Lewd Expression Modifier

A part-concept, part-style LoRA that provides a modified style of lewd expressions. Note that the trigger words are different across versions, so always reference the version notes on the side as I place the trigger words in the details table for each version. This LoRA will include major updates (megapack versions) and supplemental updates which will be single-concept LoRA dedicated to individual expressions. Use the supplemental LoRA releases if there is a specific expression you are interested in, or the megapack if you want to experiment with all of them. Note that the megapacks will have some amount of concept bleeding, so the supplemental versions will always be the most "pure" of concept.

Megapack Version 3

Recommended weight - anywhere from 0 to 2.0. See below for comparisons, but different expressions will require different weights to reliably prompt them. 1.0 is a solid start with everything and will output the expression. You will start to see style shift around 0.80 and higher.

Expressions Included:

Ahegao
Aroused
Ecstasy
Moaning
Thunderstruck
Torogao

The Lewd Expression Megapack consists of a refined dataset that is equally spread amongst the individual LoRA release concepts. The benefits of this are that you can mix and match expression types by combining their trigger words all with a single LoRA. This has the added benefit of saving disk space due to the condensed file size. The downside is that there is bleeding between concepts. Due to the amount of concepts in V3 this is more prevalent between certain concepts. The XYZ plot images in the showcase sample give a decent indication as to which expressions have overlap. You can use negative prompts or just carefully craft your positive prompt to rule out some of this overlap.

One issue you will find with a few of the expressions is that the eyes get "distorted" with a line through them. I refer to this as "dashed eyes" within my training dataset. I ran into this with a few of the individual release LoRA and found that including that carefully throughout the dataset allowed this to be filtered out by including "Dashed eyes" with the negative prompt. This is still the case for V3; however, I found that you may need to increase the weight of the "dashed eyes" keyword in your negative prompt.

Weight examples

Individual V1 Releases

The additional V1 releases are supplemental LoRA with specific expressions to avoid dilution in the major releases and also to help me finetune the datasets for each of the specific concepts. Expect V1 releases for multiple expressions before a V3 with my LoRAs. Note that trigger words may change between releases as I experiment with them. Always check the version details for the trigger words for a specific LoRA release.

There may be V2 releases of the individual LoRA. TBD as I'm undecided whether I will work on those of a V3 megapack first. Likely I will work on another concept LoRA once the first batch of V1 individual are released.

V2

Significant cleanup of dataset and refinement of trigger words. Note that the trigger word pleasureface in V2 has been replaced with lewd expression instead. Trigger word naming may change in future versions as well as concepts get added, merged, or removed from the LoRA.

See showcase in this version for more trigger word examples.