made on 375 high resolution pictures including humans (male and female), landscapes, animals and some NSFW stuff. Tagging is double (Pony and wd14) with a aesthetic rating from source_1 to source_9. Details can be seen when checking the LoRA using your favorite GUI.

This LoRA is also intended to be used as a merge for realistic checkpoints.