This is my LoRA for the spooning position. It's a bit volatile, but inpainting the small off-details should do the trick. It is also rather prone to doing anal, and I suspect that this was an oversight in my dataset. Try '(anal)' in your negative prompts to counteract this.

I use 0.75 strength, but it is usable at 0.6-0.8 comfortably. Notably, the testicles are a bit off at times, but this is a failure of generation along the line of hands/feet in which it's more luck dependent. This model doesn't require any specific keywords except general sex or sex adjacent tags. Provided in the keyword section is tags I believe will work in tandem well.

Images generated on my Galena, Euclase, and Sardonyx models. Since these models are based on AOM3, and this LoRA was trained on NovelAI, expect it to work with other anime based models. All images are generated with Hires fix and the Orangemix VAE.