EDIT: V2 released (trained with a dataset that fixes the face similarity). Make sure that you use the trigger words as given.

Here's another LoRA for a concept that Pony models do not understand natively. This one works with a wide range of resolutions (I prefer 1216 x 832), and is much easier to use than some of my other LoRAs. This time, it was baked well because of the highly curated synthetic data generated over a few days.

First, start with 0.75 (most consistent) strength for this LoRA and move upwards from there, if needed. I'd also encourage using different strengths and experimenting more than what I could do with my GTX 1650.

The secret to getting the best results depends mostly on LoRA strength ratios between this one and the character LoRA you use in tandem with it. Experiment yourself to get the best values for different LoRAs.

Cons:

• Mention of "*****" or "penis" can sometimes break the pose (happens rarely). Most Pony models are overtrained on these tags.

• Using 832 x 1216 (portrait) is really hard. Only works rarely. Use landscape or 1024 x 1024.

• I forgor to blur the faces while making the dataset XD. Use higher character LoRA weights to bypass this. Shouldn't be a problem most of the time.

• Don't see any other cons with V2, tell me if you find any.

How to use (tested prompting structure):

Start the prompt with the trigger words (Try shuffling them for variety):

p0vth1ghj0b, *****, grinding, thigh_sex, sitting_on_person, on_back, 1boy, lying, POV, penis between thighs, 1girl,

Then comes your character descriptors. Here's an example:

aqua \(konosuba\),

Last comes the description of the background. Example:

sunrise, beautiful flowers, hanging arch, starry, mountain, background

In case of any doubts, feel free to ask or comment.

If you want to support more models like this, or you just like what you see: ko-fi (｡♥‿♥｡)

Also, I don't plan on charging buzz for my LoRAs. So, I'd really appreciate if by any chance you could tip the model.