Kissing Concept Augment LoRA

Details

Download Files

Model description

Concept LoRA that manipulates image output for kissing actions. Reference the sample pictures in the showcase for examples of how the output changes.

Version 2.0

Updated dataset with new images. Added more variation to reduce impact on style. Weights between 1.0 ~ 1.2 do not impact style as much. Recommended not to exceed 1.5 in most cases. Changed the trigger words for certain concepts.

Concepts & Notes:

  • Sloppysmooch

    • Intertwined tongues is now "sloppysmooch" so that "tongues" in the previous trigger word does not affect the output.

    • Steers the image so that the tongues have a lot of contact and are entering or close to entering the opposing subject's mouth

    • See sample XYZ plots in showcase for V2 to get an idea of how this affects the image. Increasing LoRA weight and weight of the sloppysmooch should fill the mouth up with tongue.

    • At a certain point increasing the weight might cause extra tongues.

    • The end goal of this concept is to get the tongues to wrap around each other. V2 doesn't achieve this.

  • French Kiss

    • Steers the output so that the tongues are touching, mostly at their tip

    • Opposed to sloppysmooch in that the subjects are a bit further away from one another and their tongues should not have significant contact or enter the other's mouth.

  • Kiss

    • Meant to produce an output that the lips are together and sealed or mostly sealed depending on whether you prompt for the tongue to be visible.

      • Prompting for the tongue to be visible should only barely show it with normal weights

      • Heavily trained with images of lips together with tongue not being visible

  • Implied Kiss

    • Intended to be similar to kiss but help steer the output if you want the subjects to be shown in a way that does not clearly show them kissing.

    • Minority of images in the full dataset (~8%), so the concept will be easily overwritten by the rest of your prompt if it conflicts with the concept.

  • After Kiss

    • Displays the subjects having just pulled apart from an intense kiss

    • Pruned Saliva & Saliva Trail from the tags. This differentiates the concept from Imminent Kiss

  • Imminent Kiss

    • Displays the subjects in a sensual mood, close to each other, and near kissing

Version 1.0

Didn't quite come out as I had intended. I changed the training settings slightly from what I usually use which may have been part of the reason why. I need to take another pass on the training dataset and captions used for tagging for sure though. I'll release an update later to improve it.

Images made by this model

No Images Found.